anytext init

iopaint/model/anytext/anytext_pipeline.py

@@ -0,0 +1,395 @@
+"""
+AnyText: Multilingual Visual Text Generation And Editing
+Paper: https://arxiv.org/abs/2311.03054
+Code: https://github.com/tyxsspa/AnyText
+Copyright (c) Alibaba, Inc. and its affiliates.
+"""
+import os
+
+os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
+import torch
+import random
+import re
+import numpy as np
+import cv2
+import einops
+import time
+from PIL import ImageFont
+from iopaint.model.anytext.cldm.model import create_model, load_state_dict
+from iopaint.model.anytext.cldm.ddim_hacked import DDIMSampler
+from iopaint.model.anytext.utils import (
+    resize_image,
+    check_channels,
+    draw_glyph,
+    draw_glyph2,
+)
+
+
+BBOX_MAX_NUM = 8
+PLACE_HOLDER = "*"
+max_chars = 20
+
+
+class AnyTextPipeline:
+    def __init__(self, cfg_path, model_dir, font_path, device, use_fp16=True):
+        self.cfg_path = cfg_path
+        self.model_dir = model_dir
+        self.font_path = font_path
+        self.use_fp16 = use_fp16
+        self.device = device
+        self.init_model()
+
+    """
+    return:
+        result: list of images in numpy.ndarray format
+        rst_code: 0: normal -1: error 1:warning
+        rst_info: string of error or warning
+        debug_info: string for debug, only valid if show_debug=True
+    """
+
+    def __call__(self, input_tensor, **forward_params):
+        tic = time.time()
+        str_warning = ""
+        # get inputs
+        seed = input_tensor.get("seed", -1)
+        if seed == -1:
+            seed = random.randint(0, 99999999)
+        # seed_everything(seed)
+        prompt = input_tensor.get("prompt")
+        draw_pos = input_tensor.get("draw_pos")
+        ori_image = input_tensor.get("ori_image")
+
+        mode = forward_params.get("mode")
+        sort_priority = forward_params.get("sort_priority", "↕")
+        show_debug = forward_params.get("show_debug", False)
+        revise_pos = forward_params.get("revise_pos", False)
+        img_count = forward_params.get("image_count", 4)
+        ddim_steps = forward_params.get("ddim_steps", 20)
+        w = forward_params.get("image_width", 512)
+        h = forward_params.get("image_height", 512)
+        strength = forward_params.get("strength", 1.0)
+        cfg_scale = forward_params.get("cfg_scale", 9.0)
+        eta = forward_params.get("eta", 0.0)
+        a_prompt = forward_params.get(
+            "a_prompt",
+            "best quality, extremely detailed,4k, HD, supper legible text,  clear text edges,  clear strokes, neat writing, no watermarks",
+        )
+        n_prompt = forward_params.get(
+            "n_prompt",
+            "low-res, bad anatomy, extra digit, fewer digits, cropped, worst quality, low quality, watermark, unreadable text, messy words, distorted text, disorganized writing, advertising picture",
+        )
+
+        prompt, texts = self.modify_prompt(prompt)
+        if prompt is None and texts is None:
+            return (
+                None,
+                -1,
+                "You have input Chinese prompt but the translator is not loaded!",
+                "",
+            )
+        n_lines = len(texts)
+        if mode in ["text-generation", "gen"]:
+            edit_image = np.ones((h, w, 3)) * 127.5  # empty mask image
+        elif mode in ["text-editing", "edit"]:
+            if draw_pos is None or ori_image is None:
+                return (
+                    None,
+                    -1,
+                    "Reference image and position image are needed for text editing!",
+                    "",
+                )
+            if isinstance(ori_image, str):
+                ori_image = cv2.imread(ori_image)[..., ::-1]
+                assert (
+                    ori_image is not None
+                ), f"Can't read ori_image image from{ori_image}!"
+            elif isinstance(ori_image, torch.Tensor):
+                ori_image = ori_image.cpu().numpy()
+            else:
+                assert isinstance(
+                    ori_image, np.ndarray
+                ), f"Unknown format of ori_image: {type(ori_image)}"
+            edit_image = ori_image.clip(1, 255)  # for mask reason
+            edit_image = check_channels(edit_image)
+            edit_image = resize_image(
+                edit_image, max_length=768
+            )  # make w h multiple of 64, resize if w or h > max_length
+            h, w = edit_image.shape[:2]  # change h, w by input ref_img
+        # preprocess pos_imgs(if numpy, make sure it's white pos in black bg)
+        if draw_pos is None:
+            pos_imgs = np.zeros((w, h, 1))
+        if isinstance(draw_pos, str):
+            draw_pos = cv2.imread(draw_pos)[..., ::-1]
+            assert draw_pos is not None, f"Can't read draw_pos image from{draw_pos}!"
+            pos_imgs = 255 - draw_pos
+        elif isinstance(draw_pos, torch.Tensor):
+            pos_imgs = draw_pos.cpu().numpy()
+        else:
+            assert isinstance(
+                draw_pos, np.ndarray
+            ), f"Unknown format of draw_pos: {type(draw_pos)}"
+        pos_imgs = pos_imgs[..., 0:1]
+        pos_imgs = cv2.convertScaleAbs(pos_imgs)
+        _, pos_imgs = cv2.threshold(pos_imgs, 254, 255, cv2.THRESH_BINARY)
+        # seprate pos_imgs
+        pos_imgs = self.separate_pos_imgs(pos_imgs, sort_priority)
+        if len(pos_imgs) == 0:
+            pos_imgs = [np.zeros((h, w, 1))]
+        if len(pos_imgs) < n_lines:
+            if n_lines == 1 and texts[0] == " ":
+                pass  # text-to-image without text
+            else:
+                return (
+                    None,
+                    -1,
+                    f"Found {len(pos_imgs)} positions that < needed {n_lines} from prompt, check and try again!",
+                    "",
+                )
+        elif len(pos_imgs) > n_lines:
+            str_warning = f"Warning: found {len(pos_imgs)} positions that > needed {n_lines} from prompt."
+        # get pre_pos, poly_list, hint that needed for anytext
+        pre_pos = []
+        poly_list = []
+        for input_pos in pos_imgs:
+            if input_pos.mean() != 0:
+                input_pos = (
+                    input_pos[..., np.newaxis]
+                    if len(input_pos.shape) == 2
+                    else input_pos
+                )
+                poly, pos_img = self.find_polygon(input_pos)
+                pre_pos += [pos_img / 255.0]
+                poly_list += [poly]
+            else:
+                pre_pos += [np.zeros((h, w, 1))]
+                poly_list += [None]
+        np_hint = np.sum(pre_pos, axis=0).clip(0, 1)
+        # prepare info dict
+        info = {}
+        info["glyphs"] = []
+        info["gly_line"] = []
+        info["positions"] = []
+        info["n_lines"] = [len(texts)] * img_count
+        gly_pos_imgs = []
+        for i in range(len(texts)):
+            text = texts[i]
+            if len(text) > max_chars:
+                str_warning = (
+                    f'"{text}" length > max_chars: {max_chars}, will be cut off...'
+                )
+                text = text[:max_chars]
+            gly_scale = 2
+            if pre_pos[i].mean() != 0:
+                gly_line = draw_glyph(self.font, text)
+                glyphs = draw_glyph2(
+                    self.font,
+                    text,
+                    poly_list[i],
+                    scale=gly_scale,
+                    width=w,
+                    height=h,
+                    add_space=False,
+                )
+                gly_pos_img = cv2.drawContours(
+                    glyphs * 255, [poly_list[i] * gly_scale], 0, (255, 255, 255), 1
+                )
+                if revise_pos:
+                    resize_gly = cv2.resize(
+                        glyphs, (pre_pos[i].shape[1], pre_pos[i].shape[0])
+                    )
+                    new_pos = cv2.morphologyEx(
+                        (resize_gly * 255).astype(np.uint8),
+                        cv2.MORPH_CLOSE,
+                        kernel=np.ones(
+                            (resize_gly.shape[0] // 10, resize_gly.shape[1] // 10),
+                            dtype=np.uint8,
+                        ),
+                        iterations=1,
+                    )
+                    new_pos = (
+                        new_pos[..., np.newaxis] if len(new_pos.shape) == 2 else new_pos
+                    )
+                    contours, _ = cv2.findContours(
+                        new_pos, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE
+                    )
+                    if len(contours) != 1:
+                        str_warning = f"Fail to revise position {i} to bounding rect, remain position unchanged..."
+                    else:
+                        rect = cv2.minAreaRect(contours[0])
+                        poly = np.int0(cv2.boxPoints(rect))
+                        pre_pos[i] = (
+                            cv2.drawContours(new_pos, [poly], -1, 255, -1) / 255.0
+                        )
+                        gly_pos_img = cv2.drawContours(
+                            glyphs * 255, [poly * gly_scale], 0, (255, 255, 255), 1
+                        )
+                gly_pos_imgs += [gly_pos_img]  # for show
+            else:
+                glyphs = np.zeros((h * gly_scale, w * gly_scale, 1))
+                gly_line = np.zeros((80, 512, 1))
+                gly_pos_imgs += [
+                    np.zeros((h * gly_scale, w * gly_scale, 1))
+                ]  # for show
+            pos = pre_pos[i]
+            info["glyphs"] += [self.arr2tensor(glyphs, img_count)]
+            info["gly_line"] += [self.arr2tensor(gly_line, img_count)]
+            info["positions"] += [self.arr2tensor(pos, img_count)]
+        # get masked_x
+        masked_img = ((edit_image.astype(np.float32) / 127.5) - 1.0) * (1 - np_hint)
+        masked_img = np.transpose(masked_img, (2, 0, 1))
+        masked_img = torch.from_numpy(masked_img.copy()).float().to(self.device)
+        if self.use_fp16:
+            masked_img = masked_img.half()
+        encoder_posterior = self.model.encode_first_stage(masked_img[None, ...])
+        masked_x = self.model.get_first_stage_encoding(encoder_posterior).detach()
+        if self.use_fp16:
+            masked_x = masked_x.half()
+        info["masked_x"] = torch.cat([masked_x for _ in range(img_count)], dim=0)
+
+        hint = self.arr2tensor(np_hint, img_count)
+        cond = self.model.get_learned_conditioning(
+            dict(
+                c_concat=[hint],
+                c_crossattn=[[prompt + " , " + a_prompt] * img_count],
+                text_info=info,
+            )
+        )
+        un_cond = self.model.get_learned_conditioning(
+            dict(c_concat=[hint], c_crossattn=[[n_prompt] * img_count], text_info=info)
+        )
+        shape = (4, h // 8, w // 8)
+        self.model.control_scales = [strength] * 13
+        samples, intermediates = self.ddim_sampler.sample(
+            ddim_steps,
+            img_count,
+            shape,
+            cond,
+            verbose=False,
+            eta=eta,
+            unconditional_guidance_scale=cfg_scale,
+            unconditional_conditioning=un_cond,
+        )
+        if self.use_fp16:
+            samples = samples.half()
+        x_samples = self.model.decode_first_stage(samples)
+        x_samples = (
+            (einops.rearrange(x_samples, "b c h w -> b h w c") * 127.5 + 127.5)
+            .cpu()
+            .numpy()
+            .clip(0, 255)
+            .astype(np.uint8)
+        )
+        results = [x_samples[i] for i in range(img_count)]
+        if (
+            mode == "edit" and False
+        ):  # replace backgound in text editing but not ideal yet
+            results = [r * np_hint + edit_image * (1 - np_hint) for r in results]
+            results = [r.clip(0, 255).astype(np.uint8) for r in results]
+        if len(gly_pos_imgs) > 0 and show_debug:
+            glyph_bs = np.stack(gly_pos_imgs, axis=2)
+            glyph_img = np.sum(glyph_bs, axis=2) * 255
+            glyph_img = glyph_img.clip(0, 255).astype(np.uint8)
+            results += [np.repeat(glyph_img, 3, axis=2)]
+        # debug_info
+        if not show_debug:
+            debug_info = ""
+        else:
+            input_prompt = prompt
+            for t in texts:
+                input_prompt = input_prompt.replace("*", f'"{t}"', 1)
+            debug_info = f'<span style="color:black;font-size:18px">Prompt: </span>{input_prompt}<br> \
+                           <span style="color:black;font-size:18px">Size: </span>{w}x{h}<br> \
+                           <span style="color:black;font-size:18px">Image Count: </span>{img_count}<br> \
+                           <span style="color:black;font-size:18px">Seed: </span>{seed}<br> \
+                           <span style="color:black;font-size:18px">Use FP16: </span>{self.use_fp16}<br> \
+                           <span style="color:black;font-size:18px">Cost Time: </span>{(time.time()-tic):.2f}s'
+        rst_code = 1 if str_warning else 0
+        return results, rst_code, str_warning, debug_info
+
+    def init_model(self):
+        font_path = self.font_path
+        self.font = ImageFont.truetype(font_path, size=60)
+        cfg_path = self.cfg_path
+        ckpt_path = os.path.join(self.model_dir, "anytext_v1.1.ckpt")
+        clip_path = os.path.join(self.model_dir, "clip-vit-large-patch14")
+        self.model = create_model(
+            cfg_path,
+            device=self.device,
+            cond_stage_path=clip_path,
+            use_fp16=self.use_fp16,
+        )
+        if self.use_fp16:
+            self.model = self.model.half()
+        self.model.load_state_dict(
+            load_state_dict(ckpt_path, location=self.device), strict=False
+        )
+        self.model.eval()
+        self.model = self.model.to(self.device)
+        self.ddim_sampler = DDIMSampler(self.model, device=self.device)
+
+    def modify_prompt(self, prompt):
+        prompt = prompt.replace("“", '"')
+        prompt = prompt.replace("”", '"')
+        p = '"(.*?)"'
+        strs = re.findall(p, prompt)
+        if len(strs) == 0:
+            strs = [" "]
+        else:
+            for s in strs:
+                prompt = prompt.replace(f'"{s}"', f" {PLACE_HOLDER} ", 1)
+        # if self.is_chinese(prompt):
+        #     if self.trans_pipe is None:
+        #         return None, None
+        #     old_prompt = prompt
+        #     prompt = self.trans_pipe(input=prompt + " .")["translation"][:-1]
+        #     print(f"Translate: {old_prompt} --> {prompt}")
+        return prompt, strs
+
+    # def is_chinese(self, text):
+    #     text = checker._clean_text(text)
+    #     for char in text:
+    #         cp = ord(char)
+    #         if checker._is_chinese_char(cp):
+    #             return True
+    #     return False
+
+    def separate_pos_imgs(self, img, sort_priority, gap=102):
+        num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(img)
+        components = []
+        for label in range(1, num_labels):
+            component = np.zeros_like(img)
+            component[labels == label] = 255
+            components.append((component, centroids[label]))
+        if sort_priority == "↕":
+            fir, sec = 1, 0  # top-down first
+        elif sort_priority == "↔":
+            fir, sec = 0, 1  # left-right first
+        components.sort(key=lambda c: (c[1][fir] // gap, c[1][sec] // gap))
+        sorted_components = [c[0] for c in components]
+        return sorted_components
+
+    def find_polygon(self, image, min_rect=False):
+        contours, hierarchy = cv2.findContours(
+            image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE
+        )
+        max_contour = max(contours, key=cv2.contourArea)  # get contour with max area
+        if min_rect:
+            # get minimum enclosing rectangle
+            rect = cv2.minAreaRect(max_contour)
+            poly = np.int0(cv2.boxPoints(rect))
+        else:
+            # get approximate polygon
+            epsilon = 0.01 * cv2.arcLength(max_contour, True)
+            poly = cv2.approxPolyDP(max_contour, epsilon, True)
+            n, _, xy = poly.shape
+            poly = poly.reshape(n, xy)
+        cv2.drawContours(image, [poly], -1, 255, -1)
+        return poly, image
+
+    def arr2tensor(self, arr, bs):
+        arr = np.transpose(arr, (2, 0, 1))
+        _arr = torch.from_numpy(arr.copy()).float().to(self.device)
+        if self.use_fp16:
+            _arr = _arr.half()
+        _arr = torch.stack([_arr for _ in range(bs)], dim=0)
+        return _arr

iopaint/model/anytext/anytext_sd15.yaml

@@ -0,0 +1,99 @@
+model:
+  target: iopaint.model.anytext.cldm.cldm.ControlLDM
+  params:
+    linear_start: 0.00085
+    linear_end: 0.0120
+    num_timesteps_cond: 1
+    log_every_t: 200
+    timesteps: 1000
+    first_stage_key: "img"
+    cond_stage_key: "caption"
+    control_key: "hint"
+    glyph_key: "glyphs"
+    position_key: "positions"
+    image_size: 64
+    channels: 4
+    cond_stage_trainable: true  # need be true when embedding_manager is valid
+    conditioning_key: crossattn
+    monitor: val/loss_simple_ema
+    scale_factor: 0.18215
+    use_ema: False
+    only_mid_control: False
+    loss_alpha: 0  # perceptual loss, 0.003
+    loss_beta: 0  # ctc loss
+    latin_weight: 1.0  # latin text line may need smaller weigth
+    with_step_weight: true
+    use_vae_upsample: true
+    embedding_manager_config:
+      target: iopaint.model.anytext.cldm.embedding_manager.EmbeddingManager
+      params:
+        valid: true  # v6
+        emb_type: ocr  # ocr, vit, conv
+        glyph_channels: 1
+        position_channels: 1
+        add_pos: false
+        placeholder_string: '*'
+
+    control_stage_config:
+      target: iopaint.model.anytext.cldm.cldm.ControlNet
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        model_channels: 320
+        glyph_channels: 1
+        position_channels: 1
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    unet_config:
+      target: iopaint.model.anytext.cldm.cldm.ControlledUnetModel
+      params:
+        image_size: 32 # unused
+        in_channels: 4
+        out_channels: 4
+        model_channels: 320
+        attention_resolutions: [ 4, 2, 1 ]
+        num_res_blocks: 2
+        channel_mult: [ 1, 2, 4, 4 ]
+        num_heads: 8
+        use_spatial_transformer: True
+        transformer_depth: 1
+        context_dim: 768
+        use_checkpoint: True
+        legacy: False
+
+    first_stage_config:
+      target: iopaint.model.anytext.ldm.models.autoencoder.AutoencoderKL
+      params:
+        embed_dim: 4
+        monitor: val/rec_loss
+        ddconfig:
+          double_z: true
+          z_channels: 4
+          resolution: 256
+          in_channels: 3
+          out_ch: 3
+          ch: 128
+          ch_mult:
+          - 1
+          - 2
+          - 4
+          - 4
+          num_res_blocks: 2
+          attn_resolutions: []
+          dropout: 0.0
+        lossconfig:
+          target: torch.nn.Identity
+
+    cond_stage_config:
+      target: iopaint.model.anytext.ldm.modules.encoders.modules.FrozenCLIPEmbedderT3
+      params:
+        version: ./models/clip-vit-large-patch14
+        use_vision: false  # v6

iopaint/model/anytext/cldm/cldm.py

@@ -0,0 +1,630 @@
+import os
+from pathlib import Path
+
+import einops
+import torch
+import torch as th
+import torch.nn as nn
+import copy
+from easydict import EasyDict as edict
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import (
+    conv_nd,
+    linear,
+    zero_module,
+    timestep_embedding,
+)
+
+from einops import rearrange, repeat
+from iopaint.model.anytext.ldm.modules.attention import SpatialTransformer
+from iopaint.model.anytext.ldm.modules.diffusionmodules.openaimodel import UNetModel, TimestepEmbedSequential, ResBlock, Downsample, AttentionBlock
+from iopaint.model.anytext.ldm.models.diffusion.ddpm import LatentDiffusion
+from iopaint.model.anytext.ldm.util import log_txt_as_img, exists, instantiate_from_config
+from iopaint.model.anytext.ldm.models.diffusion.ddim import DDIMSampler
+from iopaint.model.anytext.ldm.modules.distributions.distributions import DiagonalGaussianDistribution
+from .recognizer import TextRecognizer, create_predictor
+
+CURRENT_DIR = Path(os.path.dirname(os.path.abspath(__file__)))
+
+
+def count_parameters(model):
+    return sum(p.numel() for p in model.parameters() if p.requires_grad)
+
+
+class ControlledUnetModel(UNetModel):
+    def forward(self, x, timesteps=None, context=None, control=None, only_mid_control=False, **kwargs):
+        hs = []
+        with torch.no_grad():
+            t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+            if self.use_fp16:
+                t_emb = t_emb.half()
+            emb = self.time_embed(t_emb)
+            h = x.type(self.dtype)
+            for module in self.input_blocks:
+                h = module(h, emb, context)
+                hs.append(h)
+            h = self.middle_block(h, emb, context)
+
+        if control is not None:
+            h += control.pop()
+
+        for i, module in enumerate(self.output_blocks):
+            if only_mid_control or control is None:
+                h = torch.cat([h, hs.pop()], dim=1)
+            else:
+                h = torch.cat([h, hs.pop() + control.pop()], dim=1)
+            h = module(h, emb, context)
+
+        h = h.type(x.dtype)
+        return self.out(h)
+
+
+class ControlNet(nn.Module):
+    def __init__(
+            self,
+            image_size,
+            in_channels,
+            model_channels,
+            glyph_channels,
+            position_channels,
+            num_res_blocks,
+            attention_resolutions,
+            dropout=0,
+            channel_mult=(1, 2, 4, 8),
+            conv_resample=True,
+            dims=2,
+            use_checkpoint=False,
+            use_fp16=False,
+            num_heads=-1,
+            num_head_channels=-1,
+            num_heads_upsample=-1,
+            use_scale_shift_norm=False,
+            resblock_updown=False,
+            use_new_attention_order=False,
+            use_spatial_transformer=False,  # custom transformer support
+            transformer_depth=1,  # custom transformer support
+            context_dim=None,  # custom transformer support
+            n_embed=None,  # custom support for prediction of discrete ids into codebook of first stage vq model
+            legacy=True,
+            disable_self_attentions=None,
+            num_attention_blocks=None,
+            disable_middle_self_attn=False,
+            use_linear_in_transformer=False,
+    ):
+        super().__init__()
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            from omegaconf.listconfig import ListConfig
+            if type(context_dim) == ListConfig:
+                context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+        self.dims = dims
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        if isinstance(num_res_blocks, int):
+            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
+        else:
+            if len(num_res_blocks) != len(channel_mult):
+                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
+                                 "as a list/tuple (per-level) with the same length as channel_mult")
+            self.num_res_blocks = num_res_blocks
+        if disable_self_attentions is not None:
+            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
+            assert len(disable_self_attentions) == len(channel_mult)
+        if num_attention_blocks is not None:
+            assert len(num_attention_blocks) == len(self.num_res_blocks)
+            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
+            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
+                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
+                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
+                  f"attention will still not be set.")
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.use_checkpoint = use_checkpoint
+        self.use_fp16 = use_fp16
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.predict_codebook_ids = n_embed is not None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
+                )
+            ]
+        )
+        self.zero_convs = nn.ModuleList([self.make_zero_conv(model_channels)])
+
+        self.glyph_block = TimestepEmbedSequential(
+            conv_nd(dims, glyph_channels, 8, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 8, 8, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 8, 16, 3, padding=1, stride=2),
+            nn.SiLU(),
+            conv_nd(dims, 16, 16, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 16, 32, 3, padding=1, stride=2),
+            nn.SiLU(),
+            conv_nd(dims, 32, 32, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 32, 96, 3, padding=1, stride=2),
+            nn.SiLU(),
+            conv_nd(dims, 96, 96, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 96, 256, 3, padding=1, stride=2),
+            nn.SiLU(),
+        )
+
+        self.position_block = TimestepEmbedSequential(
+            conv_nd(dims, position_channels, 8, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 8, 8, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 8, 16, 3, padding=1, stride=2),
+            nn.SiLU(),
+            conv_nd(dims, 16, 16, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 16, 32, 3, padding=1, stride=2),
+            nn.SiLU(),
+            conv_nd(dims, 32, 32, 3, padding=1),
+            nn.SiLU(),
+            conv_nd(dims, 32, 64, 3, padding=1, stride=2),
+            nn.SiLU(),
+        )
+
+        self.fuse_block = zero_module(conv_nd(dims, 256+64+4, model_channels, 3, padding=1))
+
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for nr in range(self.num_res_blocks[level]):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = mult * model_channels
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        # num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
+                        layers.append(
+                            AttentionBlock(
+                                ch,
+                                use_checkpoint=use_checkpoint,
+                                num_heads=num_heads,
+                                num_head_channels=dim_head,
+                                use_new_attention_order=use_new_attention_order,
+                            ) if not use_spatial_transformer else SpatialTransformer(
+                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint
+                            )
+                        )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self.zero_convs.append(self.make_zero_conv(ch))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                self.zero_convs.append(self.make_zero_conv(ch))
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            # num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=dim_head,
+                use_new_attention_order=use_new_attention_order,
+            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
+                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
+                use_checkpoint=use_checkpoint
+            ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self.middle_block_out = self.make_zero_conv(ch)
+        self._feature_size += ch
+
+    def make_zero_conv(self, channels):
+        return TimestepEmbedSequential(zero_module(conv_nd(self.dims, channels, channels, 1, padding=0)))
+
+    def forward(self, x, hint, text_info, timesteps, context, **kwargs):
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+        if self.use_fp16:
+            t_emb = t_emb.half()
+        emb = self.time_embed(t_emb)
+
+        # guided_hint from text_info
+        B, C, H, W = x.shape
+        glyphs = torch.cat(text_info['glyphs'], dim=1).sum(dim=1, keepdim=True)
+        positions = torch.cat(text_info['positions'], dim=1).sum(dim=1, keepdim=True)
+        enc_glyph = self.glyph_block(glyphs, emb, context)
+        enc_pos = self.position_block(positions, emb, context)
+        guided_hint = self.fuse_block(torch.cat([enc_glyph, enc_pos, text_info['masked_x']], dim=1))
+
+        outs = []
+
+        h = x.type(self.dtype)
+        for module, zero_conv in zip(self.input_blocks, self.zero_convs):
+            if guided_hint is not None:
+                h = module(h, emb, context)
+                h += guided_hint
+                guided_hint = None
+            else:
+                h = module(h, emb, context)
+            outs.append(zero_conv(h, emb, context))
+
+        h = self.middle_block(h, emb, context)
+        outs.append(self.middle_block_out(h, emb, context))
+
+        return outs
+
+
+class ControlLDM(LatentDiffusion):
+
+    def __init__(self, control_stage_config, control_key, glyph_key, position_key, only_mid_control, loss_alpha=0, loss_beta=0, with_step_weight=False, use_vae_upsample=False, latin_weight=1.0, embedding_manager_config=None, *args, **kwargs):
+        self.use_fp16 = kwargs.pop('use_fp16', False)
+        super().__init__(*args, **kwargs)
+        self.control_model = instantiate_from_config(control_stage_config)
+        self.control_key = control_key
+        self.glyph_key = glyph_key
+        self.position_key = position_key
+        self.only_mid_control = only_mid_control
+        self.control_scales = [1.0] * 13
+        self.loss_alpha = loss_alpha
+        self.loss_beta = loss_beta
+        self.with_step_weight = with_step_weight
+        self.use_vae_upsample = use_vae_upsample
+        self.latin_weight = latin_weight
+
+        if embedding_manager_config is not None and embedding_manager_config.params.valid:
+            self.embedding_manager = self.instantiate_embedding_manager(embedding_manager_config, self.cond_stage_model)
+            for param in self.embedding_manager.embedding_parameters():
+                param.requires_grad = True
+        else:
+            self.embedding_manager = None
+        if self.loss_alpha > 0 or self.loss_beta > 0 or self.embedding_manager:
+            if embedding_manager_config.params.emb_type == 'ocr':
+                self.text_predictor = create_predictor().eval()
+                args = edict()
+                args.rec_image_shape = "3, 48, 320"
+                args.rec_batch_num = 6
+                args.rec_char_dict_path = str(CURRENT_DIR.parent / "ocr_recog" / "ppocr_keys_v1.txt")
+                args.use_fp16 = self.use_fp16
+                self.cn_recognizer = TextRecognizer(args, self.text_predictor)
+                for param in self.text_predictor.parameters():
+                    param.requires_grad = False
+                if self.embedding_manager:
+                    self.embedding_manager.recog = self.cn_recognizer
+
+    @torch.no_grad()
+    def get_input(self, batch, k, bs=None, *args, **kwargs):
+        if self.embedding_manager is None:  # fill in full caption
+            self.fill_caption(batch)
+        x, c, mx = super().get_input(batch, self.first_stage_key, mask_k='masked_img', *args, **kwargs)
+        control = batch[self.control_key]  # for log_images and loss_alpha, not real control
+        if bs is not None:
+            control = control[:bs]
+        control = control.to(self.device)
+        control = einops.rearrange(control, 'b h w c -> b c h w')
+        control = control.to(memory_format=torch.contiguous_format).float()
+
+        inv_mask = batch['inv_mask']
+        if bs is not None:
+            inv_mask = inv_mask[:bs]
+        inv_mask = inv_mask.to(self.device)
+        inv_mask = einops.rearrange(inv_mask, 'b h w c -> b c h w')
+        inv_mask = inv_mask.to(memory_format=torch.contiguous_format).float()
+
+        glyphs = batch[self.glyph_key]
+        gly_line = batch['gly_line']
+        positions = batch[self.position_key]
+        n_lines = batch['n_lines']
+        language = batch['language']
+        texts = batch['texts']
+        assert len(glyphs) == len(positions)
+        for i in range(len(glyphs)):
+            if bs is not None:
+                glyphs[i] = glyphs[i][:bs]
+                gly_line[i] = gly_line[i][:bs]
+                positions[i] = positions[i][:bs]
+                n_lines = n_lines[:bs]
+            glyphs[i] = glyphs[i].to(self.device)
+            gly_line[i] = gly_line[i].to(self.device)
+            positions[i] = positions[i].to(self.device)
+            glyphs[i] = einops.rearrange(glyphs[i], 'b h w c -> b c h w')
+            gly_line[i] = einops.rearrange(gly_line[i], 'b h w c -> b c h w')
+            positions[i] = einops.rearrange(positions[i], 'b h w c -> b c h w')
+            glyphs[i] = glyphs[i].to(memory_format=torch.contiguous_format).float()
+            gly_line[i] = gly_line[i].to(memory_format=torch.contiguous_format).float()
+            positions[i] = positions[i].to(memory_format=torch.contiguous_format).float()
+        info = {}
+        info['glyphs'] = glyphs
+        info['positions'] = positions
+        info['n_lines'] = n_lines
+        info['language'] = language
+        info['texts'] = texts
+        info['img'] = batch['img']  # nhwc, (-1,1)
+        info['masked_x'] = mx
+        info['gly_line'] = gly_line
+        info['inv_mask'] = inv_mask
+        return x, dict(c_crossattn=[c], c_concat=[control], text_info=info)
+
+    def apply_model(self, x_noisy, t, cond, *args, **kwargs):
+        assert isinstance(cond, dict)
+        diffusion_model = self.model.diffusion_model
+        _cond = torch.cat(cond['c_crossattn'], 1)
+        _hint = torch.cat(cond['c_concat'], 1)
+        if self.use_fp16:
+            x_noisy = x_noisy.half()
+        control = self.control_model(x=x_noisy, timesteps=t, context=_cond, hint=_hint, text_info=cond['text_info'])
+        control = [c * scale for c, scale in zip(control, self.control_scales)]
+        eps = diffusion_model(x=x_noisy, timesteps=t, context=_cond, control=control, only_mid_control=self.only_mid_control)
+
+        return eps
+
+    def instantiate_embedding_manager(self, config, embedder):
+        model = instantiate_from_config(config, embedder=embedder)
+        return model
+
+    @torch.no_grad()
+    def get_unconditional_conditioning(self, N):
+        return self.get_learned_conditioning(dict(c_crossattn=[[""] * N], text_info=None))
+
+    def get_learned_conditioning(self, c):
+        if self.cond_stage_forward is None:
+            if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode):
+                if self.embedding_manager is not None and c['text_info'] is not None:
+                    self.embedding_manager.encode_text(c['text_info'])
+                if isinstance(c, dict):
+                    cond_txt = c['c_crossattn'][0]
+                else:
+                    cond_txt = c
+                if self.embedding_manager is not None:
+                    cond_txt = self.cond_stage_model.encode(cond_txt, embedding_manager=self.embedding_manager)
+                else:
+                    cond_txt = self.cond_stage_model.encode(cond_txt)
+                if isinstance(c, dict):
+                    c['c_crossattn'][0] = cond_txt
+                else:
+                    c = cond_txt
+                if isinstance(c, DiagonalGaussianDistribution):
+                    c = c.mode()
+            else:
+                c = self.cond_stage_model(c)
+        else:
+            assert hasattr(self.cond_stage_model, self.cond_stage_forward)
+            c = getattr(self.cond_stage_model, self.cond_stage_forward)(c)
+        return c
+
+    def fill_caption(self, batch, place_holder='*'):
+        bs = len(batch['n_lines'])
+        cond_list = copy.deepcopy(batch[self.cond_stage_key])
+        for i in range(bs):
+            n_lines = batch['n_lines'][i]
+            if n_lines == 0:
+                continue
+            cur_cap = cond_list[i]
+            for j in range(n_lines):
+                r_txt = batch['texts'][j][i]
+                cur_cap = cur_cap.replace(place_holder, f'"{r_txt}"', 1)
+            cond_list[i] = cur_cap
+        batch[self.cond_stage_key] = cond_list
+
+    @torch.no_grad()
+    def log_images(self, batch, N=4, n_row=2, sample=False, ddim_steps=50, ddim_eta=0.0, return_keys=None,
+                   quantize_denoised=True, inpaint=True, plot_denoise_rows=False, plot_progressive_rows=True,
+                   plot_diffusion_rows=False, unconditional_guidance_scale=9.0, unconditional_guidance_label=None,
+                   use_ema_scope=True,
+                   **kwargs):
+        use_ddim = ddim_steps is not None
+
+        log = dict()
+        z, c = self.get_input(batch, self.first_stage_key, bs=N)
+        if self.cond_stage_trainable:
+            with torch.no_grad():
+                c = self.get_learned_conditioning(c)
+        c_crossattn = c["c_crossattn"][0][:N]
+        c_cat = c["c_concat"][0][:N]
+        text_info = c["text_info"]
+        text_info['glyphs'] = [i[:N] for i in text_info['glyphs']]
+        text_info['gly_line'] = [i[:N] for i in text_info['gly_line']]
+        text_info['positions'] = [i[:N] for i in text_info['positions']]
+        text_info['n_lines'] = text_info['n_lines'][:N]
+        text_info['masked_x'] = text_info['masked_x'][:N]
+        text_info['img'] = text_info['img'][:N]
+
+        N = min(z.shape[0], N)
+        n_row = min(z.shape[0], n_row)
+        log["reconstruction"] = self.decode_first_stage(z)
+        log["masked_image"] = self.decode_first_stage(text_info['masked_x'])
+        log["control"] = c_cat * 2.0 - 1.0
+        log["img"] = text_info['img'].permute(0, 3, 1, 2)  # log source image if needed
+        # get glyph
+        glyph_bs = torch.stack(text_info['glyphs'])
+        glyph_bs = torch.sum(glyph_bs, dim=0) * 2.0 - 1.0
+        log["glyph"] = torch.nn.functional.interpolate(glyph_bs, size=(512, 512), mode='bilinear', align_corners=True,)
+        # fill caption
+        if not self.embedding_manager:
+            self.fill_caption(batch)
+        captions = batch[self.cond_stage_key]
+        log["conditioning"] = log_txt_as_img((512, 512), captions, size=16)
+
+        if plot_diffusion_rows:
+            # get diffusion row
+            diffusion_row = list()
+            z_start = z[:n_row]
+            for t in range(self.num_timesteps):
+                if t % self.log_every_t == 0 or t == self.num_timesteps - 1:
+                    t = repeat(torch.tensor([t]), '1 -> b', b=n_row)
+                    t = t.to(self.device).long()
+                    noise = torch.randn_like(z_start)
+                    z_noisy = self.q_sample(x_start=z_start, t=t, noise=noise)
+                    diffusion_row.append(self.decode_first_stage(z_noisy))
+
+            diffusion_row = torch.stack(diffusion_row)  # n_log_step, n_row, C, H, W
+            diffusion_grid = rearrange(diffusion_row, 'n b c h w -> b n c h w')
+            diffusion_grid = rearrange(diffusion_grid, 'b n c h w -> (b n) c h w')
+            diffusion_grid = make_grid(diffusion_grid, nrow=diffusion_row.shape[0])
+            log["diffusion_row"] = diffusion_grid
+
+        if sample:
+            # get denoise row
+            samples, z_denoise_row = self.sample_log(cond={"c_concat": [c_cat], "c_crossattn": [c], "text_info": text_info},
+                                                     batch_size=N, ddim=use_ddim,
+                                                     ddim_steps=ddim_steps, eta=ddim_eta)
+            x_samples = self.decode_first_stage(samples)
+            log["samples"] = x_samples
+            if plot_denoise_rows:
+                denoise_grid = self._get_denoise_row_from_list(z_denoise_row)
+                log["denoise_row"] = denoise_grid
+
+        if unconditional_guidance_scale > 1.0:
+            uc_cross = self.get_unconditional_conditioning(N)
+            uc_cat = c_cat  # torch.zeros_like(c_cat)
+            uc_full = {"c_concat": [uc_cat], "c_crossattn": [uc_cross['c_crossattn'][0]], "text_info": text_info}
+            samples_cfg, tmps = self.sample_log(cond={"c_concat": [c_cat], "c_crossattn": [c_crossattn], "text_info": text_info},
+                                                batch_size=N, ddim=use_ddim,
+                                                ddim_steps=ddim_steps, eta=ddim_eta,
+                                                unconditional_guidance_scale=unconditional_guidance_scale,
+                                                unconditional_conditioning=uc_full,
+                                                )
+            x_samples_cfg = self.decode_first_stage(samples_cfg)
+            log[f"samples_cfg_scale_{unconditional_guidance_scale:.2f}"] = x_samples_cfg
+            pred_x0 = False  # wether log pred_x0
+            if pred_x0:
+                for idx in range(len(tmps['pred_x0'])):
+                    pred_x0 = self.decode_first_stage(tmps['pred_x0'][idx])
+                    log[f"pred_x0_{tmps['index'][idx]}"] = pred_x0
+
+        return log
+
+    @torch.no_grad()
+    def sample_log(self, cond, batch_size, ddim, ddim_steps, **kwargs):
+        ddim_sampler = DDIMSampler(self)
+        b, c, h, w = cond["c_concat"][0].shape
+        shape = (self.channels, h // 8, w // 8)
+        samples, intermediates = ddim_sampler.sample(ddim_steps, batch_size, shape, cond, verbose=False, log_every_t=5, **kwargs)
+        return samples, intermediates
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        params = list(self.control_model.parameters())
+        if self.embedding_manager:
+            params += list(self.embedding_manager.embedding_parameters())
+        if not self.sd_locked:
+            # params += list(self.model.diffusion_model.input_blocks.parameters())
+            # params += list(self.model.diffusion_model.middle_block.parameters())
+            params += list(self.model.diffusion_model.output_blocks.parameters())
+            params += list(self.model.diffusion_model.out.parameters())
+        if self.unlockKV:
+            nCount = 0
+            for name, param in self.model.diffusion_model.named_parameters():
+                if 'attn2.to_k' in name or 'attn2.to_v' in name:
+                    params += [param]
+                    nCount += 1
+            print(f'Cross attention is unlocked, and {nCount} Wk or Wv are added to potimizers!!!')
+
+        opt = torch.optim.AdamW(params, lr=lr)
+        return opt
+
+    def low_vram_shift(self, is_diffusing):
+        if is_diffusing:
+            self.model = self.model.cuda()
+            self.control_model = self.control_model.cuda()
+            self.first_stage_model = self.first_stage_model.cpu()
+            self.cond_stage_model = self.cond_stage_model.cpu()
+        else:
+            self.model = self.model.cpu()
+            self.control_model = self.control_model.cpu()
+            self.first_stage_model = self.first_stage_model.cuda()
+            self.cond_stage_model = self.cond_stage_model.cuda()

iopaint/model/anytext/cldm/ddim_hacked.py

@@ -0,0 +1,486 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import (
+    make_ddim_sampling_parameters,
+    make_ddim_timesteps,
+    noise_like,
+    extract_into_tensor,
+)
+
+
+class DDIMSampler(object):
+    def __init__(self, model, device, schedule="linear", **kwargs):
+        super().__init__()
+        self.device = device
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device(self.device):
+                attr = attr.to(torch.device(self.device))
+        setattr(self, name, attr)
+
+    def make_schedule(
+        self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0.0, verbose=True
+    ):
+        self.ddim_timesteps = make_ddim_timesteps(
+            ddim_discr_method=ddim_discretize,
+            num_ddim_timesteps=ddim_num_steps,
+            num_ddpm_timesteps=self.ddpm_num_timesteps,
+            verbose=verbose,
+        )
+        alphas_cumprod = self.model.alphas_cumprod
+        assert (
+            alphas_cumprod.shape[0] == self.ddpm_num_timesteps
+        ), "alphas have to be defined for each timestep"
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.device)
+
+        self.register_buffer("betas", to_torch(self.model.betas))
+        self.register_buffer("alphas_cumprod", to_torch(alphas_cumprod))
+        self.register_buffer(
+            "alphas_cumprod_prev", to_torch(self.model.alphas_cumprod_prev)
+        )
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer(
+            "sqrt_alphas_cumprod", to_torch(np.sqrt(alphas_cumprod.cpu()))
+        )
+        self.register_buffer(
+            "sqrt_one_minus_alphas_cumprod",
+            to_torch(np.sqrt(1.0 - alphas_cumprod.cpu())),
+        )
+        self.register_buffer(
+            "log_one_minus_alphas_cumprod", to_torch(np.log(1.0 - alphas_cumprod.cpu()))
+        )
+        self.register_buffer(
+            "sqrt_recip_alphas_cumprod", to_torch(np.sqrt(1.0 / alphas_cumprod.cpu()))
+        )
+        self.register_buffer(
+            "sqrt_recipm1_alphas_cumprod",
+            to_torch(np.sqrt(1.0 / alphas_cumprod.cpu() - 1)),
+        )
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(
+            alphacums=alphas_cumprod.cpu(),
+            ddim_timesteps=self.ddim_timesteps,
+            eta=ddim_eta,
+            verbose=verbose,
+        )
+        self.register_buffer("ddim_sigmas", ddim_sigmas)
+        self.register_buffer("ddim_alphas", ddim_alphas)
+        self.register_buffer("ddim_alphas_prev", ddim_alphas_prev)
+        self.register_buffer("ddim_sqrt_one_minus_alphas", np.sqrt(1.0 - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev)
+            / (1 - self.alphas_cumprod)
+            * (1 - self.alphas_cumprod / self.alphas_cumprod_prev)
+        )
+        self.register_buffer(
+            "ddim_sigmas_for_original_num_steps", sigmas_for_original_sampling_steps
+        )
+
+    @torch.no_grad()
+    def sample(
+        self,
+        S,
+        batch_size,
+        shape,
+        conditioning=None,
+        callback=None,
+        normals_sequence=None,
+        img_callback=None,
+        quantize_x0=False,
+        eta=0.0,
+        mask=None,
+        x0=None,
+        temperature=1.0,
+        noise_dropout=0.0,
+        score_corrector=None,
+        corrector_kwargs=None,
+        verbose=True,
+        x_T=None,
+        log_every_t=100,
+        unconditional_guidance_scale=1.0,
+        unconditional_conditioning=None,  # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+        dynamic_threshold=None,
+        ucg_schedule=None,
+        **kwargs,
+    ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                ctmp = conditioning[list(conditioning.keys())[0]]
+                while isinstance(ctmp, list):
+                    ctmp = ctmp[0]
+                cbs = ctmp.shape[0]
+                if cbs != batch_size:
+                    print(
+                        f"Warning: Got {cbs} conditionings but batch-size is {batch_size}"
+                    )
+
+            elif isinstance(conditioning, list):
+                for ctmp in conditioning:
+                    if ctmp.shape[0] != batch_size:
+                        print(
+                            f"Warning: Got {cbs} conditionings but batch-size is {batch_size}"
+                        )
+
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(
+                        f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}"
+                    )
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f"Data shape for DDIM sampling is {size}, eta {eta}")
+
+        samples, intermediates = self.ddim_sampling(
+            conditioning,
+            size,
+            callback=callback,
+            img_callback=img_callback,
+            quantize_denoised=quantize_x0,
+            mask=mask,
+            x0=x0,
+            ddim_use_original_steps=False,
+            noise_dropout=noise_dropout,
+            temperature=temperature,
+            score_corrector=score_corrector,
+            corrector_kwargs=corrector_kwargs,
+            x_T=x_T,
+            log_every_t=log_every_t,
+            unconditional_guidance_scale=unconditional_guidance_scale,
+            unconditional_conditioning=unconditional_conditioning,
+            dynamic_threshold=dynamic_threshold,
+            ucg_schedule=ucg_schedule,
+        )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def ddim_sampling(
+        self,
+        cond,
+        shape,
+        x_T=None,
+        ddim_use_original_steps=False,
+        callback=None,
+        timesteps=None,
+        quantize_denoised=False,
+        mask=None,
+        x0=None,
+        img_callback=None,
+        log_every_t=100,
+        temperature=1.0,
+        noise_dropout=0.0,
+        score_corrector=None,
+        corrector_kwargs=None,
+        unconditional_guidance_scale=1.0,
+        unconditional_conditioning=None,
+        dynamic_threshold=None,
+        ucg_schedule=None,
+    ):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = (
+                self.ddpm_num_timesteps
+                if ddim_use_original_steps
+                else self.ddim_timesteps
+            )
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = (
+                int(
+                    min(timesteps / self.ddim_timesteps.shape[0], 1)
+                    * self.ddim_timesteps.shape[0]
+                )
+                - 1
+            )
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {"x_inter": [img], "pred_x0": [img]}
+        time_range = (
+            reversed(range(0, timesteps))
+            if ddim_use_original_steps
+            else np.flip(timesteps)
+        )
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc="DDIM Sampler", total=total_steps)
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(
+                    x0, ts
+                )  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1.0 - mask) * img
+
+            if ucg_schedule is not None:
+                assert len(ucg_schedule) == len(time_range)
+                unconditional_guidance_scale = ucg_schedule[i]
+
+            outs = self.p_sample_ddim(
+                img,
+                cond,
+                ts,
+                index=index,
+                use_original_steps=ddim_use_original_steps,
+                quantize_denoised=quantize_denoised,
+                temperature=temperature,
+                noise_dropout=noise_dropout,
+                score_corrector=score_corrector,
+                corrector_kwargs=corrector_kwargs,
+                unconditional_guidance_scale=unconditional_guidance_scale,
+                unconditional_conditioning=unconditional_conditioning,
+                dynamic_threshold=dynamic_threshold,
+            )
+            img, pred_x0 = outs
+            if callback:
+                callback(i)
+            if img_callback:
+                img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates["x_inter"].append(img)
+                intermediates["pred_x0"].append(pred_x0)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_ddim(
+        self,
+        x,
+        c,
+        t,
+        index,
+        repeat_noise=False,
+        use_original_steps=False,
+        quantize_denoised=False,
+        temperature=1.0,
+        noise_dropout=0.0,
+        score_corrector=None,
+        corrector_kwargs=None,
+        unconditional_guidance_scale=1.0,
+        unconditional_conditioning=None,
+        dynamic_threshold=None,
+    ):
+        b, *_, device = *x.shape, x.device
+
+        if unconditional_conditioning is None or unconditional_guidance_scale == 1.0:
+            model_output = self.model.apply_model(x, t, c)
+        else:
+            model_t = self.model.apply_model(x, t, c)
+            model_uncond = self.model.apply_model(x, t, unconditional_conditioning)
+            model_output = model_uncond + unconditional_guidance_scale * (
+                model_t - model_uncond
+            )
+
+        if self.model.parameterization == "v":
+            e_t = self.model.predict_eps_from_z_and_v(x, t, model_output)
+        else:
+            e_t = model_output
+
+        if score_corrector is not None:
+            assert self.model.parameterization == "eps", "not implemented"
+            e_t = score_corrector.modify_score(
+                self.model, e_t, x, t, c, **corrector_kwargs
+            )
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = (
+            self.model.alphas_cumprod_prev
+            if use_original_steps
+            else self.ddim_alphas_prev
+        )
+        sqrt_one_minus_alphas = (
+            self.model.sqrt_one_minus_alphas_cumprod
+            if use_original_steps
+            else self.ddim_sqrt_one_minus_alphas
+        )
+        sigmas = (
+            self.model.ddim_sigmas_for_original_num_steps
+            if use_original_steps
+            else self.ddim_sigmas
+        )
+        # select parameters corresponding to the currently considered timestep
+        a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+        a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+        sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+        sqrt_one_minus_at = torch.full(
+            (b, 1, 1, 1), sqrt_one_minus_alphas[index], device=device
+        )
+
+        # current prediction for x_0
+        if self.model.parameterization != "v":
+            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+        else:
+            pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output)
+
+        if quantize_denoised:
+            pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+
+        if dynamic_threshold is not None:
+            raise NotImplementedError()
+
+        # direction pointing to x_t
+        dir_xt = (1.0 - a_prev - sigma_t**2).sqrt() * e_t
+        noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+        if noise_dropout > 0.0:
+            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+        x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+        return x_prev, pred_x0
+
+    @torch.no_grad()
+    def encode(
+        self,
+        x0,
+        c,
+        t_enc,
+        use_original_steps=False,
+        return_intermediates=None,
+        unconditional_guidance_scale=1.0,
+        unconditional_conditioning=None,
+        callback=None,
+    ):
+        timesteps = (
+            np.arange(self.ddpm_num_timesteps)
+            if use_original_steps
+            else self.ddim_timesteps
+        )
+        num_reference_steps = timesteps.shape[0]
+
+        assert t_enc <= num_reference_steps
+        num_steps = t_enc
+
+        if use_original_steps:
+            alphas_next = self.alphas_cumprod[:num_steps]
+            alphas = self.alphas_cumprod_prev[:num_steps]
+        else:
+            alphas_next = self.ddim_alphas[:num_steps]
+            alphas = torch.tensor(self.ddim_alphas_prev[:num_steps])
+
+        x_next = x0
+        intermediates = []
+        inter_steps = []
+        for i in tqdm(range(num_steps), desc="Encoding Image"):
+            t = torch.full(
+                (x0.shape[0],), timesteps[i], device=self.model.device, dtype=torch.long
+            )
+            if unconditional_guidance_scale == 1.0:
+                noise_pred = self.model.apply_model(x_next, t, c)
+            else:
+                assert unconditional_conditioning is not None
+                e_t_uncond, noise_pred = torch.chunk(
+                    self.model.apply_model(
+                        torch.cat((x_next, x_next)),
+                        torch.cat((t, t)),
+                        torch.cat((unconditional_conditioning, c)),
+                    ),
+                    2,
+                )
+                noise_pred = e_t_uncond + unconditional_guidance_scale * (
+                    noise_pred - e_t_uncond
+                )
+
+            xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next
+            weighted_noise_pred = (
+                alphas_next[i].sqrt()
+                * ((1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt())
+                * noise_pred
+            )
+            x_next = xt_weighted + weighted_noise_pred
+            if (
+                return_intermediates
+                and i % (num_steps // return_intermediates) == 0
+                and i < num_steps - 1
+            ):
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            elif return_intermediates and i >= num_steps - 2:
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            if callback:
+                callback(i)
+
+        out = {"x_encoded": x_next, "intermediate_steps": inter_steps}
+        if return_intermediates:
+            out.update({"intermediates": intermediates})
+        return x_next, out
+
+    @torch.no_grad()
+    def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
+        # fast, but does not allow for exact reconstruction
+        # t serves as an index to gather the correct alphas
+        if use_original_steps:
+            sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
+            sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
+        else:
+            sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
+            sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas
+
+        if noise is None:
+            noise = torch.randn_like(x0)
+        return (
+            extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0
+            + extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise
+        )
+
+    @torch.no_grad()
+    def decode(
+        self,
+        x_latent,
+        cond,
+        t_start,
+        unconditional_guidance_scale=1.0,
+        unconditional_conditioning=None,
+        use_original_steps=False,
+        callback=None,
+    ):
+        timesteps = (
+            np.arange(self.ddpm_num_timesteps)
+            if use_original_steps
+            else self.ddim_timesteps
+        )
+        timesteps = timesteps[:t_start]
+
+        time_range = np.flip(timesteps)
+        total_steps = timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc="Decoding image", total=total_steps)
+        x_dec = x_latent
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full(
+                (x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long
+            )
+            x_dec, _ = self.p_sample_ddim(
+                x_dec,
+                cond,
+                ts,
+                index=index,
+                use_original_steps=use_original_steps,
+                unconditional_guidance_scale=unconditional_guidance_scale,
+                unconditional_conditioning=unconditional_conditioning,
+            )
+            if callback:
+                callback(i)
+        return x_dec

iopaint/model/anytext/cldm/embedding_manager.py

@@ -0,0 +1,165 @@
+'''
+Copyright (c) Alibaba, Inc. and its affiliates.
+'''
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from functools import partial
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import conv_nd, linear
+
+
+def get_clip_token_for_string(tokenizer, string):
+    batch_encoding = tokenizer(string, truncation=True, max_length=77, return_length=True,
+                               return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+    tokens = batch_encoding["input_ids"]
+    assert torch.count_nonzero(tokens - 49407) == 2, f"String '{string}' maps to more than a single token. Please use another string"
+    return tokens[0, 1]
+
+
+def get_bert_token_for_string(tokenizer, string):
+    token = tokenizer(string)
+    assert torch.count_nonzero(token) == 3, f"String '{string}' maps to more than a single token. Please use another string"
+    token = token[0, 1]
+    return token
+
+
+def get_clip_vision_emb(encoder, processor, img):
+    _img = img.repeat(1, 3, 1, 1)*255
+    inputs = processor(images=_img, return_tensors="pt")
+    inputs['pixel_values'] = inputs['pixel_values'].to(img.device)
+    outputs = encoder(**inputs)
+    emb = outputs.image_embeds
+    return emb
+
+
+def get_recog_emb(encoder, img_list):
+    _img_list = [(img.repeat(1, 3, 1, 1)*255)[0] for img in img_list]
+    encoder.predictor.eval()
+    _, preds_neck = encoder.pred_imglist(_img_list, show_debug=False)
+    return preds_neck
+
+
+def pad_H(x):
+    _, _, H, W = x.shape
+    p_top = (W - H) // 2
+    p_bot = W - H - p_top
+    return F.pad(x, (0, 0, p_top, p_bot))
+
+
+class EncodeNet(nn.Module):
+    def __init__(self, in_channels, out_channels):
+        super(EncodeNet, self).__init__()
+        chan = 16
+        n_layer = 4  # downsample
+
+        self.conv1 = conv_nd(2, in_channels, chan, 3, padding=1)
+        self.conv_list = nn.ModuleList([])
+        _c = chan
+        for i in range(n_layer):
+            self.conv_list.append(conv_nd(2, _c, _c*2, 3, padding=1, stride=2))
+            _c *= 2
+        self.conv2 = conv_nd(2, _c, out_channels, 3, padding=1)
+        self.avgpool = nn.AdaptiveAvgPool2d(1)
+        self.act = nn.SiLU()
+
+    def forward(self, x):
+        x = self.act(self.conv1(x))
+        for layer in self.conv_list:
+            x = self.act(layer(x))
+        x = self.act(self.conv2(x))
+        x = self.avgpool(x)
+        x = x.view(x.size(0), -1)
+        return x
+
+
+class EmbeddingManager(nn.Module):
+    def __init__(
+            self,
+            embedder,
+            valid=True,
+            glyph_channels=20,
+            position_channels=1,
+            placeholder_string='*',
+            add_pos=False,
+            emb_type='ocr',
+            **kwargs
+    ):
+        super().__init__()
+        if hasattr(embedder, 'tokenizer'):  # using Stable Diffusion's CLIP encoder
+            get_token_for_string = partial(get_clip_token_for_string, embedder.tokenizer)
+            token_dim = 768
+            if hasattr(embedder, 'vit'):
+                assert emb_type == 'vit'
+                self.get_vision_emb = partial(get_clip_vision_emb, embedder.vit, embedder.processor)
+            self.get_recog_emb = None
+        else:  # using LDM's BERT encoder
+            get_token_for_string = partial(get_bert_token_for_string, embedder.tknz_fn)
+            token_dim = 1280
+        self.token_dim = token_dim
+        self.emb_type = emb_type
+
+        self.add_pos = add_pos
+        if add_pos:
+            self.position_encoder = EncodeNet(position_channels, token_dim)
+        if emb_type == 'ocr':
+            self.proj = linear(40*64, token_dim)
+        if emb_type == 'conv':
+            self.glyph_encoder = EncodeNet(glyph_channels, token_dim)
+
+        self.placeholder_token = get_token_for_string(placeholder_string)
+
+    def encode_text(self, text_info):
+        if self.get_recog_emb is None and self.emb_type == 'ocr':
+            self.get_recog_emb = partial(get_recog_emb, self.recog)
+
+        gline_list = []
+        pos_list = []
+        for i in range(len(text_info['n_lines'])):  # sample index in a batch
+            n_lines = text_info['n_lines'][i]
+            for j in range(n_lines):  # line
+                gline_list += [text_info['gly_line'][j][i:i+1]]
+                if self.add_pos:
+                    pos_list += [text_info['positions'][j][i:i+1]]
+
+        if len(gline_list) > 0:
+            if self.emb_type == 'ocr':
+                recog_emb = self.get_recog_emb(gline_list)
+                enc_glyph = self.proj(recog_emb.reshape(recog_emb.shape[0], -1))
+            elif self.emb_type == 'vit':
+                enc_glyph = self.get_vision_emb(pad_H(torch.cat(gline_list, dim=0)))
+            elif self.emb_type == 'conv':
+                enc_glyph = self.glyph_encoder(pad_H(torch.cat(gline_list, dim=0)))
+            if self.add_pos:
+                enc_pos = self.position_encoder(torch.cat(gline_list, dim=0))
+                enc_glyph = enc_glyph+enc_pos
+
+        self.text_embs_all = []
+        n_idx = 0
+        for i in range(len(text_info['n_lines'])):  # sample index in a batch
+            n_lines = text_info['n_lines'][i]
+            text_embs = []
+            for j in range(n_lines):  # line
+                text_embs += [enc_glyph[n_idx:n_idx+1]]
+                n_idx += 1
+            self.text_embs_all += [text_embs]
+
+    def forward(
+            self,
+            tokenized_text,
+            embedded_text,
+    ):
+        b, device = tokenized_text.shape[0], tokenized_text.device
+        for i in range(b):
+            idx = tokenized_text[i] == self.placeholder_token.to(device)
+            if sum(idx) > 0:
+                if i >= len(self.text_embs_all):
+                    print('truncation for log images...')
+                    break
+                text_emb = torch.cat(self.text_embs_all[i], dim=0)
+                if sum(idx) != len(text_emb):
+                    print('truncation for long caption...')
+                embedded_text[i][idx] = text_emb[:sum(idx)]
+        return embedded_text
+
+    def embedding_parameters(self):
+        return self.parameters()

iopaint/model/anytext/cldm/hack.py

@@ -0,0 +1,111 @@
+import torch
+import einops
+
+import iopaint.model.anytext.ldm.modules.encoders.modules
+import iopaint.model.anytext.ldm.modules.attention
+
+from transformers import logging
+from iopaint.model.anytext.ldm.modules.attention import default
+
+
+def disable_verbosity():
+    logging.set_verbosity_error()
+    print('logging improved.')
+    return
+
+
+def enable_sliced_attention():
+    iopaint.model.anytext.ldm.modules.attention.CrossAttention.forward = _hacked_sliced_attentin_forward
+    print('Enabled sliced_attention.')
+    return
+
+
+def hack_everything(clip_skip=0):
+    disable_verbosity()
+    iopaint.model.anytext.ldm.modules.encoders.modules.FrozenCLIPEmbedder.forward = _hacked_clip_forward
+    iopaint.model.anytext.ldm.modules.encoders.modules.FrozenCLIPEmbedder.clip_skip = clip_skip
+    print('Enabled clip hacks.')
+    return
+
+
+# Written by Lvmin
+def _hacked_clip_forward(self, text):
+    PAD = self.tokenizer.pad_token_id
+    EOS = self.tokenizer.eos_token_id
+    BOS = self.tokenizer.bos_token_id
+
+    def tokenize(t):
+        return self.tokenizer(t, truncation=False, add_special_tokens=False)["input_ids"]
+
+    def transformer_encode(t):
+        if self.clip_skip > 1:
+            rt = self.transformer(input_ids=t, output_hidden_states=True)
+            return self.transformer.text_model.final_layer_norm(rt.hidden_states[-self.clip_skip])
+        else:
+            return self.transformer(input_ids=t, output_hidden_states=False).last_hidden_state
+
+    def split(x):
+        return x[75 * 0: 75 * 1], x[75 * 1: 75 * 2], x[75 * 2: 75 * 3]
+
+    def pad(x, p, i):
+        return x[:i] if len(x) >= i else x + [p] * (i - len(x))
+
+    raw_tokens_list = tokenize(text)
+    tokens_list = []
+
+    for raw_tokens in raw_tokens_list:
+        raw_tokens_123 = split(raw_tokens)
+        raw_tokens_123 = [[BOS] + raw_tokens_i + [EOS] for raw_tokens_i in raw_tokens_123]
+        raw_tokens_123 = [pad(raw_tokens_i, PAD, 77) for raw_tokens_i in raw_tokens_123]
+        tokens_list.append(raw_tokens_123)
+
+    tokens_list = torch.IntTensor(tokens_list).to(self.device)
+
+    feed = einops.rearrange(tokens_list, 'b f i -> (b f) i')
+    y = transformer_encode(feed)
+    z = einops.rearrange(y, '(b f) i c -> b (f i) c', f=3)
+
+    return z
+
+
+# Stolen from https://github.com/basujindal/stable-diffusion/blob/main/optimizedSD/splitAttention.py
+def _hacked_sliced_attentin_forward(self, x, context=None, mask=None):
+    h = self.heads
+
+    q = self.to_q(x)
+    context = default(context, x)
+    k = self.to_k(context)
+    v = self.to_v(context)
+    del context, x
+
+    q, k, v = map(lambda t: einops.rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
+
+    limit = k.shape[0]
+    att_step = 1
+    q_chunks = list(torch.tensor_split(q, limit // att_step, dim=0))
+    k_chunks = list(torch.tensor_split(k, limit // att_step, dim=0))
+    v_chunks = list(torch.tensor_split(v, limit // att_step, dim=0))
+
+    q_chunks.reverse()
+    k_chunks.reverse()
+    v_chunks.reverse()
+    sim = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device)
+    del k, q, v
+    for i in range(0, limit, att_step):
+        q_buffer = q_chunks.pop()
+        k_buffer = k_chunks.pop()
+        v_buffer = v_chunks.pop()
+        sim_buffer = torch.einsum('b i d, b j d -> b i j', q_buffer, k_buffer) * self.scale
+
+        del k_buffer, q_buffer
+        # attention, what we cannot get enough of, by chunks
+
+        sim_buffer = sim_buffer.softmax(dim=-1)
+
+        sim_buffer = torch.einsum('b i j, b j d -> b i d', sim_buffer, v_buffer)
+        del v_buffer
+        sim[i:i + att_step, :, :] = sim_buffer
+
+        del sim_buffer
+    sim = einops.rearrange(sim, '(b h) n d -> b n (h d)', h=h)
+    return self.to_out(sim)

iopaint/model/anytext/cldm/model.py

@@ -0,0 +1,40 @@
+import os
+import torch
+
+from omegaconf import OmegaConf
+from iopaint.model.anytext.ldm.util import instantiate_from_config
+
+
+def get_state_dict(d):
+    return d.get("state_dict", d)
+
+
+def load_state_dict(ckpt_path, location="cpu"):
+    _, extension = os.path.splitext(ckpt_path)
+    if extension.lower() == ".safetensors":
+        import safetensors.torch
+
+        state_dict = safetensors.torch.load_file(ckpt_path, device=location)
+    else:
+        state_dict = get_state_dict(
+            torch.load(ckpt_path, map_location=torch.device(location))
+        )
+    state_dict = get_state_dict(state_dict)
+    print(f"Loaded state_dict from [{ckpt_path}]")
+    return state_dict
+
+
+def create_model(config_path, device, cond_stage_path=None, use_fp16=False):
+    config = OmegaConf.load(config_path)
+    if cond_stage_path:
+        config.model.params.cond_stage_config.params.version = (
+            cond_stage_path  # use pre-downloaded ckpts, in case blocked
+        )
+        config.model.params.cond_stage_config.params.device = device
+    if use_fp16:
+        config.model.params.use_fp16 = True
+        config.model.params.control_stage_config.params.use_fp16 = True
+        config.model.params.unet_config.params.use_fp16 = True
+    model = instantiate_from_config(config.model).cpu()
+    print(f"Loaded model config from [{config_path}]")
+    return model

iopaint/model/anytext/cldm/recognizer.py

@@ -0,0 +1,300 @@
+"""
+Copyright (c) Alibaba, Inc. and its affiliates.
+"""
+import os
+import cv2
+import numpy as np
+import math
+import traceback
+from easydict import EasyDict as edict
+import time
+from iopaint.model.anytext.ocr_recog.RecModel import RecModel
+import torch
+import torch.nn.functional as F
+
+
+def min_bounding_rect(img):
+    ret, thresh = cv2.threshold(img, 127, 255, 0)
+    contours, hierarchy = cv2.findContours(
+        thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
+    )
+    if len(contours) == 0:
+        print("Bad contours, using fake bbox...")
+        return np.array([[0, 0], [100, 0], [100, 100], [0, 100]])
+    max_contour = max(contours, key=cv2.contourArea)
+    rect = cv2.minAreaRect(max_contour)
+    box = cv2.boxPoints(rect)
+    box = np.int0(box)
+    # sort
+    x_sorted = sorted(box, key=lambda x: x[0])
+    left = x_sorted[:2]
+    right = x_sorted[2:]
+    left = sorted(left, key=lambda x: x[1])
+    (tl, bl) = left
+    right = sorted(right, key=lambda x: x[1])
+    (tr, br) = right
+    if tl[1] > bl[1]:
+        (tl, bl) = (bl, tl)
+    if tr[1] > br[1]:
+        (tr, br) = (br, tr)
+    return np.array([tl, tr, br, bl])
+
+
+def create_predictor(model_dir=None, model_lang="ch", is_onnx=False):
+    model_file_path = model_dir
+    if model_file_path is not None and not os.path.exists(model_file_path):
+        raise ValueError("not find model file path {}".format(model_file_path))
+
+    if is_onnx:
+        import onnxruntime as ort
+
+        sess = ort.InferenceSession(
+            model_file_path, providers=["CPUExecutionProvider"]
+        )  # 'TensorrtExecutionProvider', 'CUDAExecutionProvider', 'CPUExecutionProvider'
+        return sess
+    else:
+        if model_lang == "ch":
+            n_class = 6625
+        elif model_lang == "en":
+            n_class = 97
+        else:
+            raise ValueError(f"Unsupported OCR recog model_lang: {model_lang}")
+        rec_config = edict(
+            in_channels=3,
+            backbone=edict(
+                type="MobileNetV1Enhance",
+                scale=0.5,
+                last_conv_stride=[1, 2],
+                last_pool_type="avg",
+            ),
+            neck=edict(
+                type="SequenceEncoder",
+                encoder_type="svtr",
+                dims=64,
+                depth=2,
+                hidden_dims=120,
+                use_guide=True,
+            ),
+            head=edict(
+                type="CTCHead",
+                fc_decay=0.00001,
+                out_channels=n_class,
+                return_feats=True,
+            ),
+        )
+
+        rec_model = RecModel(rec_config)
+        if model_file_path is not None:
+            rec_model.load_state_dict(torch.load(model_file_path, map_location="cpu"))
+            rec_model.eval()
+        return rec_model.eval()
+
+
+def _check_image_file(path):
+    img_end = {"jpg", "bmp", "png", "jpeg", "rgb", "tif", "tiff"}
+    return any([path.lower().endswith(e) for e in img_end])
+
+
+def get_image_file_list(img_file):
+    imgs_lists = []
+    if img_file is None or not os.path.exists(img_file):
+        raise Exception("not found any img file in {}".format(img_file))
+    if os.path.isfile(img_file) and _check_image_file(img_file):
+        imgs_lists.append(img_file)
+    elif os.path.isdir(img_file):
+        for single_file in os.listdir(img_file):
+            file_path = os.path.join(img_file, single_file)
+            if os.path.isfile(file_path) and _check_image_file(file_path):
+                imgs_lists.append(file_path)
+    if len(imgs_lists) == 0:
+        raise Exception("not found any img file in {}".format(img_file))
+    imgs_lists = sorted(imgs_lists)
+    return imgs_lists
+
+
+class TextRecognizer(object):
+    def __init__(self, args, predictor):
+        self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
+        self.rec_batch_num = args.rec_batch_num
+        self.predictor = predictor
+        self.chars = self.get_char_dict(args.rec_char_dict_path)
+        self.char2id = {x: i for i, x in enumerate(self.chars)}
+        self.is_onnx = not isinstance(self.predictor, torch.nn.Module)
+        self.use_fp16 = args.use_fp16
+
+    # img: CHW
+    def resize_norm_img(self, img, max_wh_ratio):
+        imgC, imgH, imgW = self.rec_image_shape
+        assert imgC == img.shape[0]
+        imgW = int((imgH * max_wh_ratio))
+
+        h, w = img.shape[1:]
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = torch.nn.functional.interpolate(
+            img.unsqueeze(0),
+            size=(imgH, resized_w),
+            mode="bilinear",
+            align_corners=True,
+        )
+        resized_image /= 255.0
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = torch.zeros((imgC, imgH, imgW), dtype=torch.float32).to(img.device)
+        padding_im[:, :, 0:resized_w] = resized_image[0]
+        return padding_im
+
+    # img_list: list of tensors with shape chw 0-255
+    def pred_imglist(self, img_list, show_debug=False, is_ori=False):
+        img_num = len(img_list)
+        assert img_num > 0
+        # Calculate the aspect ratio of all text bars
+        width_list = []
+        for img in img_list:
+            width_list.append(img.shape[2] / float(img.shape[1]))
+        # Sorting can speed up the recognition process
+        indices = torch.from_numpy(np.argsort(np.array(width_list)))
+        batch_num = self.rec_batch_num
+        preds_all = [None] * img_num
+        preds_neck_all = [None] * img_num
+        for beg_img_no in range(0, img_num, batch_num):
+            end_img_no = min(img_num, beg_img_no + batch_num)
+            norm_img_batch = []
+
+            imgC, imgH, imgW = self.rec_image_shape[:3]
+            max_wh_ratio = imgW / imgH
+            for ino in range(beg_img_no, end_img_no):
+                h, w = img_list[indices[ino]].shape[1:]
+                if h > w * 1.2:
+                    img = img_list[indices[ino]]
+                    img = torch.transpose(img, 1, 2).flip(dims=[1])
+                    img_list[indices[ino]] = img
+                    h, w = img.shape[1:]
+                # wh_ratio = w * 1.0 / h
+                # max_wh_ratio = max(max_wh_ratio, wh_ratio)  # comment to not use different ratio
+            for ino in range(beg_img_no, end_img_no):
+                norm_img = self.resize_norm_img(img_list[indices[ino]], max_wh_ratio)
+                if self.use_fp16:
+                    norm_img = norm_img.half()
+                norm_img = norm_img.unsqueeze(0)
+                norm_img_batch.append(norm_img)
+            norm_img_batch = torch.cat(norm_img_batch, dim=0)
+            if show_debug:
+                for i in range(len(norm_img_batch)):
+                    _img = norm_img_batch[i].permute(1, 2, 0).detach().cpu().numpy()
+                    _img = (_img + 0.5) * 255
+                    _img = _img[:, :, ::-1]
+                    file_name = f"{indices[beg_img_no + i]}"
+                    file_name = file_name + "_ori" if is_ori else file_name
+                    cv2.imwrite(file_name + ".jpg", _img)
+            if self.is_onnx:
+                input_dict = {}
+                input_dict[self.predictor.get_inputs()[0].name] = (
+                    norm_img_batch.detach().cpu().numpy()
+                )
+                outputs = self.predictor.run(None, input_dict)
+                preds = {}
+                preds["ctc"] = torch.from_numpy(outputs[0])
+                preds["ctc_neck"] = [torch.zeros(1)] * img_num
+            else:
+                preds = self.predictor(norm_img_batch)
+            for rno in range(preds["ctc"].shape[0]):
+                preds_all[indices[beg_img_no + rno]] = preds["ctc"][rno]
+                preds_neck_all[indices[beg_img_no + rno]] = preds["ctc_neck"][rno]
+
+        return torch.stack(preds_all, dim=0), torch.stack(preds_neck_all, dim=0)
+
+    def get_char_dict(self, character_dict_path):
+        character_str = []
+        with open(character_dict_path, "rb") as fin:
+            lines = fin.readlines()
+            for line in lines:
+                line = line.decode("utf-8").strip("\n").strip("\r\n")
+                character_str.append(line)
+        dict_character = list(character_str)
+        dict_character = ["sos"] + dict_character + [" "]  # eos is space
+        return dict_character
+
+    def get_text(self, order):
+        char_list = [self.chars[text_id] for text_id in order]
+        return "".join(char_list)
+
+    def decode(self, mat):
+        text_index = mat.detach().cpu().numpy().argmax(axis=1)
+        ignored_tokens = [0]
+        selection = np.ones(len(text_index), dtype=bool)
+        selection[1:] = text_index[1:] != text_index[:-1]
+        for ignored_token in ignored_tokens:
+            selection &= text_index != ignored_token
+        return text_index[selection], np.where(selection)[0]
+
+    def get_ctcloss(self, preds, gt_text, weight):
+        if not isinstance(weight, torch.Tensor):
+            weight = torch.tensor(weight).to(preds.device)
+        ctc_loss = torch.nn.CTCLoss(reduction="none")
+        log_probs = preds.log_softmax(dim=2).permute(1, 0, 2)  # NTC-->TNC
+        targets = []
+        target_lengths = []
+        for t in gt_text:
+            targets += [self.char2id.get(i, len(self.chars) - 1) for i in t]
+            target_lengths += [len(t)]
+        targets = torch.tensor(targets).to(preds.device)
+        target_lengths = torch.tensor(target_lengths).to(preds.device)
+        input_lengths = torch.tensor([log_probs.shape[0]] * (log_probs.shape[1])).to(
+            preds.device
+        )
+        loss = ctc_loss(log_probs, targets, input_lengths, target_lengths)
+        loss = loss / input_lengths * weight
+        return loss
+
+
+def main():
+    rec_model_dir = "./ocr_weights/ppv3_rec.pth"
+    predictor = create_predictor(rec_model_dir)
+    args = edict()
+    args.rec_image_shape = "3, 48, 320"
+    args.rec_char_dict_path = "./ocr_weights/ppocr_keys_v1.txt"
+    args.rec_batch_num = 6
+    text_recognizer = TextRecognizer(args, predictor)
+    image_dir = "./test_imgs_cn"
+    gt_text = ["韩国小馆"] * 14
+
+    image_file_list = get_image_file_list(image_dir)
+    valid_image_file_list = []
+    img_list = []
+
+    for image_file in image_file_list:
+        img = cv2.imread(image_file)
+        if img is None:
+            print("error in loading image:{}".format(image_file))
+            continue
+        valid_image_file_list.append(image_file)
+        img_list.append(torch.from_numpy(img).permute(2, 0, 1).float())
+    try:
+        tic = time.time()
+        times = []
+        for i in range(10):
+            preds, _ = text_recognizer.pred_imglist(img_list)  # get text
+            preds_all = preds.softmax(dim=2)
+            times += [(time.time() - tic) * 1000.0]
+            tic = time.time()
+        print(times)
+        print(np.mean(times[1:]) / len(preds_all))
+        weight = np.ones(len(gt_text))
+        loss = text_recognizer.get_ctcloss(preds, gt_text, weight)
+        for i in range(len(valid_image_file_list)):
+            pred = preds_all[i]
+            order, idx = text_recognizer.decode(pred)
+            text = text_recognizer.get_text(order)
+            print(
+                f'{valid_image_file_list[i]}: pred/gt="{text}"/"{gt_text[i]}", loss={loss[i]:.2f}'
+            )
+    except Exception as E:
+        print(traceback.format_exc(), E)
+
+
+if __name__ == "__main__":
+    main()

iopaint/model/anytext/ldm/models/autoencoder.py

@@ -0,0 +1,218 @@
+import torch
+import torch.nn.functional as F
+from contextlib import contextmanager
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.model import Encoder, Decoder
+from iopaint.model.anytext.ldm.modules.distributions.distributions import DiagonalGaussianDistribution
+
+from iopaint.model.anytext.ldm.util import instantiate_from_config
+from iopaint.model.anytext.ldm.modules.ema import LitEma
+
+
+class AutoencoderKL(torch.nn.Module):
+    def __init__(self,
+                 ddconfig,
+                 lossconfig,
+                 embed_dim,
+                 ckpt_path=None,
+                 ignore_keys=[],
+                 image_key="image",
+                 colorize_nlabels=None,
+                 monitor=None,
+                 ema_decay=None,
+                 learn_logvar=False
+                 ):
+        super().__init__()
+        self.learn_logvar = learn_logvar
+        self.image_key = image_key
+        self.encoder = Encoder(**ddconfig)
+        self.decoder = Decoder(**ddconfig)
+        self.loss = instantiate_from_config(lossconfig)
+        assert ddconfig["double_z"]
+        self.quant_conv = torch.nn.Conv2d(2*ddconfig["z_channels"], 2*embed_dim, 1)
+        self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1)
+        self.embed_dim = embed_dim
+        if colorize_nlabels is not None:
+            assert type(colorize_nlabels)==int
+            self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1))
+        if monitor is not None:
+            self.monitor = monitor
+
+        self.use_ema = ema_decay is not None
+        if self.use_ema:
+            self.ema_decay = ema_decay
+            assert 0. < ema_decay < 1.
+            self.model_ema = LitEma(self, decay=ema_decay)
+            print(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.")
+
+        if ckpt_path is not None:
+            self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys)
+
+    def init_from_ckpt(self, path, ignore_keys=list()):
+        sd = torch.load(path, map_location="cpu")["state_dict"]
+        keys = list(sd.keys())
+        for k in keys:
+            for ik in ignore_keys:
+                if k.startswith(ik):
+                    print("Deleting key {} from state_dict.".format(k))
+                    del sd[k]
+        self.load_state_dict(sd, strict=False)
+        print(f"Restored from {path}")
+
+    @contextmanager
+    def ema_scope(self, context=None):
+        if self.use_ema:
+            self.model_ema.store(self.parameters())
+            self.model_ema.copy_to(self)
+            if context is not None:
+                print(f"{context}: Switched to EMA weights")
+        try:
+            yield None
+        finally:
+            if self.use_ema:
+                self.model_ema.restore(self.parameters())
+                if context is not None:
+                    print(f"{context}: Restored training weights")
+
+    def on_train_batch_end(self, *args, **kwargs):
+        if self.use_ema:
+            self.model_ema(self)
+
+    def encode(self, x):
+        h = self.encoder(x)
+        moments = self.quant_conv(h)
+        posterior = DiagonalGaussianDistribution(moments)
+        return posterior
+
+    def decode(self, z):
+        z = self.post_quant_conv(z)
+        dec = self.decoder(z)
+        return dec
+
+    def forward(self, input, sample_posterior=True):
+        posterior = self.encode(input)
+        if sample_posterior:
+            z = posterior.sample()
+        else:
+            z = posterior.mode()
+        dec = self.decode(z)
+        return dec, posterior
+
+    def get_input(self, batch, k):
+        x = batch[k]
+        if len(x.shape) == 3:
+            x = x[..., None]
+        x = x.permute(0, 3, 1, 2).to(memory_format=torch.contiguous_format).float()
+        return x
+
+    def training_step(self, batch, batch_idx, optimizer_idx):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+
+        if optimizer_idx == 0:
+            # train encoder+decoder+logvar
+            aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                            last_layer=self.get_last_layer(), split="train")
+            self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return aeloss
+
+        if optimizer_idx == 1:
+            # train the discriminator
+            discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step,
+                                                last_layer=self.get_last_layer(), split="train")
+
+            self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True)
+            self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False)
+            return discloss
+
+    def validation_step(self, batch, batch_idx):
+        log_dict = self._validation_step(batch, batch_idx)
+        with self.ema_scope():
+            log_dict_ema = self._validation_step(batch, batch_idx, postfix="_ema")
+        return log_dict
+
+    def _validation_step(self, batch, batch_idx, postfix=""):
+        inputs = self.get_input(batch, self.image_key)
+        reconstructions, posterior = self(inputs)
+        aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step,
+                                        last_layer=self.get_last_layer(), split="val"+postfix)
+
+        discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step,
+                                            last_layer=self.get_last_layer(), split="val"+postfix)
+
+        self.log(f"val{postfix}/rec_loss", log_dict_ae[f"val{postfix}/rec_loss"])
+        self.log_dict(log_dict_ae)
+        self.log_dict(log_dict_disc)
+        return self.log_dict
+
+    def configure_optimizers(self):
+        lr = self.learning_rate
+        ae_params_list = list(self.encoder.parameters()) + list(self.decoder.parameters()) + list(
+            self.quant_conv.parameters()) + list(self.post_quant_conv.parameters())
+        if self.learn_logvar:
+            print(f"{self.__class__.__name__}: Learning logvar")
+            ae_params_list.append(self.loss.logvar)
+        opt_ae = torch.optim.Adam(ae_params_list,
+                                  lr=lr, betas=(0.5, 0.9))
+        opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(),
+                                    lr=lr, betas=(0.5, 0.9))
+        return [opt_ae, opt_disc], []
+
+    def get_last_layer(self):
+        return self.decoder.conv_out.weight
+
+    @torch.no_grad()
+    def log_images(self, batch, only_inputs=False, log_ema=False, **kwargs):
+        log = dict()
+        x = self.get_input(batch, self.image_key)
+        x = x.to(self.device)
+        if not only_inputs:
+            xrec, posterior = self(x)
+            if x.shape[1] > 3:
+                # colorize with random projection
+                assert xrec.shape[1] > 3
+                x = self.to_rgb(x)
+                xrec = self.to_rgb(xrec)
+            log["samples"] = self.decode(torch.randn_like(posterior.sample()))
+            log["reconstructions"] = xrec
+            if log_ema or self.use_ema:
+                with self.ema_scope():
+                    xrec_ema, posterior_ema = self(x)
+                    if x.shape[1] > 3:
+                        # colorize with random projection
+                        assert xrec_ema.shape[1] > 3
+                        xrec_ema = self.to_rgb(xrec_ema)
+                    log["samples_ema"] = self.decode(torch.randn_like(posterior_ema.sample()))
+                    log["reconstructions_ema"] = xrec_ema
+        log["inputs"] = x
+        return log
+
+    def to_rgb(self, x):
+        assert self.image_key == "segmentation"
+        if not hasattr(self, "colorize"):
+            self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x))
+        x = F.conv2d(x, weight=self.colorize)
+        x = 2.*(x-x.min())/(x.max()-x.min()) - 1.
+        return x
+
+
+class IdentityFirstStage(torch.nn.Module):
+    def __init__(self, *args, vq_interface=False, **kwargs):
+        self.vq_interface = vq_interface
+        super().__init__()
+
+    def encode(self, x, *args, **kwargs):
+        return x
+
+    def decode(self, x, *args, **kwargs):
+        return x
+
+    def quantize(self, x, *args, **kwargs):
+        if self.vq_interface:
+            return x, None, [None, None, None]
+        return x
+
+    def forward(self, x, *args, **kwargs):
+        return x
+

iopaint/model/anytext/ldm/models/diffusion/ddim.py

@@ -0,0 +1,354 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, extract_into_tensor
+
+
+class DDIMSampler(object):
+    def __init__(self, model, schedule="linear", **kwargs):
+        super().__init__()
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
+        self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
+                                                  num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
+        alphas_cumprod = self.model.alphas_cumprod
+        assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
+
+        self.register_buffer('betas', to_torch(self.model.betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
+                                                                                   ddim_timesteps=self.ddim_timesteps,
+                                                                                   eta=ddim_eta,verbose=verbose)
+        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
+                        1 - self.alphas_cumprod / self.alphas_cumprod_prev))
+        self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               dynamic_threshold=None,
+               ucg_schedule=None,
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                ctmp = conditioning[list(conditioning.keys())[0]]
+                while isinstance(ctmp, list): ctmp = ctmp[0]
+                cbs = ctmp.shape[0]
+                # cbs = len(ctmp[0])
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+
+            elif isinstance(conditioning, list):
+                for ctmp in conditioning:
+                    if ctmp.shape[0] != batch_size:
+                        print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f'Data shape for DDIM sampling is {size}, eta {eta}')
+
+        samples, intermediates = self.ddim_sampling(conditioning, size,
+                                                    callback=callback,
+                                                    img_callback=img_callback,
+                                                    quantize_denoised=quantize_x0,
+                                                    mask=mask, x0=x0,
+                                                    ddim_use_original_steps=False,
+                                                    noise_dropout=noise_dropout,
+                                                    temperature=temperature,
+                                                    score_corrector=score_corrector,
+                                                    corrector_kwargs=corrector_kwargs,
+                                                    x_T=x_T,
+                                                    log_every_t=log_every_t,
+                                                    unconditional_guidance_scale=unconditional_guidance_scale,
+                                                    unconditional_conditioning=unconditional_conditioning,
+                                                    dynamic_threshold=dynamic_threshold,
+                                                    ucg_schedule=ucg_schedule
+                                                    )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def ddim_sampling(self, cond, shape,
+                      x_T=None, ddim_use_original_steps=False,
+                      callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, log_every_t=100,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None, dynamic_threshold=None,
+                      ucg_schedule=None):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {'x_inter': [img], 'pred_x0': [img], "index": [10000]}
+        time_range = reversed(range(0, timesteps)) if ddim_use_original_steps else np.flip(timesteps)
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(x0, ts)  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1. - mask) * img
+
+            if ucg_schedule is not None:
+                assert len(ucg_schedule) == len(time_range)
+                unconditional_guidance_scale = ucg_schedule[i]
+
+            outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
+                                      quantize_denoised=quantize_denoised, temperature=temperature,
+                                      noise_dropout=noise_dropout, score_corrector=score_corrector,
+                                      corrector_kwargs=corrector_kwargs,
+                                      unconditional_guidance_scale=unconditional_guidance_scale,
+                                      unconditional_conditioning=unconditional_conditioning,
+                                      dynamic_threshold=dynamic_threshold)
+            img, pred_x0 = outs
+            if callback:
+                callback(i)
+            if img_callback:
+                img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates['x_inter'].append(img)
+                intermediates['pred_x0'].append(pred_x0)
+                intermediates['index'].append(index)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None,
+                      dynamic_threshold=None):
+        b, *_, device = *x.shape, x.device
+
+        if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
+            model_output = self.model.apply_model(x, t, c)
+        else:
+            x_in = torch.cat([x] * 2)
+            t_in = torch.cat([t] * 2)
+            if isinstance(c, dict):
+                assert isinstance(unconditional_conditioning, dict)
+                c_in = dict()
+                for k in c:
+                    if isinstance(c[k], list):
+                        c_in[k] = [torch.cat([
+                            unconditional_conditioning[k][i],
+                            c[k][i]]) for i in range(len(c[k]))]
+                    elif isinstance(c[k], dict):
+                        c_in[k] = dict()
+                        for key in c[k]:
+                            if isinstance(c[k][key], list):
+                                if not isinstance(c[k][key][0], torch.Tensor):
+                                    continue
+                                c_in[k][key] = [torch.cat([
+                                    unconditional_conditioning[k][key][i],
+                                    c[k][key][i]]) for i in range(len(c[k][key]))]
+                            else:
+                                c_in[k][key] = torch.cat([
+                                    unconditional_conditioning[k][key],
+                                    c[k][key]])
+
+                    else:
+                        c_in[k] = torch.cat([
+                                unconditional_conditioning[k],
+                                c[k]])
+            elif isinstance(c, list):
+                c_in = list()
+                assert isinstance(unconditional_conditioning, list)
+                for i in range(len(c)):
+                    c_in.append(torch.cat([unconditional_conditioning[i], c[i]]))
+            else:
+                c_in = torch.cat([unconditional_conditioning, c])
+            model_uncond, model_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
+            model_output = model_uncond + unconditional_guidance_scale * (model_t - model_uncond)
+
+        if self.model.parameterization == "v":
+            e_t = self.model.predict_eps_from_z_and_v(x, t, model_output)
+        else:
+            e_t = model_output
+
+        if score_corrector is not None:
+            assert self.model.parameterization == "eps", 'not implemented'
+            e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
+        sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
+        sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
+        # select parameters corresponding to the currently considered timestep
+        a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+        a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+        sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+        sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
+
+        # current prediction for x_0
+        if self.model.parameterization != "v":
+            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+        else:
+            pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output)
+
+        if quantize_denoised:
+            pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+
+        if dynamic_threshold is not None:
+            raise NotImplementedError()
+
+        # direction pointing to x_t
+        dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
+        noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+        if noise_dropout > 0.:
+            noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+        x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+        return x_prev, pred_x0
+
+    @torch.no_grad()
+    def encode(self, x0, c, t_enc, use_original_steps=False, return_intermediates=None,
+               unconditional_guidance_scale=1.0, unconditional_conditioning=None, callback=None):
+        num_reference_steps = self.ddpm_num_timesteps if use_original_steps else self.ddim_timesteps.shape[0]
+
+        assert t_enc <= num_reference_steps
+        num_steps = t_enc
+
+        if use_original_steps:
+            alphas_next = self.alphas_cumprod[:num_steps]
+            alphas = self.alphas_cumprod_prev[:num_steps]
+        else:
+            alphas_next = self.ddim_alphas[:num_steps]
+            alphas = torch.tensor(self.ddim_alphas_prev[:num_steps])
+
+        x_next = x0
+        intermediates = []
+        inter_steps = []
+        for i in tqdm(range(num_steps), desc='Encoding Image'):
+            t = torch.full((x0.shape[0],), i, device=self.model.device, dtype=torch.long)
+            if unconditional_guidance_scale == 1.:
+                noise_pred = self.model.apply_model(x_next, t, c)
+            else:
+                assert unconditional_conditioning is not None
+                e_t_uncond, noise_pred = torch.chunk(
+                    self.model.apply_model(torch.cat((x_next, x_next)), torch.cat((t, t)),
+                                           torch.cat((unconditional_conditioning, c))), 2)
+                noise_pred = e_t_uncond + unconditional_guidance_scale * (noise_pred - e_t_uncond)
+
+            xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next
+            weighted_noise_pred = alphas_next[i].sqrt() * (
+                    (1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt()) * noise_pred
+            x_next = xt_weighted + weighted_noise_pred
+            if return_intermediates and i % (
+                    num_steps // return_intermediates) == 0 and i < num_steps - 1:
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            elif return_intermediates and i >= num_steps - 2:
+                intermediates.append(x_next)
+                inter_steps.append(i)
+            if callback: callback(i)
+
+        out = {'x_encoded': x_next, 'intermediate_steps': inter_steps}
+        if return_intermediates:
+            out.update({'intermediates': intermediates})
+        return x_next, out
+
+    @torch.no_grad()
+    def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
+        # fast, but does not allow for exact reconstruction
+        # t serves as an index to gather the correct alphas
+        if use_original_steps:
+            sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
+            sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
+        else:
+            sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
+            sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas
+
+        if noise is None:
+            noise = torch.randn_like(x0)
+        return (extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0 +
+                extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise)
+
+    @torch.no_grad()
+    def decode(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None,
+               use_original_steps=False, callback=None):
+
+        timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
+        timesteps = timesteps[:t_start]
+
+        time_range = np.flip(timesteps)
+        total_steps = timesteps.shape[0]
+        print(f"Running DDIM Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='Decoding image', total=total_steps)
+        x_dec = x_latent
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long)
+            x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps,
+                                          unconditional_guidance_scale=unconditional_guidance_scale,
+                                          unconditional_conditioning=unconditional_conditioning)
+            if callback: callback(i)
+        return x_dec

iopaint/model/anytext/ldm/models/diffusion/dpm_solver/__init__.py

@@ -0,0 +1 @@
+from .sampler import DPMSolverSampler

iopaint/model/anytext/ldm/models/diffusion/dpm_solver/sampler.py

@@ -0,0 +1,87 @@
+"""SAMPLING ONLY."""
+import torch
+
+from .dpm_solver import NoiseScheduleVP, model_wrapper, DPM_Solver
+
+
+MODEL_TYPES = {
+    "eps": "noise",
+    "v": "v"
+}
+
+
+class DPMSolverSampler(object):
+    def __init__(self, model, **kwargs):
+        super().__init__()
+        self.model = model
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(model.device)
+        self.register_buffer('alphas_cumprod', to_torch(model.alphas_cumprod))
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None,
+               # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                cbs = conditioning[list(conditioning.keys())[0]].shape[0]
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+
+        print(f'Data shape for DPM-Solver sampling is {size}, sampling steps {S}')
+
+        device = self.model.betas.device
+        if x_T is None:
+            img = torch.randn(size, device=device)
+        else:
+            img = x_T
+
+        ns = NoiseScheduleVP('discrete', alphas_cumprod=self.alphas_cumprod)
+
+        model_fn = model_wrapper(
+            lambda x, t, c: self.model.apply_model(x, t, c),
+            ns,
+            model_type=MODEL_TYPES[self.model.parameterization],
+            guidance_type="classifier-free",
+            condition=conditioning,
+            unconditional_condition=unconditional_conditioning,
+            guidance_scale=unconditional_guidance_scale,
+        )
+
+        dpm_solver = DPM_Solver(model_fn, ns, predict_x0=True, thresholding=False)
+        x = dpm_solver.sample(img, steps=S, skip_type="time_uniform", method="multistep", order=2, lower_order_final=True)
+
+        return x.to(device), None

iopaint/model/anytext/ldm/models/diffusion/plms.py

@@ -0,0 +1,244 @@
+"""SAMPLING ONLY."""
+
+import torch
+import numpy as np
+from tqdm import tqdm
+from functools import partial
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like
+from iopaint.model.anytext.ldm.models.diffusion.sampling_util import norm_thresholding
+
+
+class PLMSSampler(object):
+    def __init__(self, model, schedule="linear", **kwargs):
+        super().__init__()
+        self.model = model
+        self.ddpm_num_timesteps = model.num_timesteps
+        self.schedule = schedule
+
+    def register_buffer(self, name, attr):
+        if type(attr) == torch.Tensor:
+            if attr.device != torch.device("cuda"):
+                attr = attr.to(torch.device("cuda"))
+        setattr(self, name, attr)
+
+    def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
+        if ddim_eta != 0:
+            raise ValueError('ddim_eta must be 0 for PLMS')
+        self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
+                                                  num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
+        alphas_cumprod = self.model.alphas_cumprod
+        assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
+        to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
+
+        self.register_buffer('betas', to_torch(self.model.betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
+
+        # ddim sampling parameters
+        ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
+                                                                                   ddim_timesteps=self.ddim_timesteps,
+                                                                                   eta=ddim_eta,verbose=verbose)
+        self.register_buffer('ddim_sigmas', ddim_sigmas)
+        self.register_buffer('ddim_alphas', ddim_alphas)
+        self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
+        self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
+        sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
+            (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
+                        1 - self.alphas_cumprod / self.alphas_cumprod_prev))
+        self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
+
+    @torch.no_grad()
+    def sample(self,
+               S,
+               batch_size,
+               shape,
+               conditioning=None,
+               callback=None,
+               normals_sequence=None,
+               img_callback=None,
+               quantize_x0=False,
+               eta=0.,
+               mask=None,
+               x0=None,
+               temperature=1.,
+               noise_dropout=0.,
+               score_corrector=None,
+               corrector_kwargs=None,
+               verbose=True,
+               x_T=None,
+               log_every_t=100,
+               unconditional_guidance_scale=1.,
+               unconditional_conditioning=None,
+               # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
+               dynamic_threshold=None,
+               **kwargs
+               ):
+        if conditioning is not None:
+            if isinstance(conditioning, dict):
+                cbs = conditioning[list(conditioning.keys())[0]].shape[0]
+                if cbs != batch_size:
+                    print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
+            else:
+                if conditioning.shape[0] != batch_size:
+                    print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
+
+        self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
+        # sampling
+        C, H, W = shape
+        size = (batch_size, C, H, W)
+        print(f'Data shape for PLMS sampling is {size}')
+
+        samples, intermediates = self.plms_sampling(conditioning, size,
+                                                    callback=callback,
+                                                    img_callback=img_callback,
+                                                    quantize_denoised=quantize_x0,
+                                                    mask=mask, x0=x0,
+                                                    ddim_use_original_steps=False,
+                                                    noise_dropout=noise_dropout,
+                                                    temperature=temperature,
+                                                    score_corrector=score_corrector,
+                                                    corrector_kwargs=corrector_kwargs,
+                                                    x_T=x_T,
+                                                    log_every_t=log_every_t,
+                                                    unconditional_guidance_scale=unconditional_guidance_scale,
+                                                    unconditional_conditioning=unconditional_conditioning,
+                                                    dynamic_threshold=dynamic_threshold,
+                                                    )
+        return samples, intermediates
+
+    @torch.no_grad()
+    def plms_sampling(self, cond, shape,
+                      x_T=None, ddim_use_original_steps=False,
+                      callback=None, timesteps=None, quantize_denoised=False,
+                      mask=None, x0=None, img_callback=None, log_every_t=100,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None,
+                      dynamic_threshold=None):
+        device = self.model.betas.device
+        b = shape[0]
+        if x_T is None:
+            img = torch.randn(shape, device=device)
+        else:
+            img = x_T
+
+        if timesteps is None:
+            timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
+        elif timesteps is not None and not ddim_use_original_steps:
+            subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
+            timesteps = self.ddim_timesteps[:subset_end]
+
+        intermediates = {'x_inter': [img], 'pred_x0': [img]}
+        time_range = list(reversed(range(0,timesteps))) if ddim_use_original_steps else np.flip(timesteps)
+        total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
+        print(f"Running PLMS Sampling with {total_steps} timesteps")
+
+        iterator = tqdm(time_range, desc='PLMS Sampler', total=total_steps)
+        old_eps = []
+
+        for i, step in enumerate(iterator):
+            index = total_steps - i - 1
+            ts = torch.full((b,), step, device=device, dtype=torch.long)
+            ts_next = torch.full((b,), time_range[min(i + 1, len(time_range) - 1)], device=device, dtype=torch.long)
+
+            if mask is not None:
+                assert x0 is not None
+                img_orig = self.model.q_sample(x0, ts)  # TODO: deterministic forward pass?
+                img = img_orig * mask + (1. - mask) * img
+
+            outs = self.p_sample_plms(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
+                                      quantize_denoised=quantize_denoised, temperature=temperature,
+                                      noise_dropout=noise_dropout, score_corrector=score_corrector,
+                                      corrector_kwargs=corrector_kwargs,
+                                      unconditional_guidance_scale=unconditional_guidance_scale,
+                                      unconditional_conditioning=unconditional_conditioning,
+                                      old_eps=old_eps, t_next=ts_next,
+                                      dynamic_threshold=dynamic_threshold)
+            img, pred_x0, e_t = outs
+            old_eps.append(e_t)
+            if len(old_eps) >= 4:
+                old_eps.pop(0)
+            if callback: callback(i)
+            if img_callback: img_callback(pred_x0, i)
+
+            if index % log_every_t == 0 or index == total_steps - 1:
+                intermediates['x_inter'].append(img)
+                intermediates['pred_x0'].append(pred_x0)
+
+        return img, intermediates
+
+    @torch.no_grad()
+    def p_sample_plms(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
+                      temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
+                      unconditional_guidance_scale=1., unconditional_conditioning=None, old_eps=None, t_next=None,
+                      dynamic_threshold=None):
+        b, *_, device = *x.shape, x.device
+
+        def get_model_output(x, t):
+            if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
+                e_t = self.model.apply_model(x, t, c)
+            else:
+                x_in = torch.cat([x] * 2)
+                t_in = torch.cat([t] * 2)
+                c_in = torch.cat([unconditional_conditioning, c])
+                e_t_uncond, e_t = self.model.apply_model(x_in, t_in, c_in).chunk(2)
+                e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond)
+
+            if score_corrector is not None:
+                assert self.model.parameterization == "eps"
+                e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
+
+            return e_t
+
+        alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
+        alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
+        sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
+        sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
+
+        def get_x_prev_and_pred_x0(e_t, index):
+            # select parameters corresponding to the currently considered timestep
+            a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
+            a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
+            sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
+            sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
+
+            # current prediction for x_0
+            pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
+            if quantize_denoised:
+                pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
+            if dynamic_threshold is not None:
+                pred_x0 = norm_thresholding(pred_x0, dynamic_threshold)
+            # direction pointing to x_t
+            dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
+            noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
+            if noise_dropout > 0.:
+                noise = torch.nn.functional.dropout(noise, p=noise_dropout)
+            x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
+            return x_prev, pred_x0
+
+        e_t = get_model_output(x, t)
+        if len(old_eps) == 0:
+            # Pseudo Improved Euler (2nd order)
+            x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t, index)
+            e_t_next = get_model_output(x_prev, t_next)
+            e_t_prime = (e_t + e_t_next) / 2
+        elif len(old_eps) == 1:
+            # 2nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (3 * e_t - old_eps[-1]) / 2
+        elif len(old_eps) == 2:
+            # 3nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (23 * e_t - 16 * old_eps[-1] + 5 * old_eps[-2]) / 12
+        elif len(old_eps) >= 3:
+            # 4nd order Pseudo Linear Multistep (Adams-Bashforth)
+            e_t_prime = (55 * e_t - 59 * old_eps[-1] + 37 * old_eps[-2] - 9 * old_eps[-3]) / 24
+
+        x_prev, pred_x0 = get_x_prev_and_pred_x0(e_t_prime, index)
+
+        return x_prev, pred_x0, e_t

iopaint/model/anytext/ldm/models/diffusion/sampling_util.py

@@ -0,0 +1,22 @@
+import torch
+import numpy as np
+
+
+def append_dims(x, target_dims):
+    """Appends dimensions to the end of a tensor until it has target_dims dimensions.
+    From https://github.com/crowsonkb/k-diffusion/blob/master/k_diffusion/utils.py"""
+    dims_to_append = target_dims - x.ndim
+    if dims_to_append < 0:
+        raise ValueError(f'input has {x.ndim} dims but target_dims is {target_dims}, which is less')
+    return x[(...,) + (None,) * dims_to_append]
+
+
+def norm_thresholding(x0, value):
+    s = append_dims(x0.pow(2).flatten(1).mean(1).sqrt().clamp(min=value), x0.ndim)
+    return x0 * (value / s)
+
+
+def spatial_norm_thresholding(x0, value):
+    # b c h w
+    s = x0.pow(2).mean(1, keepdim=True).sqrt().clamp(min=value)
+    return x0 * (value / s)

iopaint/model/anytext/ldm/modules/attention.py

@@ -0,0 +1,360 @@
+from inspect import isfunction
+import math
+import torch
+import torch.nn.functional as F
+from torch import nn, einsum
+from einops import rearrange, repeat
+from typing import Optional, Any
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import checkpoint
+
+
+# CrossAttn precision handling
+import os
+
+_ATTN_PRECISION = os.environ.get("ATTN_PRECISION", "fp32")
+
+
+def exists(val):
+    return val is not None
+
+
+def uniq(arr):
+    return {el: True for el in arr}.keys()
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def max_neg_value(t):
+    return -torch.finfo(t.dtype).max
+
+
+def init_(tensor):
+    dim = tensor.shape[-1]
+    std = 1 / math.sqrt(dim)
+    tensor.uniform_(-std, std)
+    return tensor
+
+
+# feedforward
+class GEGLU(nn.Module):
+    def __init__(self, dim_in, dim_out):
+        super().__init__()
+        self.proj = nn.Linear(dim_in, dim_out * 2)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, dim=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Module):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.0):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = (
+            nn.Sequential(nn.Linear(dim, inner_dim), nn.GELU())
+            if not glu
+            else GEGLU(dim, inner_dim)
+        )
+
+        self.net = nn.Sequential(
+            project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+def zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+
+def Normalize(in_channels):
+    return torch.nn.GroupNorm(
+        num_groups=32, num_channels=in_channels, eps=1e-6, affine=True
+    )
+
+
+class SpatialSelfAttention(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.k = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.v = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.proj_out = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        b, c, h, w = q.shape
+        q = rearrange(q, "b c h w -> b (h w) c")
+        k = rearrange(k, "b c h w -> b c (h w)")
+        w_ = torch.einsum("bij,bjk->bik", q, k)
+
+        w_ = w_ * (int(c) ** (-0.5))
+        w_ = torch.nn.functional.softmax(w_, dim=2)
+
+        # attend to values
+        v = rearrange(v, "b c h w -> b c (h w)")
+        w_ = rearrange(w_, "b i j -> b j i")
+        h_ = torch.einsum("bij,bjk->bik", v, w_)
+        h_ = rearrange(h_, "b c (h w) -> b c h w", h=h)
+        h_ = self.proj_out(h_)
+
+        return x + h_
+
+
+class CrossAttention(nn.Module):
+    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.0):
+        super().__init__()
+        inner_dim = dim_head * heads
+        context_dim = default(context_dim, query_dim)
+
+        self.scale = dim_head**-0.5
+        self.heads = heads
+
+        self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
+        self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
+        self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
+
+        self.to_out = nn.Sequential(
+            nn.Linear(inner_dim, query_dim), nn.Dropout(dropout)
+        )
+
+    def forward(self, x, context=None, mask=None):
+        h = self.heads
+
+        q = self.to_q(x)
+        context = default(context, x)
+        k = self.to_k(context)
+        v = self.to_v(context)
+
+        q, k, v = map(lambda t: rearrange(t, "b n (h d) -> (b h) n d", h=h), (q, k, v))
+
+        # force cast to fp32 to avoid overflowing
+        if _ATTN_PRECISION == "fp32":
+            with torch.autocast(enabled=False, device_type="cuda"):
+                q, k = q.float(), k.float()
+                sim = einsum("b i d, b j d -> b i j", q, k) * self.scale
+        else:
+            sim = einsum("b i d, b j d -> b i j", q, k) * self.scale
+
+        del q, k
+
+        if exists(mask):
+            mask = rearrange(mask, "b ... -> b (...)")
+            max_neg_value = -torch.finfo(sim.dtype).max
+            mask = repeat(mask, "b j -> (b h) () j", h=h)
+            sim.masked_fill_(~mask, max_neg_value)
+
+        # attention, what we cannot get enough of
+        sim = sim.softmax(dim=-1)
+
+        out = einsum("b i j, b j d -> b i d", sim, v)
+        out = rearrange(out, "(b h) n d -> b n (h d)", h=h)
+        return self.to_out(out)
+
+
+class SDPACrossAttention(CrossAttention):
+    def forward(self, x, context=None, mask=None):
+        batch_size, sequence_length, inner_dim = x.shape
+
+        if mask is not None:
+            mask = self.prepare_attention_mask(mask, sequence_length, batch_size)
+            mask = mask.view(batch_size, self.heads, -1, mask.shape[-1])
+
+        h = self.heads
+        q_in = self.to_q(x)
+        context = default(context, x)
+
+        k_in = self.to_k(context)
+        v_in = self.to_v(context)
+
+        head_dim = inner_dim // h
+        q = q_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
+        k = k_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
+        v = v_in.view(batch_size, -1, h, head_dim).transpose(1, 2)
+
+        del q_in, k_in, v_in
+
+        dtype = q.dtype
+        if _ATTN_PRECISION == "fp32":
+            q, k, v = q.float(), k.float(), v.float()
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        hidden_states = torch.nn.functional.scaled_dot_product_attention(
+            q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(
+            batch_size, -1, h * head_dim
+        )
+        hidden_states = hidden_states.to(dtype)
+
+        # linear proj
+        hidden_states = self.to_out[0](hidden_states)
+        # dropout
+        hidden_states = self.to_out[1](hidden_states)
+        return hidden_states
+
+
+class BasicTransformerBlock(nn.Module):
+    def __init__(
+        self,
+        dim,
+        n_heads,
+        d_head,
+        dropout=0.0,
+        context_dim=None,
+        gated_ff=True,
+        checkpoint=True,
+        disable_self_attn=False,
+    ):
+        super().__init__()
+
+        if hasattr(torch.nn.functional, "scaled_dot_product_attention"):
+            attn_cls = SDPACrossAttention
+        else:
+            attn_cls = CrossAttention
+
+        self.disable_self_attn = disable_self_attn
+        self.attn1 = attn_cls(
+            query_dim=dim,
+            heads=n_heads,
+            dim_head=d_head,
+            dropout=dropout,
+            context_dim=context_dim if self.disable_self_attn else None,
+        )  # is a self-attention if not self.disable_self_attn
+        self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
+        self.attn2 = attn_cls(
+            query_dim=dim,
+            context_dim=context_dim,
+            heads=n_heads,
+            dim_head=d_head,
+            dropout=dropout,
+        )  # is self-attn if context is none
+        self.norm1 = nn.LayerNorm(dim)
+        self.norm2 = nn.LayerNorm(dim)
+        self.norm3 = nn.LayerNorm(dim)
+        self.checkpoint = checkpoint
+
+    def forward(self, x, context=None):
+        return checkpoint(
+            self._forward, (x, context), self.parameters(), self.checkpoint
+        )
+
+    def _forward(self, x, context=None):
+        x = (
+            self.attn1(
+                self.norm1(x), context=context if self.disable_self_attn else None
+            )
+            + x
+        )
+        x = self.attn2(self.norm2(x), context=context) + x
+        x = self.ff(self.norm3(x)) + x
+        return x
+
+
+class SpatialTransformer(nn.Module):
+    """
+    Transformer block for image-like data.
+    First, project the input (aka embedding)
+    and reshape to b, t, d.
+    Then apply standard transformer action.
+    Finally, reshape to image
+    NEW: use_linear for more efficiency instead of the 1x1 convs
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        n_heads,
+        d_head,
+        depth=1,
+        dropout=0.0,
+        context_dim=None,
+        disable_self_attn=False,
+        use_linear=False,
+        use_checkpoint=True,
+    ):
+        super().__init__()
+        if exists(context_dim) and not isinstance(context_dim, list):
+            context_dim = [context_dim]
+        self.in_channels = in_channels
+        inner_dim = n_heads * d_head
+        self.norm = Normalize(in_channels)
+        if not use_linear:
+            self.proj_in = nn.Conv2d(
+                in_channels, inner_dim, kernel_size=1, stride=1, padding=0
+            )
+        else:
+            self.proj_in = nn.Linear(in_channels, inner_dim)
+
+        self.transformer_blocks = nn.ModuleList(
+            [
+                BasicTransformerBlock(
+                    inner_dim,
+                    n_heads,
+                    d_head,
+                    dropout=dropout,
+                    context_dim=context_dim[d],
+                    disable_self_attn=disable_self_attn,
+                    checkpoint=use_checkpoint,
+                )
+                for d in range(depth)
+            ]
+        )
+        if not use_linear:
+            self.proj_out = zero_module(
+                nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)
+            )
+        else:
+            self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
+        self.use_linear = use_linear
+
+    def forward(self, x, context=None):
+        # note: if no context is given, cross-attention defaults to self-attention
+        if not isinstance(context, list):
+            context = [context]
+        b, c, h, w = x.shape
+        x_in = x
+        x = self.norm(x)
+        if not self.use_linear:
+            x = self.proj_in(x)
+        x = rearrange(x, "b c h w -> b (h w) c").contiguous()
+        if self.use_linear:
+            x = self.proj_in(x)
+        for i, block in enumerate(self.transformer_blocks):
+            x = block(x, context=context[i])
+        if self.use_linear:
+            x = self.proj_out(x)
+        x = rearrange(x, "b (h w) c -> b c h w", h=h, w=w).contiguous()
+        if not self.use_linear:
+            x = self.proj_out(x)
+        return x + x_in

iopaint/model/anytext/ldm/modules/diffusionmodules/model.py

@@ -0,0 +1,973 @@
+# pytorch_diffusion + derived encoder decoder
+import math
+
+import numpy as np
+import torch
+import torch.nn as nn
+
+
+def get_timestep_embedding(timesteps, embedding_dim):
+    """
+    This matches the implementation in Denoising Diffusion Probabilistic Models:
+    From Fairseq.
+    Build sinusoidal embeddings.
+    This matches the implementation in tensor2tensor, but differs slightly
+    from the description in Section 3.5 of "Attention Is All You Need".
+    """
+    assert len(timesteps.shape) == 1
+
+    half_dim = embedding_dim // 2
+    emb = math.log(10000) / (half_dim - 1)
+    emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
+    emb = emb.to(device=timesteps.device)
+    emb = timesteps.float()[:, None] * emb[None, :]
+    emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+    if embedding_dim % 2 == 1:  # zero pad
+        emb = torch.nn.functional.pad(emb, (0, 1, 0, 0))
+    return emb
+
+
+def nonlinearity(x):
+    # swish
+    return x * torch.sigmoid(x)
+
+
+def Normalize(in_channels, num_groups=32):
+    return torch.nn.GroupNorm(
+        num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True
+    )
+
+
+class Upsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            self.conv = torch.nn.Conv2d(
+                in_channels, in_channels, kernel_size=3, stride=1, padding=1
+            )
+
+    def forward(self, x):
+        x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
+        if self.with_conv:
+            x = self.conv(x)
+        return x
+
+
+class Downsample(nn.Module):
+    def __init__(self, in_channels, with_conv):
+        super().__init__()
+        self.with_conv = with_conv
+        if self.with_conv:
+            # no asymmetric padding in torch conv, must do it ourselves
+            self.conv = torch.nn.Conv2d(
+                in_channels, in_channels, kernel_size=3, stride=2, padding=0
+            )
+
+    def forward(self, x):
+        if self.with_conv:
+            pad = (0, 1, 0, 1)
+            x = torch.nn.functional.pad(x, pad, mode="constant", value=0)
+            x = self.conv(x)
+        else:
+            x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2)
+        return x
+
+
+class ResnetBlock(nn.Module):
+    def __init__(
+        self,
+        *,
+        in_channels,
+        out_channels=None,
+        conv_shortcut=False,
+        dropout,
+        temb_channels=512,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        out_channels = in_channels if out_channels is None else out_channels
+        self.out_channels = out_channels
+        self.use_conv_shortcut = conv_shortcut
+
+        self.norm1 = Normalize(in_channels)
+        self.conv1 = torch.nn.Conv2d(
+            in_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+        if temb_channels > 0:
+            self.temb_proj = torch.nn.Linear(temb_channels, out_channels)
+        self.norm2 = Normalize(out_channels)
+        self.dropout = torch.nn.Dropout(dropout)
+        self.conv2 = torch.nn.Conv2d(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                self.conv_shortcut = torch.nn.Conv2d(
+                    in_channels, out_channels, kernel_size=3, stride=1, padding=1
+                )
+            else:
+                self.nin_shortcut = torch.nn.Conv2d(
+                    in_channels, out_channels, kernel_size=1, stride=1, padding=0
+                )
+
+    def forward(self, x, temb):
+        h = x
+        h = self.norm1(h)
+        h = nonlinearity(h)
+        h = self.conv1(h)
+
+        if temb is not None:
+            h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None]
+
+        h = self.norm2(h)
+        h = nonlinearity(h)
+        h = self.dropout(h)
+        h = self.conv2(h)
+
+        if self.in_channels != self.out_channels:
+            if self.use_conv_shortcut:
+                x = self.conv_shortcut(x)
+            else:
+                x = self.nin_shortcut(x)
+
+        return x + h
+
+
+class AttnBlock(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.k = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.v = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.proj_out = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        b, c, h, w = q.shape
+        q = q.reshape(b, c, h * w)
+        q = q.permute(0, 2, 1)  # b,hw,c
+        k = k.reshape(b, c, h * w)  # b,c,hw
+        w_ = torch.bmm(q, k)  # b,hw,hw    w[b,i,j]=sum_c q[b,i,c]k[b,c,j]
+        w_ = w_ * (int(c) ** (-0.5))
+        w_ = torch.nn.functional.softmax(w_, dim=2)
+
+        # attend to values
+        v = v.reshape(b, c, h * w)
+        w_ = w_.permute(0, 2, 1)  # b,hw,hw (first hw of k, second of q)
+        h_ = torch.bmm(v, w_)  # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j]
+        h_ = h_.reshape(b, c, h, w)
+
+        h_ = self.proj_out(h_)
+
+        return x + h_
+
+
+class AttnBlock2_0(nn.Module):
+    def __init__(self, in_channels):
+        super().__init__()
+        self.in_channels = in_channels
+
+        self.norm = Normalize(in_channels)
+        self.q = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.k = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.v = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.proj_out = torch.nn.Conv2d(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+
+    def forward(self, x):
+        h_ = x
+        h_ = self.norm(h_)
+        # output: [1, 512, 64, 64]
+        q = self.q(h_)
+        k = self.k(h_)
+        v = self.v(h_)
+
+        # compute attention
+        b, c, h, w = q.shape
+
+        # q = q.reshape(b, c, h * w).transpose()
+        # q = q.permute(0, 2, 1)  # b,hw,c
+        # k = k.reshape(b, c, h * w)  # b,c,hw
+        q = q.transpose(1, 2)
+        k = k.transpose(1, 2)
+        v = v.transpose(1, 2)
+        # (batch, num_heads, seq_len, head_dim)
+        hidden_states = torch.nn.functional.scaled_dot_product_attention(
+            q, k, v, attn_mask=None, dropout_p=0.0, is_causal=False
+        )
+        hidden_states = hidden_states.transpose(1, 2)
+        hidden_states = hidden_states.to(q.dtype)
+
+        h_ = self.proj_out(hidden_states)
+
+        return x + h_
+
+
+def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
+    assert attn_type in [
+        "vanilla",
+        "vanilla-xformers",
+        "memory-efficient-cross-attn",
+        "linear",
+        "none",
+    ], f"attn_type {attn_type} unknown"
+    assert attn_kwargs is None
+    if hasattr(torch.nn.functional, "scaled_dot_product_attention"):
+        # print(f"Using torch.nn.functional.scaled_dot_product_attention")
+        return AttnBlock2_0(in_channels)
+    return AttnBlock(in_channels)
+
+
+class Model(nn.Module):
+    def __init__(
+        self,
+        *,
+        ch,
+        out_ch,
+        ch_mult=(1, 2, 4, 8),
+        num_res_blocks,
+        attn_resolutions,
+        dropout=0.0,
+        resamp_with_conv=True,
+        in_channels,
+        resolution,
+        use_timestep=True,
+        use_linear_attn=False,
+        attn_type="vanilla",
+    ):
+        super().__init__()
+        if use_linear_attn:
+            attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = self.ch * 4
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        self.use_timestep = use_timestep
+        if self.use_timestep:
+            # timestep embedding
+            self.temb = nn.Module()
+            self.temb.dense = nn.ModuleList(
+                [
+                    torch.nn.Linear(self.ch, self.temb_ch),
+                    torch.nn.Linear(self.temb_ch, self.temb_ch),
+                ]
+            )
+
+        # downsampling
+        self.conv_in = torch.nn.Conv2d(
+            in_channels, self.ch, kernel_size=3, stride=1, padding=1
+        )
+
+        curr_res = resolution
+        in_ch_mult = (1,) + tuple(ch_mult)
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch * in_ch_mult[i_level]
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(
+                    ResnetBlock(
+                        in_channels=block_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions - 1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch * ch_mult[i_level]
+            skip_in = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks + 1):
+                if i_block == self.num_res_blocks:
+                    skip_in = ch * in_ch_mult[i_level]
+                block.append(
+                    ResnetBlock(
+                        in_channels=block_in + skip_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up)  # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(
+            block_in, out_ch, kernel_size=3, stride=1, padding=1
+        )
+
+    def forward(self, x, t=None, context=None):
+        # assert x.shape[2] == x.shape[3] == self.resolution
+        if context is not None:
+            # assume aligned context, cat along channel axis
+            x = torch.cat((x, context), dim=1)
+        if self.use_timestep:
+            # timestep embedding
+            assert t is not None
+            temb = get_timestep_embedding(t, self.ch)
+            temb = self.temb.dense[0](temb)
+            temb = nonlinearity(temb)
+            temb = self.temb.dense[1](temb)
+        else:
+            temb = None
+
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1], temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions - 1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.up[i_level].block[i_block](
+                    torch.cat([h, hs.pop()], dim=1), temb
+                )
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+    def get_last_layer(self):
+        return self.conv_out.weight
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        *,
+        ch,
+        out_ch,
+        ch_mult=(1, 2, 4, 8),
+        num_res_blocks,
+        attn_resolutions,
+        dropout=0.0,
+        resamp_with_conv=True,
+        in_channels,
+        resolution,
+        z_channels,
+        double_z=True,
+        use_linear_attn=False,
+        attn_type="vanilla",
+        **ignore_kwargs,
+    ):
+        super().__init__()
+        if use_linear_attn:
+            attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+
+        # downsampling
+        self.conv_in = torch.nn.Conv2d(
+            in_channels, self.ch, kernel_size=3, stride=1, padding=1
+        )
+
+        curr_res = resolution
+        in_ch_mult = (1,) + tuple(ch_mult)
+        self.in_ch_mult = in_ch_mult
+        self.down = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_in = ch * in_ch_mult[i_level]
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks):
+                block.append(
+                    ResnetBlock(
+                        in_channels=block_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            down = nn.Module()
+            down.block = block
+            down.attn = attn
+            if i_level != self.num_resolutions - 1:
+                down.downsample = Downsample(block_in, resamp_with_conv)
+                curr_res = curr_res // 2
+            self.down.append(down)
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(
+            block_in,
+            2 * z_channels if double_z else z_channels,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+        )
+
+    def forward(self, x):
+        # timestep embedding
+        temb = None
+
+        # downsampling
+        hs = [self.conv_in(x)]
+        for i_level in range(self.num_resolutions):
+            for i_block in range(self.num_res_blocks):
+                h = self.down[i_level].block[i_block](hs[-1], temb)
+                if len(self.down[i_level].attn) > 0:
+                    h = self.down[i_level].attn[i_block](h)
+                hs.append(h)
+            if i_level != self.num_resolutions - 1:
+                hs.append(self.down[i_level].downsample(hs[-1]))
+
+        # middle
+        h = hs[-1]
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # end
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        *,
+        ch,
+        out_ch,
+        ch_mult=(1, 2, 4, 8),
+        num_res_blocks,
+        attn_resolutions,
+        dropout=0.0,
+        resamp_with_conv=True,
+        in_channels,
+        resolution,
+        z_channels,
+        give_pre_end=False,
+        tanh_out=False,
+        use_linear_attn=False,
+        attn_type="vanilla",
+        **ignorekwargs,
+    ):
+        super().__init__()
+        if use_linear_attn:
+            attn_type = "linear"
+        self.ch = ch
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        self.resolution = resolution
+        self.in_channels = in_channels
+        self.give_pre_end = give_pre_end
+        self.tanh_out = tanh_out
+
+        # compute in_ch_mult, block_in and curr_res at lowest res
+        in_ch_mult = (1,) + tuple(ch_mult)
+        block_in = ch * ch_mult[self.num_resolutions - 1]
+        curr_res = resolution // 2 ** (self.num_resolutions - 1)
+        self.z_shape = (1, z_channels, curr_res, curr_res)
+        print(
+            "Working with z of shape {} = {} dimensions.".format(
+                self.z_shape, np.prod(self.z_shape)
+            )
+        )
+
+        # z to block_in
+        self.conv_in = torch.nn.Conv2d(
+            z_channels, block_in, kernel_size=3, stride=1, padding=1
+        )
+
+        # middle
+        self.mid = nn.Module()
+        self.mid.block_1 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+        self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
+        self.mid.block_2 = ResnetBlock(
+            in_channels=block_in,
+            out_channels=block_in,
+            temb_channels=self.temb_ch,
+            dropout=dropout,
+        )
+
+        # upsampling
+        self.up = nn.ModuleList()
+        for i_level in reversed(range(self.num_resolutions)):
+            block = nn.ModuleList()
+            attn = nn.ModuleList()
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks + 1):
+                block.append(
+                    ResnetBlock(
+                        in_channels=block_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+                if curr_res in attn_resolutions:
+                    attn.append(make_attn(block_in, attn_type=attn_type))
+            up = nn.Module()
+            up.block = block
+            up.attn = attn
+            if i_level != 0:
+                up.upsample = Upsample(block_in, resamp_with_conv)
+                curr_res = curr_res * 2
+            self.up.insert(0, up)  # prepend to get consistent order
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(
+            block_in, out_ch, kernel_size=3, stride=1, padding=1
+        )
+
+    def forward(self, z):
+        # assert z.shape[1:] == self.z_shape[1:]
+        self.last_z_shape = z.shape
+
+        # timestep embedding
+        temb = None
+
+        # z to block_in
+        h = self.conv_in(z)
+
+        # middle
+        h = self.mid.block_1(h, temb)
+        h = self.mid.attn_1(h)
+        h = self.mid.block_2(h, temb)
+
+        # upsampling
+        for i_level in reversed(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.up[i_level].block[i_block](h, temb)
+                if len(self.up[i_level].attn) > 0:
+                    h = self.up[i_level].attn[i_block](h)
+            if i_level != 0:
+                h = self.up[i_level].upsample(h)
+
+        # end
+        if self.give_pre_end:
+            return h
+
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        if self.tanh_out:
+            h = torch.tanh(h)
+        return h
+
+
+class SimpleDecoder(nn.Module):
+    def __init__(self, in_channels, out_channels, *args, **kwargs):
+        super().__init__()
+        self.model = nn.ModuleList(
+            [
+                nn.Conv2d(in_channels, in_channels, 1),
+                ResnetBlock(
+                    in_channels=in_channels,
+                    out_channels=2 * in_channels,
+                    temb_channels=0,
+                    dropout=0.0,
+                ),
+                ResnetBlock(
+                    in_channels=2 * in_channels,
+                    out_channels=4 * in_channels,
+                    temb_channels=0,
+                    dropout=0.0,
+                ),
+                ResnetBlock(
+                    in_channels=4 * in_channels,
+                    out_channels=2 * in_channels,
+                    temb_channels=0,
+                    dropout=0.0,
+                ),
+                nn.Conv2d(2 * in_channels, in_channels, 1),
+                Upsample(in_channels, with_conv=True),
+            ]
+        )
+        # end
+        self.norm_out = Normalize(in_channels)
+        self.conv_out = torch.nn.Conv2d(
+            in_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+
+    def forward(self, x):
+        for i, layer in enumerate(self.model):
+            if i in [1, 2, 3]:
+                x = layer(x, None)
+            else:
+                x = layer(x)
+
+        h = self.norm_out(x)
+        h = nonlinearity(h)
+        x = self.conv_out(h)
+        return x
+
+
+class UpsampleDecoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        ch,
+        num_res_blocks,
+        resolution,
+        ch_mult=(2, 2),
+        dropout=0.0,
+    ):
+        super().__init__()
+        # upsampling
+        self.temb_ch = 0
+        self.num_resolutions = len(ch_mult)
+        self.num_res_blocks = num_res_blocks
+        block_in = in_channels
+        curr_res = resolution // 2 ** (self.num_resolutions - 1)
+        self.res_blocks = nn.ModuleList()
+        self.upsample_blocks = nn.ModuleList()
+        for i_level in range(self.num_resolutions):
+            res_block = []
+            block_out = ch * ch_mult[i_level]
+            for i_block in range(self.num_res_blocks + 1):
+                res_block.append(
+                    ResnetBlock(
+                        in_channels=block_in,
+                        out_channels=block_out,
+                        temb_channels=self.temb_ch,
+                        dropout=dropout,
+                    )
+                )
+                block_in = block_out
+            self.res_blocks.append(nn.ModuleList(res_block))
+            if i_level != self.num_resolutions - 1:
+                self.upsample_blocks.append(Upsample(block_in, True))
+                curr_res = curr_res * 2
+
+        # end
+        self.norm_out = Normalize(block_in)
+        self.conv_out = torch.nn.Conv2d(
+            block_in, out_channels, kernel_size=3, stride=1, padding=1
+        )
+
+    def forward(self, x):
+        # upsampling
+        h = x
+        for k, i_level in enumerate(range(self.num_resolutions)):
+            for i_block in range(self.num_res_blocks + 1):
+                h = self.res_blocks[i_level][i_block](h, None)
+            if i_level != self.num_resolutions - 1:
+                h = self.upsample_blocks[k](h)
+        h = self.norm_out(h)
+        h = nonlinearity(h)
+        h = self.conv_out(h)
+        return h
+
+
+class LatentRescaler(nn.Module):
+    def __init__(self, factor, in_channels, mid_channels, out_channels, depth=2):
+        super().__init__()
+        # residual block, interpolate, residual block
+        self.factor = factor
+        self.conv_in = nn.Conv2d(
+            in_channels, mid_channels, kernel_size=3, stride=1, padding=1
+        )
+        self.res_block1 = nn.ModuleList(
+            [
+                ResnetBlock(
+                    in_channels=mid_channels,
+                    out_channels=mid_channels,
+                    temb_channels=0,
+                    dropout=0.0,
+                )
+                for _ in range(depth)
+            ]
+        )
+        self.attn = AttnBlock(mid_channels)
+        self.res_block2 = nn.ModuleList(
+            [
+                ResnetBlock(
+                    in_channels=mid_channels,
+                    out_channels=mid_channels,
+                    temb_channels=0,
+                    dropout=0.0,
+                )
+                for _ in range(depth)
+            ]
+        )
+
+        self.conv_out = nn.Conv2d(
+            mid_channels,
+            out_channels,
+            kernel_size=1,
+        )
+
+    def forward(self, x):
+        x = self.conv_in(x)
+        for block in self.res_block1:
+            x = block(x, None)
+        x = torch.nn.functional.interpolate(
+            x,
+            size=(
+                int(round(x.shape[2] * self.factor)),
+                int(round(x.shape[3] * self.factor)),
+            ),
+        )
+        x = self.attn(x)
+        for block in self.res_block2:
+            x = block(x, None)
+        x = self.conv_out(x)
+        return x
+
+
+class MergedRescaleEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        ch,
+        resolution,
+        out_ch,
+        num_res_blocks,
+        attn_resolutions,
+        dropout=0.0,
+        resamp_with_conv=True,
+        ch_mult=(1, 2, 4, 8),
+        rescale_factor=1.0,
+        rescale_module_depth=1,
+    ):
+        super().__init__()
+        intermediate_chn = ch * ch_mult[-1]
+        self.encoder = Encoder(
+            in_channels=in_channels,
+            num_res_blocks=num_res_blocks,
+            ch=ch,
+            ch_mult=ch_mult,
+            z_channels=intermediate_chn,
+            double_z=False,
+            resolution=resolution,
+            attn_resolutions=attn_resolutions,
+            dropout=dropout,
+            resamp_with_conv=resamp_with_conv,
+            out_ch=None,
+        )
+        self.rescaler = LatentRescaler(
+            factor=rescale_factor,
+            in_channels=intermediate_chn,
+            mid_channels=intermediate_chn,
+            out_channels=out_ch,
+            depth=rescale_module_depth,
+        )
+
+    def forward(self, x):
+        x = self.encoder(x)
+        x = self.rescaler(x)
+        return x
+
+
+class MergedRescaleDecoder(nn.Module):
+    def __init__(
+        self,
+        z_channels,
+        out_ch,
+        resolution,
+        num_res_blocks,
+        attn_resolutions,
+        ch,
+        ch_mult=(1, 2, 4, 8),
+        dropout=0.0,
+        resamp_with_conv=True,
+        rescale_factor=1.0,
+        rescale_module_depth=1,
+    ):
+        super().__init__()
+        tmp_chn = z_channels * ch_mult[-1]
+        self.decoder = Decoder(
+            out_ch=out_ch,
+            z_channels=tmp_chn,
+            attn_resolutions=attn_resolutions,
+            dropout=dropout,
+            resamp_with_conv=resamp_with_conv,
+            in_channels=None,
+            num_res_blocks=num_res_blocks,
+            ch_mult=ch_mult,
+            resolution=resolution,
+            ch=ch,
+        )
+        self.rescaler = LatentRescaler(
+            factor=rescale_factor,
+            in_channels=z_channels,
+            mid_channels=tmp_chn,
+            out_channels=tmp_chn,
+            depth=rescale_module_depth,
+        )
+
+    def forward(self, x):
+        x = self.rescaler(x)
+        x = self.decoder(x)
+        return x
+
+
+class Upsampler(nn.Module):
+    def __init__(self, in_size, out_size, in_channels, out_channels, ch_mult=2):
+        super().__init__()
+        assert out_size >= in_size
+        num_blocks = int(np.log2(out_size // in_size)) + 1
+        factor_up = 1.0 + (out_size % in_size)
+        print(
+            f"Building {self.__class__.__name__} with in_size: {in_size} --> out_size {out_size} and factor {factor_up}"
+        )
+        self.rescaler = LatentRescaler(
+            factor=factor_up,
+            in_channels=in_channels,
+            mid_channels=2 * in_channels,
+            out_channels=in_channels,
+        )
+        self.decoder = Decoder(
+            out_ch=out_channels,
+            resolution=out_size,
+            z_channels=in_channels,
+            num_res_blocks=2,
+            attn_resolutions=[],
+            in_channels=None,
+            ch=in_channels,
+            ch_mult=[ch_mult for _ in range(num_blocks)],
+        )
+
+    def forward(self, x):
+        x = self.rescaler(x)
+        x = self.decoder(x)
+        return x
+
+
+class Resize(nn.Module):
+    def __init__(self, in_channels=None, learned=False, mode="bilinear"):
+        super().__init__()
+        self.with_conv = learned
+        self.mode = mode
+        if self.with_conv:
+            print(
+                f"Note: {self.__class__.__name} uses learned downsampling and will ignore the fixed {mode} mode"
+            )
+            raise NotImplementedError()
+            assert in_channels is not None
+            # no asymmetric padding in torch conv, must do it ourselves
+            self.conv = torch.nn.Conv2d(
+                in_channels, in_channels, kernel_size=4, stride=2, padding=1
+            )
+
+    def forward(self, x, scale_factor=1.0):
+        if scale_factor == 1.0:
+            return x
+        else:
+            x = torch.nn.functional.interpolate(
+                x, mode=self.mode, align_corners=False, scale_factor=scale_factor
+            )
+        return x

iopaint/model/anytext/ldm/modules/diffusionmodules/openaimodel.py

@@ -0,0 +1,786 @@
+from abc import abstractmethod
+import math
+
+import numpy as np
+import torch as th
+import torch.nn as nn
+import torch.nn.functional as F
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import (
+    checkpoint,
+    conv_nd,
+    linear,
+    avg_pool_nd,
+    zero_module,
+    normalization,
+    timestep_embedding,
+)
+from iopaint.model.anytext.ldm.modules.attention import SpatialTransformer
+from iopaint.model.anytext.ldm.util import exists
+
+
+# dummy replace
+def convert_module_to_f16(x):
+    pass
+
+def convert_module_to_f32(x):
+    pass
+
+
+## go
+class AttentionPool2d(nn.Module):
+    """
+    Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
+    """
+
+    def __init__(
+        self,
+        spacial_dim: int,
+        embed_dim: int,
+        num_heads_channels: int,
+        output_dim: int = None,
+    ):
+        super().__init__()
+        self.positional_embedding = nn.Parameter(th.randn(embed_dim, spacial_dim ** 2 + 1) / embed_dim ** 0.5)
+        self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
+        self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
+        self.num_heads = embed_dim // num_heads_channels
+        self.attention = QKVAttention(self.num_heads)
+
+    def forward(self, x):
+        b, c, *_spatial = x.shape
+        x = x.reshape(b, c, -1)  # NC(HW)
+        x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1)  # NC(HW+1)
+        x = x + self.positional_embedding[None, :, :].to(x.dtype)  # NC(HW+1)
+        x = self.qkv_proj(x)
+        x = self.attention(x)
+        x = self.c_proj(x)
+        return x[:, :, 0]
+
+
+class TimestepBlock(nn.Module):
+    """
+    Any module where forward() takes timestep embeddings as a second argument.
+    """
+
+    @abstractmethod
+    def forward(self, x, emb):
+        """
+        Apply the module to `x` given `emb` timestep embeddings.
+        """
+
+
+class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
+    """
+    A sequential module that passes timestep embeddings to the children that
+    support it as an extra input.
+    """
+
+    def forward(self, x, emb, context=None):
+        for layer in self:
+            if isinstance(layer, TimestepBlock):
+                x = layer(x, emb)
+            elif isinstance(layer, SpatialTransformer):
+                x = layer(x, context)
+            else:
+                x = layer(x)
+        return x
+
+
+class Upsample(nn.Module):
+    """
+    An upsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 upsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        if use_conv:
+            self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        if self.dims == 3:
+            x = F.interpolate(
+                x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
+            )
+        else:
+            x = F.interpolate(x, scale_factor=2, mode="nearest")
+        if self.use_conv:
+            x = self.conv(x)
+        return x
+
+class TransposedUpsample(nn.Module):
+    'Learned 2x upsampling without padding'
+    def __init__(self, channels, out_channels=None, ks=5):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+
+        self.up = nn.ConvTranspose2d(self.channels,self.out_channels,kernel_size=ks,stride=2)
+
+    def forward(self,x):
+        return self.up(x)
+
+
+class Downsample(nn.Module):
+    """
+    A downsampling layer with an optional convolution.
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 downsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None,padding=1):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        stride = 2 if dims != 3 else (1, 2, 2)
+        if use_conv:
+            self.op = conv_nd(
+                dims, self.channels, self.out_channels, 3, stride=stride, padding=padding
+            )
+        else:
+            assert self.channels == self.out_channels
+            self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        return self.op(x)
+
+
+class ResBlock(TimestepBlock):
+    """
+    A residual block that can optionally change the number of channels.
+    :param channels: the number of input channels.
+    :param emb_channels: the number of timestep embedding channels.
+    :param dropout: the rate of dropout.
+    :param out_channels: if specified, the number of out channels.
+    :param use_conv: if True and out_channels is specified, use a spatial
+        convolution instead of a smaller 1x1 convolution to change the
+        channels in the skip connection.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param use_checkpoint: if True, use gradient checkpointing on this module.
+    :param up: if True, use this block for upsampling.
+    :param down: if True, use this block for downsampling.
+    """
+
+    def __init__(
+        self,
+        channels,
+        emb_channels,
+        dropout,
+        out_channels=None,
+        use_conv=False,
+        use_scale_shift_norm=False,
+        dims=2,
+        use_checkpoint=False,
+        up=False,
+        down=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.emb_channels = emb_channels
+        self.dropout = dropout
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.use_checkpoint = use_checkpoint
+        self.use_scale_shift_norm = use_scale_shift_norm
+
+        self.in_layers = nn.Sequential(
+            normalization(channels),
+            nn.SiLU(),
+            conv_nd(dims, channels, self.out_channels, 3, padding=1),
+        )
+
+        self.updown = up or down
+
+        if up:
+            self.h_upd = Upsample(channels, False, dims)
+            self.x_upd = Upsample(channels, False, dims)
+        elif down:
+            self.h_upd = Downsample(channels, False, dims)
+            self.x_upd = Downsample(channels, False, dims)
+        else:
+            self.h_upd = self.x_upd = nn.Identity()
+
+        self.emb_layers = nn.Sequential(
+            nn.SiLU(),
+            linear(
+                emb_channels,
+                2 * self.out_channels if use_scale_shift_norm else self.out_channels,
+            ),
+        )
+        self.out_layers = nn.Sequential(
+            normalization(self.out_channels),
+            nn.SiLU(),
+            nn.Dropout(p=dropout),
+            zero_module(
+                conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
+            ),
+        )
+
+        if self.out_channels == channels:
+            self.skip_connection = nn.Identity()
+        elif use_conv:
+            self.skip_connection = conv_nd(
+                dims, channels, self.out_channels, 3, padding=1
+            )
+        else:
+            self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
+
+    def forward(self, x, emb):
+        """
+        Apply the block to a Tensor, conditioned on a timestep embedding.
+        :param x: an [N x C x ...] Tensor of features.
+        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        return checkpoint(
+            self._forward, (x, emb), self.parameters(), self.use_checkpoint
+        )
+
+
+    def _forward(self, x, emb):
+        if self.updown:
+            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
+            h = in_rest(x)
+            h = self.h_upd(h)
+            x = self.x_upd(x)
+            h = in_conv(h)
+        else:
+            h = self.in_layers(x)
+        emb_out = self.emb_layers(emb).type(h.dtype)
+        while len(emb_out.shape) < len(h.shape):
+            emb_out = emb_out[..., None]
+        if self.use_scale_shift_norm:
+            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
+            scale, shift = th.chunk(emb_out, 2, dim=1)
+            h = out_norm(h) * (1 + scale) + shift
+            h = out_rest(h)
+        else:
+            h = h + emb_out
+            h = self.out_layers(h)
+        return self.skip_connection(x) + h
+
+
+class AttentionBlock(nn.Module):
+    """
+    An attention block that allows spatial positions to attend to each other.
+    Originally ported from here, but adapted to the N-d case.
+    https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
+    """
+
+    def __init__(
+        self,
+        channels,
+        num_heads=1,
+        num_head_channels=-1,
+        use_checkpoint=False,
+        use_new_attention_order=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        if num_head_channels == -1:
+            self.num_heads = num_heads
+        else:
+            assert (
+                channels % num_head_channels == 0
+            ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
+            self.num_heads = channels // num_head_channels
+        self.use_checkpoint = use_checkpoint
+        self.norm = normalization(channels)
+        self.qkv = conv_nd(1, channels, channels * 3, 1)
+        if use_new_attention_order:
+            # split qkv before split heads
+            self.attention = QKVAttention(self.num_heads)
+        else:
+            # split heads before split qkv
+            self.attention = QKVAttentionLegacy(self.num_heads)
+
+        self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
+
+    def forward(self, x):
+        return checkpoint(self._forward, (x,), self.parameters(), True)   # TODO: check checkpoint usage, is True # TODO: fix the .half call!!!
+        #return pt_checkpoint(self._forward, x)  # pytorch
+
+    def _forward(self, x):
+        b, c, *spatial = x.shape
+        x = x.reshape(b, c, -1)
+        qkv = self.qkv(self.norm(x))
+        h = self.attention(qkv)
+        h = self.proj_out(h)
+        return (x + h).reshape(b, c, *spatial)
+
+
+def count_flops_attn(model, _x, y):
+    """
+    A counter for the `thop` package to count the operations in an
+    attention operation.
+    Meant to be used like:
+        macs, params = thop.profile(
+            model,
+            inputs=(inputs, timestamps),
+            custom_ops={QKVAttention: QKVAttention.count_flops},
+        )
+    """
+    b, c, *spatial = y[0].shape
+    num_spatial = int(np.prod(spatial))
+    # We perform two matmuls with the same number of ops.
+    # The first computes the weight matrix, the second computes
+    # the combination of the value vectors.
+    matmul_ops = 2 * b * (num_spatial ** 2) * c
+    model.total_ops += th.DoubleTensor([matmul_ops])
+
+
+class QKVAttentionLegacy(nn.Module):
+    """
+    A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+        :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts", q * scale, k * scale
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v)
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class QKVAttention(nn.Module):
+    """
+    A module which performs QKV attention and splits in a different order.
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+        :param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.chunk(3, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts",
+            (q * scale).view(bs * self.n_heads, ch, length),
+            (k * scale).view(bs * self.n_heads, ch, length),
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class UNetModel(nn.Module):
+    """
+    The full UNet model with attention and timestep embedding.
+    :param in_channels: channels in the input Tensor.
+    :param model_channels: base channel count for the model.
+    :param out_channels: channels in the output Tensor.
+    :param num_res_blocks: number of residual blocks per downsample.
+    :param attention_resolutions: a collection of downsample rates at which
+        attention will take place. May be a set, list, or tuple.
+        For example, if this contains 4, then at 4x downsampling, attention
+        will be used.
+    :param dropout: the dropout probability.
+    :param channel_mult: channel multiplier for each level of the UNet.
+    :param conv_resample: if True, use learned convolutions for upsampling and
+        downsampling.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param num_classes: if specified (as an int), then this model will be
+        class-conditional with `num_classes` classes.
+    :param use_checkpoint: use gradient checkpointing to reduce memory usage.
+    :param num_heads: the number of attention heads in each attention layer.
+    :param num_heads_channels: if specified, ignore num_heads and instead use
+                               a fixed channel width per attention head.
+    :param num_heads_upsample: works with num_heads to set a different number
+                               of heads for upsampling. Deprecated.
+    :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
+    :param resblock_updown: use residual blocks for up/downsampling.
+    :param use_new_attention_order: use a different attention pattern for potentially
+                                    increased efficiency.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        attention_resolutions,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        num_classes=None,
+        use_checkpoint=False,
+        use_fp16=False,
+        num_heads=-1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        use_spatial_transformer=False,    # custom transformer support
+        transformer_depth=1,              # custom transformer support
+        context_dim=None,                 # custom transformer support
+        n_embed=None,                     # custom support for prediction of discrete ids into codebook of first stage vq model
+        legacy=True,
+        disable_self_attentions=None,
+        num_attention_blocks=None,
+        disable_middle_self_attn=False,
+        use_linear_in_transformer=False,
+    ):
+        super().__init__()
+        if use_spatial_transformer:
+            assert context_dim is not None, 'Fool!! You forgot to include the dimension of your cross-attention conditioning...'
+
+        if context_dim is not None:
+            assert use_spatial_transformer, 'Fool!! You forgot to use the spatial transformer for your cross-attention conditioning...'
+            from omegaconf.listconfig import ListConfig
+            if type(context_dim) == ListConfig:
+                context_dim = list(context_dim)
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        if num_heads == -1:
+            assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set'
+
+        if num_head_channels == -1:
+            assert num_heads != -1, 'Either num_heads or num_head_channels has to be set'
+
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+        if isinstance(num_res_blocks, int):
+            self.num_res_blocks = len(channel_mult) * [num_res_blocks]
+        else:
+            if len(num_res_blocks) != len(channel_mult):
+                raise ValueError("provide num_res_blocks either as an int (globally constant) or "
+                                 "as a list/tuple (per-level) with the same length as channel_mult")
+            self.num_res_blocks = num_res_blocks
+        if disable_self_attentions is not None:
+            # should be a list of booleans, indicating whether to disable self-attention in TransformerBlocks or not
+            assert len(disable_self_attentions) == len(channel_mult)
+        if num_attention_blocks is not None:
+            assert len(num_attention_blocks) == len(self.num_res_blocks)
+            assert all(map(lambda i: self.num_res_blocks[i] >= num_attention_blocks[i], range(len(num_attention_blocks))))
+            print(f"Constructor of UNetModel received num_attention_blocks={num_attention_blocks}. "
+                  f"This option has LESS priority than attention_resolutions {attention_resolutions}, "
+                  f"i.e., in cases where num_attention_blocks[i] > 0 but 2**i not in attention_resolutions, "
+                  f"attention will still not be set.")
+        self.use_fp16 = use_fp16
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.num_classes = num_classes
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+        self.predict_codebook_ids = n_embed is not None
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        if self.num_classes is not None:
+            if isinstance(self.num_classes, int):
+                self.label_emb = nn.Embedding(num_classes, time_embed_dim)
+            elif self.num_classes == "continuous":
+                print("setting up linear c_adm embedding layer")
+                self.label_emb = nn.Linear(1, time_embed_dim)
+            else:
+                raise ValueError()
+
+        self.input_blocks = nn.ModuleList(
+            [
+                TimestepEmbedSequential(
+                    conv_nd(dims, in_channels, model_channels, 3, padding=1)
+                )
+            ]
+        )
+        self._feature_size = model_channels
+        input_block_chans = [model_channels]
+        ch = model_channels
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for nr in range(self.num_res_blocks[level]):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=mult * model_channels,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = mult * model_channels
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or nr < num_attention_blocks[level]:
+                        layers.append(
+                            AttentionBlock(
+                                ch,
+                                use_checkpoint=use_checkpoint,
+                                num_heads=num_heads,
+                                num_head_channels=dim_head,
+                                use_new_attention_order=use_new_attention_order,
+                            ) if not use_spatial_transformer else SpatialTransformer(
+                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint
+                            )
+                        )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        if num_head_channels == -1:
+            dim_head = ch // num_heads
+        else:
+            num_heads = ch // num_head_channels
+            dim_head = num_head_channels
+        if legacy:
+            #num_heads = 1
+            dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=dim_head,
+                use_new_attention_order=use_new_attention_order,
+            ) if not use_spatial_transformer else SpatialTransformer(  # always uses a self-attn
+                            ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                            disable_self_attn=disable_middle_self_attn, use_linear=use_linear_in_transformer,
+                            use_checkpoint=use_checkpoint
+                        ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self._feature_size += ch
+
+        self.output_blocks = nn.ModuleList([])
+        for level, mult in list(enumerate(channel_mult))[::-1]:
+            for i in range(self.num_res_blocks[level] + 1):
+                ich = input_block_chans.pop()
+                layers = [
+                    ResBlock(
+                        ch + ich,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=model_channels * mult,
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = model_channels * mult
+                if ds in attention_resolutions:
+                    if num_head_channels == -1:
+                        dim_head = ch // num_heads
+                    else:
+                        num_heads = ch // num_head_channels
+                        dim_head = num_head_channels
+                    if legacy:
+                        #num_heads = 1
+                        dim_head = ch // num_heads if use_spatial_transformer else num_head_channels
+                    if exists(disable_self_attentions):
+                        disabled_sa = disable_self_attentions[level]
+                    else:
+                        disabled_sa = False
+
+                    if not exists(num_attention_blocks) or i < num_attention_blocks[level]:
+                        layers.append(
+                            AttentionBlock(
+                                ch,
+                                use_checkpoint=use_checkpoint,
+                                num_heads=num_heads_upsample,
+                                num_head_channels=dim_head,
+                                use_new_attention_order=use_new_attention_order,
+                            ) if not use_spatial_transformer else SpatialTransformer(
+                                ch, num_heads, dim_head, depth=transformer_depth, context_dim=context_dim,
+                                disable_self_attn=disabled_sa, use_linear=use_linear_in_transformer,
+                                use_checkpoint=use_checkpoint
+                            )
+                        )
+                if level and i == self.num_res_blocks[level]:
+                    out_ch = ch
+                    layers.append(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            up=True,
+                        )
+                        if resblock_updown
+                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
+                    )
+                    ds //= 2
+                self.output_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+
+        self.out = nn.Sequential(
+            normalization(ch),
+            nn.SiLU(),
+            zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)),
+        )
+        if self.predict_codebook_ids:
+            self.id_predictor = nn.Sequential(
+            normalization(ch),
+            conv_nd(dims, model_channels, n_embed, 1),
+            #nn.LogSoftmax(dim=1)  # change to cross_entropy and produce non-normalized logits
+        )
+
+    def convert_to_fp16(self):
+        """
+        Convert the torso of the model to float16.
+        """
+        self.input_blocks.apply(convert_module_to_f16)
+        self.middle_block.apply(convert_module_to_f16)
+        self.output_blocks.apply(convert_module_to_f16)
+
+    def convert_to_fp32(self):
+        """
+        Convert the torso of the model to float32.
+        """
+        self.input_blocks.apply(convert_module_to_f32)
+        self.middle_block.apply(convert_module_to_f32)
+        self.output_blocks.apply(convert_module_to_f32)
+
+    def forward(self, x, timesteps=None, context=None, y=None,**kwargs):
+        """
+        Apply the model to an input batch.
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :param context: conditioning plugged in via crossattn
+        :param y: an [N] Tensor of labels, if class-conditional.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        assert (y is not None) == (
+            self.num_classes is not None
+        ), "must specify y if and only if the model is class-conditional"
+        hs = []
+        t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False)
+        emb = self.time_embed(t_emb)
+
+        if self.num_classes is not None:
+            assert y.shape[0] == x.shape[0]
+            emb = emb + self.label_emb(y)
+
+        h = x.type(self.dtype)
+        for module in self.input_blocks:
+            h = module(h, emb, context)
+            hs.append(h)
+        h = self.middle_block(h, emb, context)
+        for module in self.output_blocks:
+            h = th.cat([h, hs.pop()], dim=1)
+            h = module(h, emb, context)
+        h = h.type(x.dtype)
+        if self.predict_codebook_ids:
+            return self.id_predictor(h)
+        else:
+            return self.out(h)

iopaint/model/anytext/ldm/modules/diffusionmodules/upscaling.py

@@ -0,0 +1,81 @@
+import torch
+import torch.nn as nn
+import numpy as np
+from functools import partial
+
+from iopaint.model.anytext.ldm.modules.diffusionmodules.util import extract_into_tensor, make_beta_schedule
+from iopaint.model.anytext.ldm.util import default
+
+
+class AbstractLowScaleModel(nn.Module):
+    # for concatenating a downsampled image to the latent representation
+    def __init__(self, noise_schedule_config=None):
+        super(AbstractLowScaleModel, self).__init__()
+        if noise_schedule_config is not None:
+            self.register_schedule(**noise_schedule_config)
+
+    def register_schedule(self, beta_schedule="linear", timesteps=1000,
+                          linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+        betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end,
+                                   cosine_s=cosine_s)
+        alphas = 1. - betas
+        alphas_cumprod = np.cumprod(alphas, axis=0)
+        alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1])
+
+        timesteps, = betas.shape
+        self.num_timesteps = int(timesteps)
+        self.linear_start = linear_start
+        self.linear_end = linear_end
+        assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep'
+
+        to_torch = partial(torch.tensor, dtype=torch.float32)
+
+        self.register_buffer('betas', to_torch(betas))
+        self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
+        self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev))
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod)))
+        self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod)))
+        self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod)))
+        self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod)))
+        self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1)))
+
+    def q_sample(self, x_start, t, noise=None):
+        noise = default(noise, lambda: torch.randn_like(x_start))
+        return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start +
+                extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise)
+
+    def forward(self, x):
+        return x, None
+
+    def decode(self, x):
+        return x
+
+
+class SimpleImageConcat(AbstractLowScaleModel):
+    # no noise level conditioning
+    def __init__(self):
+        super(SimpleImageConcat, self).__init__(noise_schedule_config=None)
+        self.max_noise_level = 0
+
+    def forward(self, x):
+        # fix to constant noise level
+        return x, torch.zeros(x.shape[0], device=x.device).long()
+
+
+class ImageConcatWithNoiseAugmentation(AbstractLowScaleModel):
+    def __init__(self, noise_schedule_config, max_noise_level=1000, to_cuda=False):
+        super().__init__(noise_schedule_config=noise_schedule_config)
+        self.max_noise_level = max_noise_level
+
+    def forward(self, x, noise_level=None):
+        if noise_level is None:
+            noise_level = torch.randint(0, self.max_noise_level, (x.shape[0],), device=x.device).long()
+        else:
+            assert isinstance(noise_level, torch.Tensor)
+        z = self.q_sample(x, noise_level)
+        return z, noise_level
+
+
+

iopaint/model/anytext/ldm/modules/diffusionmodules/util.py

@@ -0,0 +1,271 @@
+# adopted from
+# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py
+# and
+# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py
+# and
+# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py
+#
+# thanks!
+
+
+import os
+import math
+import torch
+import torch.nn as nn
+import numpy as np
+from einops import repeat
+
+from iopaint.model.anytext.ldm.util import instantiate_from_config
+
+
+def make_beta_schedule(schedule, n_timestep, linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
+    if schedule == "linear":
+        betas = (
+                torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
+        )
+
+    elif schedule == "cosine":
+        timesteps = (
+                torch.arange(n_timestep + 1, dtype=torch.float64) / n_timestep + cosine_s
+        )
+        alphas = timesteps / (1 + cosine_s) * np.pi / 2
+        alphas = torch.cos(alphas).pow(2)
+        alphas = alphas / alphas[0]
+        betas = 1 - alphas[1:] / alphas[:-1]
+        betas = np.clip(betas, a_min=0, a_max=0.999)
+
+    elif schedule == "sqrt_linear":
+        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64)
+    elif schedule == "sqrt":
+        betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5
+    else:
+        raise ValueError(f"schedule '{schedule}' unknown.")
+    return betas.numpy()
+
+
+def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True):
+    if ddim_discr_method == 'uniform':
+        c = num_ddpm_timesteps // num_ddim_timesteps
+        ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
+    elif ddim_discr_method == 'quad':
+        ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int)
+    else:
+        raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"')
+
+    # assert ddim_timesteps.shape[0] == num_ddim_timesteps
+    # add one to get the final alpha values right (the ones from first scale to data during sampling)
+    steps_out = ddim_timesteps + 1
+    if verbose:
+        print(f'Selected timesteps for ddim sampler: {steps_out}')
+    return steps_out
+
+
+def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
+    # select alphas for computing the variance schedule
+    alphas = alphacums[ddim_timesteps]
+    alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())
+
+    # according the the formula provided in https://arxiv.org/abs/2010.02502
+    sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
+    if verbose:
+        print(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')
+        print(f'For the chosen value of eta, which is {eta}, '
+              f'this results in the following sigma_t schedule for ddim sampler {sigmas}')
+    return sigmas.to(torch.float32), alphas.to(torch.float32), alphas_prev.astype(np.float32)
+
+
+def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function,
+    which defines the cumulative product of (1-beta) over time from t = [0,1].
+    :param num_diffusion_timesteps: the number of betas to produce.
+    :param alpha_bar: a lambda that takes an argument t from 0 to 1 and
+                      produces the cumulative product of (1-beta) up to that
+                      part of the diffusion process.
+    :param max_beta: the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+    """
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
+    return np.array(betas)
+
+
+def extract_into_tensor(a, t, x_shape):
+    b, *_ = t.shape
+    out = a.gather(-1, t)
+    return out.reshape(b, *((1,) * (len(x_shape) - 1)))
+
+
+def checkpoint(func, inputs, params, flag):
+    """
+    Evaluate a function without caching intermediate activations, allowing for
+    reduced memory at the expense of extra compute in the backward pass.
+    :param func: the function to evaluate.
+    :param inputs: the argument sequence to pass to `func`.
+    :param params: a sequence of parameters `func` depends on but does not
+                   explicitly take as arguments.
+    :param flag: if False, disable gradient checkpointing.
+    """
+    if flag:
+        args = tuple(inputs) + tuple(params)
+        return CheckpointFunction.apply(func, len(inputs), *args)
+    else:
+        return func(*inputs)
+
+
+class CheckpointFunction(torch.autograd.Function):
+    @staticmethod
+    def forward(ctx, run_function, length, *args):
+        ctx.run_function = run_function
+        ctx.input_tensors = list(args[:length])
+        ctx.input_params = list(args[length:])
+        ctx.gpu_autocast_kwargs = {"enabled": torch.is_autocast_enabled(),
+                                   "dtype": torch.get_autocast_gpu_dtype(),
+                                   "cache_enabled": torch.is_autocast_cache_enabled()}
+        with torch.no_grad():
+            output_tensors = ctx.run_function(*ctx.input_tensors)
+        return output_tensors
+
+    @staticmethod
+    def backward(ctx, *output_grads):
+        ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
+        with torch.enable_grad(), \
+                torch.cuda.amp.autocast(**ctx.gpu_autocast_kwargs):
+            # Fixes a bug where the first op in run_function modifies the
+            # Tensor storage in place, which is not allowed for detach()'d
+            # Tensors.
+            shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
+            output_tensors = ctx.run_function(*shallow_copies)
+        input_grads = torch.autograd.grad(
+            output_tensors,
+            ctx.input_tensors + ctx.input_params,
+            output_grads,
+            allow_unused=True,
+        )
+        del ctx.input_tensors
+        del ctx.input_params
+        del output_tensors
+        return (None, None) + input_grads
+
+
+def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
+    """
+    Create sinusoidal timestep embeddings.
+    :param timesteps: a 1-D Tensor of N indices, one per batch element.
+                      These may be fractional.
+    :param dim: the dimension of the output.
+    :param max_period: controls the minimum frequency of the embeddings.
+    :return: an [N x dim] Tensor of positional embeddings.
+    """
+    if not repeat_only:
+        half = dim // 2
+        freqs = torch.exp(
+            -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
+        ).to(device=timesteps.device)
+        args = timesteps[:, None].float() * freqs[None]
+        embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
+        if dim % 2:
+            embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
+    else:
+        embedding = repeat(timesteps, 'b -> b d', d=dim)
+    return embedding
+
+
+def zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+
+def scale_module(module, scale):
+    """
+    Scale the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().mul_(scale)
+    return module
+
+
+def mean_flat(tensor):
+    """
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def normalization(channels):
+    """
+    Make a standard normalization layer.
+    :param channels: number of input channels.
+    :return: an nn.Module for normalization.
+    """
+    return GroupNorm32(32, channels)
+
+
+# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
+class SiLU(nn.Module):
+    def forward(self, x):
+        return x * torch.sigmoid(x)
+
+
+class GroupNorm32(nn.GroupNorm):
+    def forward(self, x):
+        # return super().forward(x.float()).type(x.dtype)
+        return super().forward(x).type(x.dtype)
+
+def conv_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D convolution module.
+    """
+    if dims == 1:
+        return nn.Conv1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.Conv2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.Conv3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def linear(*args, **kwargs):
+    """
+    Create a linear module.
+    """
+    return nn.Linear(*args, **kwargs)
+
+
+def avg_pool_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D average pooling module.
+    """
+    if dims == 1:
+        return nn.AvgPool1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.AvgPool2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.AvgPool3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+class HybridConditioner(nn.Module):
+
+    def __init__(self, c_concat_config, c_crossattn_config):
+        super().__init__()
+        self.concat_conditioner = instantiate_from_config(c_concat_config)
+        self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
+
+    def forward(self, c_concat, c_crossattn):
+        c_concat = self.concat_conditioner(c_concat)
+        c_crossattn = self.crossattn_conditioner(c_crossattn)
+        return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}
+
+
+def noise_like(shape, device, repeat=False):
+    repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
+    noise = lambda: torch.randn(shape, device=device)
+    return repeat_noise() if repeat else noise()

iopaint/model/anytext/ldm/modules/distributions/distributions.py

@@ -0,0 +1,92 @@
+import torch
+import numpy as np
+
+
+class AbstractDistribution:
+    def sample(self):
+        raise NotImplementedError()
+
+    def mode(self):
+        raise NotImplementedError()
+
+
+class DiracDistribution(AbstractDistribution):
+    def __init__(self, value):
+        self.value = value
+
+    def sample(self):
+        return self.value
+
+    def mode(self):
+        return self.value
+
+
+class DiagonalGaussianDistribution(object):
+    def __init__(self, parameters, deterministic=False):
+        self.parameters = parameters
+        self.mean, self.logvar = torch.chunk(parameters, 2, dim=1)
+        self.logvar = torch.clamp(self.logvar, -30.0, 20.0)
+        self.deterministic = deterministic
+        self.std = torch.exp(0.5 * self.logvar)
+        self.var = torch.exp(self.logvar)
+        if self.deterministic:
+            self.var = self.std = torch.zeros_like(self.mean).to(device=self.parameters.device)
+
+    def sample(self):
+        x = self.mean + self.std * torch.randn(self.mean.shape).to(device=self.parameters.device)
+        return x
+
+    def kl(self, other=None):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        else:
+            if other is None:
+                return 0.5 * torch.sum(torch.pow(self.mean, 2)
+                                       + self.var - 1.0 - self.logvar,
+                                       dim=[1, 2, 3])
+            else:
+                return 0.5 * torch.sum(
+                    torch.pow(self.mean - other.mean, 2) / other.var
+                    + self.var / other.var - 1.0 - self.logvar + other.logvar,
+                    dim=[1, 2, 3])
+
+    def nll(self, sample, dims=[1,2,3]):
+        if self.deterministic:
+            return torch.Tensor([0.])
+        logtwopi = np.log(2.0 * np.pi)
+        return 0.5 * torch.sum(
+            logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var,
+            dim=dims)
+
+    def mode(self):
+        return self.mean
+
+
+def normal_kl(mean1, logvar1, mean2, logvar2):
+    """
+    source: https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/losses.py#L12
+    Compute the KL divergence between two gaussians.
+    Shapes are automatically broadcasted, so batches can be compared to
+    scalars, among other use cases.
+    """
+    tensor = None
+    for obj in (mean1, logvar1, mean2, logvar2):
+        if isinstance(obj, torch.Tensor):
+            tensor = obj
+            break
+    assert tensor is not None, "at least one argument must be a Tensor"
+
+    # Force variances to be Tensors. Broadcasting helps convert scalars to
+    # Tensors, but it does not work for torch.exp().
+    logvar1, logvar2 = [
+        x if isinstance(x, torch.Tensor) else torch.tensor(x).to(tensor)
+        for x in (logvar1, logvar2)
+    ]
+
+    return 0.5 * (
+        -1.0
+        + logvar2
+        - logvar1
+        + torch.exp(logvar1 - logvar2)
+        + ((mean1 - mean2) ** 2) * torch.exp(-logvar2)
+    )

iopaint/model/anytext/ldm/modules/ema.py

@@ -0,0 +1,80 @@
+import torch
+from torch import nn
+
+
+class LitEma(nn.Module):
+    def __init__(self, model, decay=0.9999, use_num_upates=True):
+        super().__init__()
+        if decay < 0.0 or decay > 1.0:
+            raise ValueError('Decay must be between 0 and 1')
+
+        self.m_name2s_name = {}
+        self.register_buffer('decay', torch.tensor(decay, dtype=torch.float32))
+        self.register_buffer('num_updates', torch.tensor(0, dtype=torch.int) if use_num_upates
+        else torch.tensor(-1, dtype=torch.int))
+
+        for name, p in model.named_parameters():
+            if p.requires_grad:
+                # remove as '.'-character is not allowed in buffers
+                s_name = name.replace('.', '')
+                self.m_name2s_name.update({name: s_name})
+                self.register_buffer(s_name, p.clone().detach().data)
+
+        self.collected_params = []
+
+    def reset_num_updates(self):
+        del self.num_updates
+        self.register_buffer('num_updates', torch.tensor(0, dtype=torch.int))
+
+    def forward(self, model):
+        decay = self.decay
+
+        if self.num_updates >= 0:
+            self.num_updates += 1
+            decay = min(self.decay, (1 + self.num_updates) / (10 + self.num_updates))
+
+        one_minus_decay = 1.0 - decay
+
+        with torch.no_grad():
+            m_param = dict(model.named_parameters())
+            shadow_params = dict(self.named_buffers())
+
+            for key in m_param:
+                if m_param[key].requires_grad:
+                    sname = self.m_name2s_name[key]
+                    shadow_params[sname] = shadow_params[sname].type_as(m_param[key])
+                    shadow_params[sname].sub_(one_minus_decay * (shadow_params[sname] - m_param[key]))
+                else:
+                    assert not key in self.m_name2s_name
+
+    def copy_to(self, model):
+        m_param = dict(model.named_parameters())
+        shadow_params = dict(self.named_buffers())
+        for key in m_param:
+            if m_param[key].requires_grad:
+                m_param[key].data.copy_(shadow_params[self.m_name2s_name[key]].data)
+            else:
+                assert not key in self.m_name2s_name
+
+    def store(self, parameters):
+        """
+        Save the current parameters for restoring later.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            temporarily stored.
+        """
+        self.collected_params = [param.clone() for param in parameters]
+
+    def restore(self, parameters):
+        """
+        Restore the parameters stored with the `store` method.
+        Useful to validate the model with EMA parameters without affecting the
+        original optimization process. Store the parameters before the
+        `copy_to` method. After validation (or model saving), use this to
+        restore the former parameters.
+        Args:
+          parameters: Iterable of `torch.nn.Parameter`; the parameters to be
+            updated with the stored parameters.
+        """
+        for c_param, param in zip(self.collected_params, parameters):
+            param.data.copy_(c_param.data)

iopaint/model/anytext/ldm/modules/encoders/modules.py

@@ -0,0 +1,384 @@
+import torch
+import torch.nn as nn
+from torch.utils.checkpoint import checkpoint
+
+from transformers import T5Tokenizer, T5EncoderModel, CLIPTokenizer, CLIPTextModel, AutoProcessor, CLIPVisionModelWithProjection
+
+from iopaint.model.anytext.ldm.util import count_params
+
+
+def _expand_mask(mask, dtype, tgt_len=None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+def _build_causal_attention_mask(bsz, seq_len, dtype):
+    # lazily create causal attention mask, with full attention between the vision tokens
+    # pytorch uses additive attention mask; fill with -inf
+    mask = torch.empty(bsz, seq_len, seq_len, dtype=dtype)
+    mask.fill_(torch.tensor(torch.finfo(dtype).min))
+    mask.triu_(1)  # zero out the lower diagonal
+    mask = mask.unsqueeze(1)  # expand mask
+    return mask
+
+class AbstractEncoder(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+    def encode(self, *args, **kwargs):
+        raise NotImplementedError
+
+
+class IdentityEncoder(AbstractEncoder):
+
+    def encode(self, x):
+        return x
+
+
+class ClassEmbedder(nn.Module):
+    def __init__(self, embed_dim, n_classes=1000, key='class', ucg_rate=0.1):
+        super().__init__()
+        self.key = key
+        self.embedding = nn.Embedding(n_classes, embed_dim)
+        self.n_classes = n_classes
+        self.ucg_rate = ucg_rate
+
+    def forward(self, batch, key=None, disable_dropout=False):
+        if key is None:
+            key = self.key
+        # this is for use in crossattn
+        c = batch[key][:, None]
+        if self.ucg_rate > 0. and not disable_dropout:
+            mask = 1. - torch.bernoulli(torch.ones_like(c) * self.ucg_rate)
+            c = mask * c + (1-mask) * torch.ones_like(c)*(self.n_classes-1)
+            c = c.long()
+        c = self.embedding(c)
+        return c
+
+    def get_unconditional_conditioning(self, bs, device="cuda"):
+        uc_class = self.n_classes - 1  # 1000 classes --> 0 ... 999, one extra class for ucg (class 1000)
+        uc = torch.ones((bs,), device=device) * uc_class
+        uc = {self.key: uc}
+        return uc
+
+
+def disabled_train(self, mode=True):
+    """Overwrite model.train with this function to make sure train/eval mode
+    does not change anymore."""
+    return self
+
+
+class FrozenT5Embedder(AbstractEncoder):
+    """Uses the T5 transformer encoder for text"""
+    def __init__(self, version="google/t5-v1_1-large", device="cuda", max_length=77, freeze=True):  # others are google/t5-v1_1-xl and google/t5-v1_1-xxl
+        super().__init__()
+        self.tokenizer = T5Tokenizer.from_pretrained(version)
+        self.transformer = T5EncoderModel.from_pretrained(version)
+        self.device = device
+        self.max_length = max_length   # TODO: typical value?
+        if freeze:
+            self.freeze()
+
+    def freeze(self):
+        self.transformer = self.transformer.eval()
+        #self.train = disabled_train
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)
+        outputs = self.transformer(input_ids=tokens)
+
+        z = outputs.last_hidden_state
+        return z
+
+    def encode(self, text):
+        return self(text)
+
+
+class FrozenCLIPEmbedder(AbstractEncoder):
+    """Uses the CLIP transformer encoder for text (from huggingface)"""
+    LAYERS = [
+        "last",
+        "pooled",
+        "hidden"
+    ]
+    def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77,
+                 freeze=True, layer="last", layer_idx=None):  # clip-vit-base-patch32
+        super().__init__()
+        assert layer in self.LAYERS
+        self.tokenizer = CLIPTokenizer.from_pretrained(version)
+        self.transformer = CLIPTextModel.from_pretrained(version)
+        self.device = device
+        self.max_length = max_length
+        if freeze:
+            self.freeze()
+        self.layer = layer
+        self.layer_idx = layer_idx
+        if layer == "hidden":
+            assert layer_idx is not None
+            assert 0 <= abs(layer_idx) <= 12
+
+    def freeze(self):
+        self.transformer = self.transformer.eval()
+        # self.train = disabled_train
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)
+        outputs = self.transformer(input_ids=tokens, output_hidden_states=self.layer=="hidden")
+        if self.layer == "last":
+            z = outputs.last_hidden_state
+        elif self.layer == "pooled":
+            z = outputs.pooler_output[:, None, :]
+        else:
+            z = outputs.hidden_states[self.layer_idx]
+        return z
+
+    def encode(self, text):
+        return self(text)
+
+
+class FrozenOpenCLIPEmbedder(AbstractEncoder):
+    """
+    Uses the OpenCLIP transformer encoder for text
+    """
+    LAYERS = [
+        # "pooled",
+        "last",
+        "penultimate"
+    ]
+
+    def __init__(self, arch="ViT-H-14", version="laion2b_s32b_b79k", device="cuda", max_length=77,
+                 freeze=True, layer="last"):
+        super().__init__()
+        assert layer in self.LAYERS
+        model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), pretrained=version)
+        del model.visual
+        self.model = model
+
+        self.device = device
+        self.max_length = max_length
+        if freeze:
+            self.freeze()
+        self.layer = layer
+        if self.layer == "last":
+            self.layer_idx = 0
+        elif self.layer == "penultimate":
+            self.layer_idx = 1
+        else:
+            raise NotImplementedError()
+
+    def freeze(self):
+        self.model = self.model.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text):
+        tokens = open_clip.tokenize(text)
+        z = self.encode_with_transformer(tokens.to(self.device))
+        return z
+
+    def encode_with_transformer(self, text):
+        x = self.model.token_embedding(text)  # [batch_size, n_ctx, d_model]
+        x = x + self.model.positional_embedding
+        x = x.permute(1, 0, 2)  # NLD -> LND
+        x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask)
+        x = x.permute(1, 0, 2)  # LND -> NLD
+        x = self.model.ln_final(x)
+        return x
+
+    def text_transformer_forward(self, x: torch.Tensor, attn_mask=None):
+        for i, r in enumerate(self.model.transformer.resblocks):
+            if i == len(self.model.transformer.resblocks) - self.layer_idx:
+                break
+            if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting():
+                x = checkpoint(r, x, attn_mask)
+            else:
+                x = r(x, attn_mask=attn_mask)
+        return x
+
+    def encode(self, text):
+        return self(text)
+
+
+class FrozenCLIPT5Encoder(AbstractEncoder):
+    def __init__(self, clip_version="openai/clip-vit-large-patch14", t5_version="google/t5-v1_1-xl", device="cuda",
+                 clip_max_length=77, t5_max_length=77):
+        super().__init__()
+        self.clip_encoder = FrozenCLIPEmbedder(clip_version, device, max_length=clip_max_length)
+        self.t5_encoder = FrozenT5Embedder(t5_version, device, max_length=t5_max_length)
+        print(f"{self.clip_encoder.__class__.__name__} has {count_params(self.clip_encoder)*1.e-6:.2f} M parameters, "
+              f"{self.t5_encoder.__class__.__name__} comes with {count_params(self.t5_encoder)*1.e-6:.2f} M params.")
+
+    def encode(self, text):
+        return self(text)
+
+    def forward(self, text):
+        clip_z = self.clip_encoder.encode(text)
+        t5_z = self.t5_encoder.encode(text)
+        return [clip_z, t5_z]
+
+
+class FrozenCLIPEmbedderT3(AbstractEncoder):
+    """Uses the CLIP transformer encoder for text (from Hugging Face)"""
+    def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77, freeze=True, use_vision=False):
+        super().__init__()
+        self.tokenizer = CLIPTokenizer.from_pretrained(version)
+        self.transformer = CLIPTextModel.from_pretrained(version)
+        if use_vision:
+            self.vit = CLIPVisionModelWithProjection.from_pretrained(version)
+            self.processor = AutoProcessor.from_pretrained(version)
+        self.device = device
+        self.max_length = max_length
+        if freeze:
+            self.freeze()
+
+        def embedding_forward(
+            self,
+            input_ids=None,
+            position_ids=None,
+            inputs_embeds=None,
+            embedding_manager=None,
+        ):
+            seq_length = input_ids.shape[-1] if input_ids is not None else inputs_embeds.shape[-2]
+            if position_ids is None:
+                position_ids = self.position_ids[:, :seq_length]
+            if inputs_embeds is None:
+                inputs_embeds = self.token_embedding(input_ids)
+            if embedding_manager is not None:
+                inputs_embeds = embedding_manager(input_ids, inputs_embeds)
+            position_embeddings = self.position_embedding(position_ids)
+            embeddings = inputs_embeds + position_embeddings
+            return embeddings
+
+        self.transformer.text_model.embeddings.forward = embedding_forward.__get__(self.transformer.text_model.embeddings)
+
+        def encoder_forward(
+            self,
+            inputs_embeds,
+            attention_mask=None,
+            causal_attention_mask=None,
+            output_attentions=None,
+            output_hidden_states=None,
+            return_dict=None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+            encoder_states = () if output_hidden_states else None
+            all_attentions = () if output_attentions else None
+            hidden_states = inputs_embeds
+            for idx, encoder_layer in enumerate(self.layers):
+                if output_hidden_states:
+                    encoder_states = encoder_states + (hidden_states,)
+                layer_outputs = encoder_layer(
+                    hidden_states,
+                    attention_mask,
+                    causal_attention_mask,
+                    output_attentions=output_attentions,
+                )
+                hidden_states = layer_outputs[0]
+                if output_attentions:
+                    all_attentions = all_attentions + (layer_outputs[1],)
+            if output_hidden_states:
+                encoder_states = encoder_states + (hidden_states,)
+            return hidden_states
+
+        self.transformer.text_model.encoder.forward = encoder_forward.__get__(self.transformer.text_model.encoder)
+
+        def text_encoder_forward(
+            self,
+            input_ids=None,
+            attention_mask=None,
+            position_ids=None,
+            output_attentions=None,
+            output_hidden_states=None,
+            return_dict=None,
+            embedding_manager=None,
+        ):
+            output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+            output_hidden_states = (
+                output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+            )
+            return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+            if input_ids is None:
+                raise ValueError("You have to specify either input_ids")
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+            hidden_states = self.embeddings(input_ids=input_ids, position_ids=position_ids, embedding_manager=embedding_manager)
+            bsz, seq_len = input_shape
+            # CLIP's text model uses causal mask, prepare it here.
+            # https://github.com/openai/CLIP/blob/cfcffb90e69f37bf2ff1e988237a0fbe41f33c04/clip/model.py#L324
+            causal_attention_mask = _build_causal_attention_mask(bsz, seq_len, hidden_states.dtype).to(
+                hidden_states.device
+            )
+            # expand attention_mask
+            if attention_mask is not None:
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                attention_mask = _expand_mask(attention_mask, hidden_states.dtype)
+            last_hidden_state = self.encoder(
+                inputs_embeds=hidden_states,
+                attention_mask=attention_mask,
+                causal_attention_mask=causal_attention_mask,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+            )
+            last_hidden_state = self.final_layer_norm(last_hidden_state)
+            return last_hidden_state
+
+        self.transformer.text_model.forward = text_encoder_forward.__get__(self.transformer.text_model)
+
+        def transformer_forward(
+            self,
+            input_ids=None,
+            attention_mask=None,
+            position_ids=None,
+            output_attentions=None,
+            output_hidden_states=None,
+            return_dict=None,
+            embedding_manager=None,
+        ):
+            return self.text_model(
+                input_ids=input_ids,
+                attention_mask=attention_mask,
+                position_ids=position_ids,
+                output_attentions=output_attentions,
+                output_hidden_states=output_hidden_states,
+                return_dict=return_dict,
+                embedding_manager=embedding_manager
+            )
+
+        self.transformer.forward = transformer_forward.__get__(self.transformer)
+
+    def freeze(self):
+        self.transformer = self.transformer.eval()
+        for param in self.parameters():
+            param.requires_grad = False
+
+    def forward(self, text, **kwargs):
+        batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
+                                        return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
+        tokens = batch_encoding["input_ids"].to(self.device)
+        z = self.transformer(input_ids=tokens, **kwargs)
+        return z
+
+    def encode(self, text, **kwargs):
+        return self(text, **kwargs)

iopaint/model/anytext/ldm/util.py

@@ -0,0 +1,197 @@
+import importlib
+
+import torch
+from torch import optim
+import numpy as np
+
+from inspect import isfunction
+from PIL import Image, ImageDraw, ImageFont
+
+
+def log_txt_as_img(wh, xc, size=10):
+    # wh a tuple of (width, height)
+    # xc a list of captions to plot
+    b = len(xc)
+    txts = list()
+    for bi in range(b):
+        txt = Image.new("RGB", wh, color="white")
+        draw = ImageDraw.Draw(txt)
+        font = ImageFont.truetype('font/Arial_Unicode.ttf', size=size)
+        nc = int(32 * (wh[0] / 256))
+        lines = "\n".join(xc[bi][start:start + nc] for start in range(0, len(xc[bi]), nc))
+
+        try:
+            draw.text((0, 0), lines, fill="black", font=font)
+        except UnicodeEncodeError:
+            print("Cant encode string for logging. Skipping.")
+
+        txt = np.array(txt).transpose(2, 0, 1) / 127.5 - 1.0
+        txts.append(txt)
+    txts = np.stack(txts)
+    txts = torch.tensor(txts)
+    return txts
+
+
+def ismap(x):
+    if not isinstance(x, torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] > 3)
+
+
+def isimage(x):
+    if not isinstance(x,torch.Tensor):
+        return False
+    return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1)
+
+
+def exists(x):
+    return x is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+def mean_flat(tensor):
+    """
+    https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/nn.py#L86
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def count_params(model, verbose=False):
+    total_params = sum(p.numel() for p in model.parameters())
+    if verbose:
+        print(f"{model.__class__.__name__} has {total_params*1.e-6:.2f} M params.")
+    return total_params
+
+
+def instantiate_from_config(config, **kwargs):
+    if "target" not in config:
+        if config == '__is_first_stage__':
+            return None
+        elif config == "__is_unconditional__":
+            return None
+        raise KeyError("Expected key `target` to instantiate.")
+    return get_obj_from_str(config["target"])(**config.get("params", dict()), **kwargs)
+
+
+def get_obj_from_str(string, reload=False):
+    module, cls = string.rsplit(".", 1)
+    if reload:
+        module_imp = importlib.import_module(module)
+        importlib.reload(module_imp)
+    return getattr(importlib.import_module(module, package=None), cls)
+
+
+class AdamWwithEMAandWings(optim.Optimizer):
+    # credit to https://gist.github.com/crowsonkb/65f7265353f403714fce3b2595e0b298
+    def __init__(self, params, lr=1.e-3, betas=(0.9, 0.999), eps=1.e-8,  # TODO: check hyperparameters before using
+                 weight_decay=1.e-2, amsgrad=False, ema_decay=0.9999,   # ema decay to match previous code
+                 ema_power=1., param_names=()):
+        """AdamW that saves EMA versions of the parameters."""
+        if not 0.0 <= lr:
+            raise ValueError("Invalid learning rate: {}".format(lr))
+        if not 0.0 <= eps:
+            raise ValueError("Invalid epsilon value: {}".format(eps))
+        if not 0.0 <= betas[0] < 1.0:
+            raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
+        if not 0.0 <= betas[1] < 1.0:
+            raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
+        if not 0.0 <= weight_decay:
+            raise ValueError("Invalid weight_decay value: {}".format(weight_decay))
+        if not 0.0 <= ema_decay <= 1.0:
+            raise ValueError("Invalid ema_decay value: {}".format(ema_decay))
+        defaults = dict(lr=lr, betas=betas, eps=eps,
+                        weight_decay=weight_decay, amsgrad=amsgrad, ema_decay=ema_decay,
+                        ema_power=ema_power, param_names=param_names)
+        super().__init__(params, defaults)
+
+    def __setstate__(self, state):
+        super().__setstate__(state)
+        for group in self.param_groups:
+            group.setdefault('amsgrad', False)
+
+    @torch.no_grad()
+    def step(self, closure=None):
+        """Performs a single optimization step.
+        Args:
+            closure (callable, optional): A closure that reevaluates the model
+                and returns the loss.
+        """
+        loss = None
+        if closure is not None:
+            with torch.enable_grad():
+                loss = closure()
+
+        for group in self.param_groups:
+            params_with_grad = []
+            grads = []
+            exp_avgs = []
+            exp_avg_sqs = []
+            ema_params_with_grad = []
+            state_sums = []
+            max_exp_avg_sqs = []
+            state_steps = []
+            amsgrad = group['amsgrad']
+            beta1, beta2 = group['betas']
+            ema_decay = group['ema_decay']
+            ema_power = group['ema_power']
+
+            for p in group['params']:
+                if p.grad is None:
+                    continue
+                params_with_grad.append(p)
+                if p.grad.is_sparse:
+                    raise RuntimeError('AdamW does not support sparse gradients')
+                grads.append(p.grad)
+
+                state = self.state[p]
+
+                # State initialization
+                if len(state) == 0:
+                    state['step'] = 0
+                    # Exponential moving average of gradient values
+                    state['exp_avg'] = torch.zeros_like(p, memory_format=torch.preserve_format)
+                    # Exponential moving average of squared gradient values
+                    state['exp_avg_sq'] = torch.zeros_like(p, memory_format=torch.preserve_format)
+                    if amsgrad:
+                        # Maintains max of all exp. moving avg. of sq. grad. values
+                        state['max_exp_avg_sq'] = torch.zeros_like(p, memory_format=torch.preserve_format)
+                    # Exponential moving average of parameter values
+                    state['param_exp_avg'] = p.detach().float().clone()
+
+                exp_avgs.append(state['exp_avg'])
+                exp_avg_sqs.append(state['exp_avg_sq'])
+                ema_params_with_grad.append(state['param_exp_avg'])
+
+                if amsgrad:
+                    max_exp_avg_sqs.append(state['max_exp_avg_sq'])
+
+                # update the steps for each param group update
+                state['step'] += 1
+                # record the step after step update
+                state_steps.append(state['step'])
+
+            optim._functional.adamw(params_with_grad,
+                    grads,
+                    exp_avgs,
+                    exp_avg_sqs,
+                    max_exp_avg_sqs,
+                    state_steps,
+                    amsgrad=amsgrad,
+                    beta1=beta1,
+                    beta2=beta2,
+                    lr=group['lr'],
+                    weight_decay=group['weight_decay'],
+                    eps=group['eps'],
+                    maximize=False)
+
+            cur_ema_decay = min(ema_decay, 1 - state['step'] ** -ema_power)
+            for param, ema_param in zip(params_with_grad, ema_params_with_grad):
+                ema_param.mul_(cur_ema_decay).add_(param.float(), alpha=1 - cur_ema_decay)
+
+        return loss

iopaint/model/anytext/main.py

@@ -0,0 +1,52 @@
+from anytext_pipeline import AnyTextPipeline
+from utils import save_images
+
+seed = 66273235
+# seed_everything(seed)
+
+pipe = AnyTextPipeline(
+    cfg_path="/Users/cwq/code/github/AnyText/anytext/models_yaMl/anytext_sd15.yaml",
+    model_dir="/Users/cwq/.cache/modelscope/hub/damo/cv_anytext_text_generation_editing",
+    # font_path="/Users/cwq/code/github/AnyText/anytext/font/Arial_Unicode.ttf",
+    # font_path="/Users/cwq/code/github/AnyText/anytext/font/SourceHanSansSC-VF.ttf",
+    font_path="/Users/cwq/code/github/AnyText/anytext/font/SourceHanSansSC-Medium.otf",
+    use_fp16=False,
+    device="mps",
+)
+
+img_save_folder = "SaveImages"
+params = {
+    "show_debug": True,
+    "image_count": 2,
+    "ddim_steps": 20,
+}
+
+# # 1. text generation
+# mode = "text-generation"
+# input_data = {
+#     "prompt": 'photo of caramel macchiato coffee on the table, top-down perspective, with "Any" "Text" written on it using cream',
+#     "seed": seed,
+#     "draw_pos": "/Users/cwq/code/github/AnyText/anytext/example_images/gen9.png",
+# }
+# results, rtn_code, rtn_warning, debug_info = pipe(input_data, mode=mode, **params)
+# if rtn_code >= 0:
+#     save_images(results, img_save_folder)
+#     print(f"Done, result images are saved in: {img_save_folder}")
+# if rtn_warning:
+#     print(rtn_warning)
+#
+# exit()
+# 2. text editing
+mode = "text-editing"
+input_data = {
+    "prompt": 'A cake with colorful characters that reads "EVERYDAY"',
+    "seed": seed,
+    "draw_pos": "/Users/cwq/code/github/AnyText/anytext/example_images/edit7.png",
+    "ori_image": "/Users/cwq/code/github/AnyText/anytext/example_images/ref7.jpg",
+}
+results, rtn_code, rtn_warning, debug_info = pipe(input_data, mode=mode, **params)
+if rtn_code >= 0:
+    save_images(results, img_save_folder)
+    print(f"Done, result images are saved in: {img_save_folder}")
+if rtn_warning:
+    print(rtn_warning)

iopaint/model/anytext/ocr_recog/RNN.py

@@ -0,0 +1,210 @@
+from torch import nn
+import torch
+from .RecSVTR import Block
+
+class Swish(nn.Module):
+    def __int__(self):
+        super(Swish, self).__int__()
+
+    def forward(self,x):
+        return x*torch.sigmoid(x)
+
+class Im2Im(nn.Module):
+    def __init__(self, in_channels, **kwargs):
+        super().__init__()
+        self.out_channels = in_channels
+
+    def forward(self, x):
+        return x
+
+class Im2Seq(nn.Module):
+    def __init__(self, in_channels, **kwargs):
+        super().__init__()
+        self.out_channels = in_channels
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        # assert H == 1
+        x = x.reshape(B, C, H * W)
+        x = x.permute((0, 2, 1))
+        return x
+
+class EncoderWithRNN(nn.Module):
+    def __init__(self, in_channels,**kwargs):
+        super(EncoderWithRNN, self).__init__()
+        hidden_size = kwargs.get('hidden_size', 256)
+        self.out_channels = hidden_size * 2
+        self.lstm = nn.LSTM(in_channels, hidden_size, bidirectional=True, num_layers=2,batch_first=True)
+
+    def forward(self, x):
+        self.lstm.flatten_parameters()
+        x, _ = self.lstm(x)
+        return x
+
+class SequenceEncoder(nn.Module):
+    def __init__(self, in_channels, encoder_type='rnn',  **kwargs):
+        super(SequenceEncoder, self).__init__()
+        self.encoder_reshape = Im2Seq(in_channels)
+        self.out_channels = self.encoder_reshape.out_channels
+        self.encoder_type = encoder_type
+        if encoder_type == 'reshape':
+            self.only_reshape = True
+        else:
+            support_encoder_dict = {
+                'reshape': Im2Seq,
+                'rnn': EncoderWithRNN,
+                'svtr': EncoderWithSVTR
+            }
+            assert encoder_type in support_encoder_dict, '{} must in {}'.format(
+                encoder_type, support_encoder_dict.keys())
+
+            self.encoder = support_encoder_dict[encoder_type](
+                self.encoder_reshape.out_channels,**kwargs)
+            self.out_channels = self.encoder.out_channels
+            self.only_reshape = False
+
+    def forward(self, x):
+        if self.encoder_type != 'svtr':
+            x = self.encoder_reshape(x)
+            if not self.only_reshape:
+                x = self.encoder(x)
+            return x
+        else:
+            x = self.encoder(x)
+            x = self.encoder_reshape(x)
+            return x
+
+class ConvBNLayer(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size=3,
+                 stride=1,
+                 padding=0,
+                 bias_attr=False,
+                 groups=1,
+                 act=nn.GELU):
+        super().__init__()
+        self.conv = nn.Conv2d(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            # weight_attr=paddle.ParamAttr(initializer=nn.initializer.KaimingUniform()),
+            bias=bias_attr)
+        self.norm = nn.BatchNorm2d(out_channels)
+        self.act = Swish()
+
+    def forward(self, inputs):
+        out = self.conv(inputs)
+        out = self.norm(out)
+        out = self.act(out)
+        return out
+
+
+class EncoderWithSVTR(nn.Module):
+    def __init__(
+            self,
+            in_channels,
+            dims=64,  # XS
+            depth=2,
+            hidden_dims=120,
+            use_guide=False,
+            num_heads=8,
+            qkv_bias=True,
+            mlp_ratio=2.0,
+            drop_rate=0.1,
+            attn_drop_rate=0.1,
+            drop_path=0.,
+            qk_scale=None):
+        super(EncoderWithSVTR, self).__init__()
+        self.depth = depth
+        self.use_guide = use_guide
+        self.conv1 = ConvBNLayer(
+            in_channels, in_channels // 8, padding=1, act='swish')
+        self.conv2 = ConvBNLayer(
+            in_channels // 8, hidden_dims, kernel_size=1, act='swish')
+
+        self.svtr_block = nn.ModuleList([
+            Block(
+                dim=hidden_dims,
+                num_heads=num_heads,
+                mixer='Global',
+                HW=None,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                act_layer='swish',
+                attn_drop=attn_drop_rate,
+                drop_path=drop_path,
+                norm_layer='nn.LayerNorm',
+                epsilon=1e-05,
+                prenorm=False) for i in range(depth)
+        ])
+        self.norm = nn.LayerNorm(hidden_dims, eps=1e-6)
+        self.conv3 = ConvBNLayer(
+            hidden_dims, in_channels, kernel_size=1, act='swish')
+        # last conv-nxn, the input is concat of input tensor and conv3 output tensor
+        self.conv4 = ConvBNLayer(
+            2 * in_channels, in_channels // 8, padding=1, act='swish')
+
+        self.conv1x1 = ConvBNLayer(
+            in_channels // 8, dims, kernel_size=1, act='swish')
+        self.out_channels = dims
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        # weight initialization
+        if isinstance(m, nn.Conv2d):
+            nn.init.kaiming_normal_(m.weight, mode='fan_out')
+            if m.bias is not None:
+                nn.init.zeros_(m.bias)
+        elif isinstance(m, nn.BatchNorm2d):
+            nn.init.ones_(m.weight)
+            nn.init.zeros_(m.bias)
+        elif isinstance(m, nn.Linear):
+            nn.init.normal_(m.weight, 0, 0.01)
+            if m.bias is not None:
+                nn.init.zeros_(m.bias)
+        elif isinstance(m, nn.ConvTranspose2d):
+            nn.init.kaiming_normal_(m.weight, mode='fan_out')
+            if m.bias is not None:
+                nn.init.zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            nn.init.ones_(m.weight)
+            nn.init.zeros_(m.bias)
+
+    def forward(self, x):
+        # for use guide
+        if self.use_guide:
+            z = x.clone()
+            z.stop_gradient = True
+        else:
+            z = x
+        # for short cut
+        h = z
+        # reduce dim
+        z = self.conv1(z)
+        z = self.conv2(z)
+        # SVTR global block
+        B, C, H, W = z.shape
+        z = z.flatten(2).permute(0, 2, 1)
+
+        for blk in self.svtr_block:
+            z = blk(z)
+
+        z = self.norm(z)
+        # last stage
+        z = z.reshape([-1, H, W, C]).permute(0, 3, 1, 2)
+        z = self.conv3(z)
+        z = torch.cat((h, z), dim=1)
+        z = self.conv1x1(self.conv4(z))
+
+        return z
+
+if __name__=="__main__":
+    svtrRNN = EncoderWithSVTR(56)
+    print(svtrRNN)

iopaint/model/anytext/ocr_recog/RecCTCHead.py

@@ -0,0 +1,48 @@
+from torch import nn
+
+
+class CTCHead(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels=6625,
+                 fc_decay=0.0004,
+                 mid_channels=None,
+                 return_feats=False,
+                 **kwargs):
+        super(CTCHead, self).__init__()
+        if mid_channels is None:
+            self.fc = nn.Linear(
+                in_channels,
+                out_channels,
+                bias=True,)
+        else:
+            self.fc1 = nn.Linear(
+                in_channels,
+                mid_channels,
+                bias=True,
+            )
+            self.fc2 = nn.Linear(
+                mid_channels,
+                out_channels,
+                bias=True,
+            )
+
+        self.out_channels = out_channels
+        self.mid_channels = mid_channels
+        self.return_feats = return_feats
+
+    def forward(self, x, labels=None):
+        if self.mid_channels is None:
+            predicts = self.fc(x)
+        else:
+            x = self.fc1(x)
+            predicts = self.fc2(x)
+
+        if self.return_feats:
+            result = dict()
+            result['ctc'] = predicts
+            result['ctc_neck'] = x
+        else:
+            result = predicts
+
+        return result

iopaint/model/anytext/ocr_recog/RecModel.py

@@ -0,0 +1,45 @@
+from torch import nn
+from .RNN import SequenceEncoder, Im2Seq, Im2Im
+from .RecMv1_enhance import MobileNetV1Enhance
+
+from .RecCTCHead import CTCHead
+
+backbone_dict = {"MobileNetV1Enhance":MobileNetV1Enhance}
+neck_dict = {'SequenceEncoder': SequenceEncoder, 'Im2Seq': Im2Seq,'None':Im2Im}
+head_dict = {'CTCHead':CTCHead}
+
+
+class RecModel(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        assert 'in_channels' in config, 'in_channels must in model config'
+        backbone_type = config.backbone.pop('type')
+        assert backbone_type in backbone_dict, f'backbone.type must in {backbone_dict}'
+        self.backbone = backbone_dict[backbone_type](config.in_channels, **config.backbone)
+
+        neck_type = config.neck.pop('type')
+        assert neck_type in neck_dict, f'neck.type must in {neck_dict}'
+        self.neck = neck_dict[neck_type](self.backbone.out_channels, **config.neck)
+
+        head_type = config.head.pop('type')
+        assert head_type in head_dict, f'head.type must in {head_dict}'
+        self.head = head_dict[head_type](self.neck.out_channels, **config.head)
+
+        self.name = f'RecModel_{backbone_type}_{neck_type}_{head_type}'
+
+    def load_3rd_state_dict(self, _3rd_name, _state):
+        self.backbone.load_3rd_state_dict(_3rd_name, _state)
+        self.neck.load_3rd_state_dict(_3rd_name, _state)
+        self.head.load_3rd_state_dict(_3rd_name, _state)
+
+    def forward(self, x):
+        x = self.backbone(x)
+        x = self.neck(x)
+        x = self.head(x)
+        return x
+
+    def encode(self, x):
+        x = self.backbone(x)
+        x = self.neck(x)
+        x = self.head.ctc_encoder(x)
+        return x

iopaint/model/anytext/ocr_recog/RecMv1_enhance.py

@@ -0,0 +1,232 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from .common import Activation
+
+
+class ConvBNLayer(nn.Module):
+    def __init__(self,
+                 num_channels,
+                 filter_size,
+                 num_filters,
+                 stride,
+                 padding,
+                 channels=None,
+                 num_groups=1,
+                 act='hard_swish'):
+        super(ConvBNLayer, self).__init__()
+        self.act = act
+        self._conv = nn.Conv2d(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=num_groups,
+            bias=False)
+
+        self._batch_norm = nn.BatchNorm2d(
+            num_filters,
+        )
+        if self.act is not None:
+            self._act = Activation(act_type=act, inplace=True)
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        if self.act is not None:
+            y = self._act(y)
+        return y
+
+
+class DepthwiseSeparable(nn.Module):
+    def __init__(self,
+                 num_channels,
+                 num_filters1,
+                 num_filters2,
+                 num_groups,
+                 stride,
+                 scale,
+                 dw_size=3,
+                 padding=1,
+                 use_se=False):
+        super(DepthwiseSeparable, self).__init__()
+        self.use_se = use_se
+        self._depthwise_conv = ConvBNLayer(
+            num_channels=num_channels,
+            num_filters=int(num_filters1 * scale),
+            filter_size=dw_size,
+            stride=stride,
+            padding=padding,
+            num_groups=int(num_groups * scale))
+        if use_se:
+            self._se = SEModule(int(num_filters1 * scale))
+        self._pointwise_conv = ConvBNLayer(
+            num_channels=int(num_filters1 * scale),
+            filter_size=1,
+            num_filters=int(num_filters2 * scale),
+            stride=1,
+            padding=0)
+
+    def forward(self, inputs):
+        y = self._depthwise_conv(inputs)
+        if self.use_se:
+            y = self._se(y)
+        y = self._pointwise_conv(y)
+        return y
+
+
+class MobileNetV1Enhance(nn.Module):
+    def __init__(self,
+                 in_channels=3,
+                 scale=0.5,
+                 last_conv_stride=1,
+                 last_pool_type='max',
+                 **kwargs):
+        super().__init__()
+        self.scale = scale
+        self.block_list = []
+
+        self.conv1 = ConvBNLayer(
+            num_channels=in_channels,
+            filter_size=3,
+            channels=3,
+            num_filters=int(32 * scale),
+            stride=2,
+            padding=1)
+
+        conv2_1 = DepthwiseSeparable(
+            num_channels=int(32 * scale),
+            num_filters1=32,
+            num_filters2=64,
+            num_groups=32,
+            stride=1,
+            scale=scale)
+        self.block_list.append(conv2_1)
+
+        conv2_2 = DepthwiseSeparable(
+            num_channels=int(64 * scale),
+            num_filters1=64,
+            num_filters2=128,
+            num_groups=64,
+            stride=1,
+            scale=scale)
+        self.block_list.append(conv2_2)
+
+        conv3_1 = DepthwiseSeparable(
+            num_channels=int(128 * scale),
+            num_filters1=128,
+            num_filters2=128,
+            num_groups=128,
+            stride=1,
+            scale=scale)
+        self.block_list.append(conv3_1)
+
+        conv3_2 = DepthwiseSeparable(
+            num_channels=int(128 * scale),
+            num_filters1=128,
+            num_filters2=256,
+            num_groups=128,
+            stride=(2, 1),
+            scale=scale)
+        self.block_list.append(conv3_2)
+
+        conv4_1 = DepthwiseSeparable(
+            num_channels=int(256 * scale),
+            num_filters1=256,
+            num_filters2=256,
+            num_groups=256,
+            stride=1,
+            scale=scale)
+        self.block_list.append(conv4_1)
+
+        conv4_2 = DepthwiseSeparable(
+            num_channels=int(256 * scale),
+            num_filters1=256,
+            num_filters2=512,
+            num_groups=256,
+            stride=(2, 1),
+            scale=scale)
+        self.block_list.append(conv4_2)
+
+        for _ in range(5):
+            conv5 = DepthwiseSeparable(
+                num_channels=int(512 * scale),
+                num_filters1=512,
+                num_filters2=512,
+                num_groups=512,
+                stride=1,
+                dw_size=5,
+                padding=2,
+                scale=scale,
+                use_se=False)
+            self.block_list.append(conv5)
+
+        conv5_6 = DepthwiseSeparable(
+            num_channels=int(512 * scale),
+            num_filters1=512,
+            num_filters2=1024,
+            num_groups=512,
+            stride=(2, 1),
+            dw_size=5,
+            padding=2,
+            scale=scale,
+            use_se=True)
+        self.block_list.append(conv5_6)
+
+        conv6 = DepthwiseSeparable(
+            num_channels=int(1024 * scale),
+            num_filters1=1024,
+            num_filters2=1024,
+            num_groups=1024,
+            stride=last_conv_stride,
+            dw_size=5,
+            padding=2,
+            use_se=True,
+            scale=scale)
+        self.block_list.append(conv6)
+
+        self.block_list = nn.Sequential(*self.block_list)
+        if last_pool_type == 'avg':
+            self.pool = nn.AvgPool2d(kernel_size=2, stride=2, padding=0)
+        else:
+            self.pool = nn.MaxPool2d(kernel_size=2, stride=2, padding=0)
+        self.out_channels = int(1024 * scale)
+
+    def forward(self, inputs):
+        y = self.conv1(inputs)
+        y = self.block_list(y)
+        y = self.pool(y)
+        return y
+
+def hardsigmoid(x):
+    return F.relu6(x + 3., inplace=True) / 6.
+
+class SEModule(nn.Module):
+    def __init__(self, channel, reduction=4):
+        super(SEModule, self).__init__()
+        self.avg_pool = nn.AdaptiveAvgPool2d(1)
+        self.conv1 = nn.Conv2d(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            bias=True)
+        self.conv2 = nn.Conv2d(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            bias=True)
+
+    def forward(self, inputs):
+        outputs = self.avg_pool(inputs)
+        outputs = self.conv1(outputs)
+        outputs = F.relu(outputs)
+        outputs = self.conv2(outputs)
+        outputs = hardsigmoid(outputs)
+        x = torch.mul(inputs, outputs)
+
+        return x

iopaint/model/anytext/ocr_recog/RecSVTR.py

@@ -0,0 +1,591 @@
+import torch
+import torch.nn as nn
+import numpy as np
+from torch.nn.init import trunc_normal_, zeros_, ones_
+from torch.nn import functional
+
+
+def drop_path(x, drop_prob=0., training=False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ...
+    """
+    if drop_prob == 0. or not training:
+        return x
+    keep_prob = torch.tensor(1 - drop_prob)
+    shape = (x.size()[0], ) + (1, ) * (x.ndim - 1)
+    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype)
+    random_tensor = torch.floor(random_tensor)  # binarize
+    output = x.divide(keep_prob) * random_tensor
+    return output
+
+
+class Swish(nn.Module):
+    def __int__(self):
+        super(Swish, self).__int__()
+
+    def forward(self,x):
+        return x*torch.sigmoid(x)
+
+
+class ConvBNLayer(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 kernel_size=3,
+                 stride=1,
+                 padding=0,
+                 bias_attr=False,
+                 groups=1,
+                 act=nn.GELU):
+        super().__init__()
+        self.conv = nn.Conv2d(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            # weight_attr=paddle.ParamAttr(initializer=nn.initializer.KaimingUniform()),
+            bias=bias_attr)
+        self.norm = nn.BatchNorm2d(out_channels)
+        self.act = act()
+
+    def forward(self, inputs):
+        out = self.conv(inputs)
+        out = self.norm(out)
+        out = self.act(out)
+        return out
+
+
+class DropPath(nn.Module):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks).
+    """
+
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+
+class Identity(nn.Module):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, input):
+        return input
+
+
+class Mlp(nn.Module):
+    def __init__(self,
+                 in_features,
+                 hidden_features=None,
+                 out_features=None,
+                 act_layer=nn.GELU,
+                 drop=0.):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        if isinstance(act_layer, str):
+            self.act = Swish()
+        else:
+            self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class ConvMixer(nn.Module):
+    def __init__(
+            self,
+            dim,
+            num_heads=8,
+            HW=(8, 25),
+            local_k=(3, 3), ):
+        super().__init__()
+        self.HW = HW
+        self.dim = dim
+        self.local_mixer = nn.Conv2d(
+            dim,
+            dim,
+            local_k,
+            1, (local_k[0] // 2, local_k[1] // 2),
+            groups=num_heads,
+            # weight_attr=ParamAttr(initializer=KaimingNormal())
+        )
+
+    def forward(self, x):
+        h = self.HW[0]
+        w = self.HW[1]
+        x = x.transpose([0, 2, 1]).reshape([0, self.dim, h, w])
+        x = self.local_mixer(x)
+        x = x.flatten(2).transpose([0, 2, 1])
+        return x
+
+
+class Attention(nn.Module):
+    def __init__(self,
+                 dim,
+                 num_heads=8,
+                 mixer='Global',
+                 HW=(8, 25),
+                 local_k=(7, 11),
+                 qkv_bias=False,
+                 qk_scale=None,
+                 attn_drop=0.,
+                 proj_drop=0.):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.HW = HW
+        if HW is not None:
+            H = HW[0]
+            W = HW[1]
+            self.N = H * W
+            self.C = dim
+        if mixer == 'Local' and HW is not None:
+            hk = local_k[0]
+            wk = local_k[1]
+            mask = torch.ones([H * W, H + hk - 1, W + wk - 1])
+            for h in range(0, H):
+                for w in range(0, W):
+                    mask[h * W + w, h:h + hk, w:w + wk] = 0.
+            mask_paddle = mask[:, hk // 2:H + hk // 2, wk // 2:W + wk //
+                               2].flatten(1)
+            mask_inf = torch.full([H * W, H * W],fill_value=float('-inf'))
+            mask = torch.where(mask_paddle < 1, mask_paddle, mask_inf)
+            self.mask = mask[None,None,:]
+            # self.mask = mask.unsqueeze([0, 1])
+        self.mixer = mixer
+
+    def forward(self, x):
+        if self.HW is not None:
+            N = self.N
+            C = self.C
+        else:
+            _, N, C = x.shape
+        qkv = self.qkv(x).reshape((-1, N, 3, self.num_heads, C //self.num_heads)).permute((2, 0, 3, 1, 4))
+        q, k, v = qkv[0] * self.scale, qkv[1], qkv[2]
+
+        attn = (q.matmul(k.permute((0, 1, 3, 2))))
+        if self.mixer == 'Local':
+            attn += self.mask
+        attn = functional.softmax(attn, dim=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).permute((0, 2, 1, 3)).reshape((-1, N, C))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Module):
+    def __init__(self,
+                 dim,
+                 num_heads,
+                 mixer='Global',
+                 local_mixer=(7, 11),
+                 HW=(8, 25),
+                 mlp_ratio=4.,
+                 qkv_bias=False,
+                 qk_scale=None,
+                 drop=0.,
+                 attn_drop=0.,
+                 drop_path=0.,
+                 act_layer=nn.GELU,
+                 norm_layer='nn.LayerNorm',
+                 epsilon=1e-6,
+                 prenorm=True):
+        super().__init__()
+        if isinstance(norm_layer, str):
+            self.norm1 = eval(norm_layer)(dim, eps=epsilon)
+        else:
+            self.norm1 = norm_layer(dim)
+        if mixer == 'Global' or mixer == 'Local':
+
+            self.mixer = Attention(
+                dim,
+                num_heads=num_heads,
+                mixer=mixer,
+                HW=HW,
+                local_k=local_mixer,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                attn_drop=attn_drop,
+                proj_drop=drop)
+        elif mixer == 'Conv':
+            self.mixer = ConvMixer(
+                dim, num_heads=num_heads, HW=HW, local_k=local_mixer)
+        else:
+            raise TypeError("The mixer must be one of [Global, Local, Conv]")
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0. else Identity()
+        if isinstance(norm_layer, str):
+            self.norm2 = eval(norm_layer)(dim, eps=epsilon)
+        else:
+            self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(in_features=dim,
+                       hidden_features=mlp_hidden_dim,
+                       act_layer=act_layer,
+                       drop=drop)
+        self.prenorm = prenorm
+
+    def forward(self, x):
+        if self.prenorm:
+            x = self.norm1(x + self.drop_path(self.mixer(x)))
+            x = self.norm2(x + self.drop_path(self.mlp(x)))
+        else:
+            x = x + self.drop_path(self.mixer(self.norm1(x)))
+            x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class PatchEmbed(nn.Module):
+    """ Image to Patch Embedding
+    """
+
+    def __init__(self,
+                 img_size=(32, 100),
+                 in_channels=3,
+                 embed_dim=768,
+                 sub_num=2):
+        super().__init__()
+        num_patches = (img_size[1] // (2 ** sub_num)) * \
+                      (img_size[0] // (2 ** sub_num))
+        self.img_size = img_size
+        self.num_patches = num_patches
+        self.embed_dim = embed_dim
+        self.norm = None
+        if sub_num == 2:
+            self.proj = nn.Sequential(
+                ConvBNLayer(
+                    in_channels=in_channels,
+                    out_channels=embed_dim // 2,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    act=nn.GELU,
+                    bias_attr=False),
+                ConvBNLayer(
+                    in_channels=embed_dim // 2,
+                    out_channels=embed_dim,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    act=nn.GELU,
+                    bias_attr=False))
+        if sub_num == 3:
+            self.proj = nn.Sequential(
+                ConvBNLayer(
+                    in_channels=in_channels,
+                    out_channels=embed_dim // 4,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    act=nn.GELU,
+                    bias_attr=False),
+                ConvBNLayer(
+                    in_channels=embed_dim // 4,
+                    out_channels=embed_dim // 2,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    act=nn.GELU,
+                    bias_attr=False),
+                ConvBNLayer(
+                    in_channels=embed_dim // 2,
+                    out_channels=embed_dim,
+                    kernel_size=3,
+                    stride=2,
+                    padding=1,
+                    act=nn.GELU,
+                    bias_attr=False))
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        assert H == self.img_size[0] and W == self.img_size[1], \
+            f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+        x = self.proj(x).flatten(2).permute(0, 2, 1)
+        return x
+
+
+class SubSample(nn.Module):
+    def __init__(self,
+                 in_channels,
+                 out_channels,
+                 types='Pool',
+                 stride=(2, 1),
+                 sub_norm='nn.LayerNorm',
+                 act=None):
+        super().__init__()
+        self.types = types
+        if types == 'Pool':
+            self.avgpool = nn.AvgPool2d(
+                kernel_size=(3, 5), stride=stride, padding=(1, 2))
+            self.maxpool = nn.MaxPool2d(
+                kernel_size=(3, 5), stride=stride, padding=(1, 2))
+            self.proj = nn.Linear(in_channels, out_channels)
+        else:
+            self.conv = nn.Conv2d(
+                in_channels,
+                out_channels,
+                kernel_size=3,
+                stride=stride,
+                padding=1,
+                # weight_attr=ParamAttr(initializer=KaimingNormal())
+            )
+        self.norm = eval(sub_norm)(out_channels)
+        if act is not None:
+            self.act = act()
+        else:
+            self.act = None
+
+    def forward(self, x):
+
+        if self.types == 'Pool':
+            x1 = self.avgpool(x)
+            x2 = self.maxpool(x)
+            x = (x1 + x2) * 0.5
+            out = self.proj(x.flatten(2).permute((0, 2, 1)))
+        else:
+            x = self.conv(x)
+            out = x.flatten(2).permute((0, 2, 1))
+        out = self.norm(out)
+        if self.act is not None:
+            out = self.act(out)
+
+        return out
+
+
+class SVTRNet(nn.Module):
+    def __init__(
+            self,
+            img_size=[48, 100],
+            in_channels=3,
+            embed_dim=[64, 128, 256],
+            depth=[3, 6, 3],
+            num_heads=[2, 4, 8],
+            mixer=['Local'] * 6 + ['Global'] *
+            6,  # Local atten, Global atten, Conv
+            local_mixer=[[7, 11], [7, 11], [7, 11]],
+            patch_merging='Conv',  # Conv, Pool, None
+            mlp_ratio=4,
+            qkv_bias=True,
+            qk_scale=None,
+            drop_rate=0.,
+            last_drop=0.1,
+            attn_drop_rate=0.,
+            drop_path_rate=0.1,
+            norm_layer='nn.LayerNorm',
+            sub_norm='nn.LayerNorm',
+            epsilon=1e-6,
+            out_channels=192,
+            out_char_num=25,
+            block_unit='Block',
+            act='nn.GELU',
+            last_stage=True,
+            sub_num=2,
+            prenorm=True,
+            use_lenhead=False,
+            **kwargs):
+        super().__init__()
+        self.img_size = img_size
+        self.embed_dim = embed_dim
+        self.out_channels = out_channels
+        self.prenorm = prenorm
+        patch_merging = None if patch_merging != 'Conv' and patch_merging != 'Pool' else patch_merging
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            in_channels=in_channels,
+            embed_dim=embed_dim[0],
+            sub_num=sub_num)
+        num_patches = self.patch_embed.num_patches
+        self.HW = [img_size[0] // (2**sub_num), img_size[1] // (2**sub_num)]
+        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim[0]))
+        # self.pos_embed = self.create_parameter(
+        #     shape=[1, num_patches, embed_dim[0]], default_initializer=zeros_)
+
+        # self.add_parameter("pos_embed", self.pos_embed)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+        Block_unit = eval(block_unit)
+
+        dpr = np.linspace(0, drop_path_rate, sum(depth))
+        self.blocks1 = nn.ModuleList(
+            [
+            Block_unit(
+                dim=embed_dim[0],
+                num_heads=num_heads[0],
+                mixer=mixer[0:depth[0]][i],
+                HW=self.HW,
+                local_mixer=local_mixer[0],
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                act_layer=eval(act),
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[0:depth[0]][i],
+                norm_layer=norm_layer,
+                epsilon=epsilon,
+                prenorm=prenorm) for i in range(depth[0])
+        ]
+        )
+        if patch_merging is not None:
+            self.sub_sample1 = SubSample(
+                embed_dim[0],
+                embed_dim[1],
+                sub_norm=sub_norm,
+                stride=[2, 1],
+                types=patch_merging)
+            HW = [self.HW[0] // 2, self.HW[1]]
+        else:
+            HW = self.HW
+        self.patch_merging = patch_merging
+        self.blocks2 = nn.ModuleList([
+            Block_unit(
+                dim=embed_dim[1],
+                num_heads=num_heads[1],
+                mixer=mixer[depth[0]:depth[0] + depth[1]][i],
+                HW=HW,
+                local_mixer=local_mixer[1],
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                act_layer=eval(act),
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[depth[0]:depth[0] + depth[1]][i],
+                norm_layer=norm_layer,
+                epsilon=epsilon,
+                prenorm=prenorm) for i in range(depth[1])
+        ])
+        if patch_merging is not None:
+            self.sub_sample2 = SubSample(
+                embed_dim[1],
+                embed_dim[2],
+                sub_norm=sub_norm,
+                stride=[2, 1],
+                types=patch_merging)
+            HW = [self.HW[0] // 4, self.HW[1]]
+        else:
+            HW = self.HW
+        self.blocks3 = nn.ModuleList([
+            Block_unit(
+                dim=embed_dim[2],
+                num_heads=num_heads[2],
+                mixer=mixer[depth[0] + depth[1]:][i],
+                HW=HW,
+                local_mixer=local_mixer[2],
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                act_layer=eval(act),
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[depth[0] + depth[1]:][i],
+                norm_layer=norm_layer,
+                epsilon=epsilon,
+                prenorm=prenorm) for i in range(depth[2])
+        ])
+        self.last_stage = last_stage
+        if last_stage:
+            self.avg_pool = nn.AdaptiveAvgPool2d((1, out_char_num))
+            self.last_conv = nn.Conv2d(
+                in_channels=embed_dim[2],
+                out_channels=self.out_channels,
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                bias=False)
+            self.hardswish = nn.Hardswish()
+            self.dropout = nn.Dropout(p=last_drop)
+        if not prenorm:
+            self.norm = eval(norm_layer)(embed_dim[-1], epsilon=epsilon)
+        self.use_lenhead = use_lenhead
+        if use_lenhead:
+            self.len_conv = nn.Linear(embed_dim[2], self.out_channels)
+            self.hardswish_len = nn.Hardswish()
+            self.dropout_len = nn.Dropout(
+                p=last_drop)
+
+        trunc_normal_(self.pos_embed,std=.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight,std=.02)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        x = self.patch_embed(x)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks1:
+            x = blk(x)
+        if self.patch_merging is not None:
+            x = self.sub_sample1(
+                x.permute([0, 2, 1]).reshape(
+                    [-1, self.embed_dim[0], self.HW[0], self.HW[1]]))
+        for blk in self.blocks2:
+            x = blk(x)
+        if self.patch_merging is not None:
+            x = self.sub_sample2(
+                x.permute([0, 2, 1]).reshape(
+                    [-1, self.embed_dim[1], self.HW[0] // 2, self.HW[1]]))
+        for blk in self.blocks3:
+            x = blk(x)
+        if not self.prenorm:
+            x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        if self.use_lenhead:
+            len_x = self.len_conv(x.mean(1))
+            len_x = self.dropout_len(self.hardswish_len(len_x))
+        if self.last_stage:
+            if self.patch_merging is not None:
+                h = self.HW[0] // 4
+            else:
+                h = self.HW[0]
+            x = self.avg_pool(
+                x.permute([0, 2, 1]).reshape(
+                    [-1, self.embed_dim[2], h, self.HW[1]]))
+            x = self.last_conv(x)
+            x = self.hardswish(x)
+            x = self.dropout(x)
+        if self.use_lenhead:
+            return x, len_x
+        return x
+
+
+if __name__=="__main__":
+    a = torch.rand(1,3,48,100)
+    svtr = SVTRNet()
+
+    out = svtr(a)
+    print(svtr)
+    print(out.size())

iopaint/model/anytext/ocr_recog/common.py

@@ -0,0 +1,74 @@
+
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class Hswish(nn.Module):
+    def __init__(self, inplace=True):
+        super(Hswish, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return x * F.relu6(x + 3., inplace=self.inplace) / 6.
+
+# out = max(0, min(1, slop*x+offset))
+# paddle.fluid.layers.hard_sigmoid(x, slope=0.2, offset=0.5, name=None)
+class Hsigmoid(nn.Module):
+    def __init__(self, inplace=True):
+        super(Hsigmoid, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        # torch: F.relu6(x + 3., inplace=self.inplace) / 6.
+        # paddle: F.relu6(1.2 * x + 3., inplace=self.inplace) / 6.
+        return F.relu6(1.2 * x + 3., inplace=self.inplace) / 6.
+
+class GELU(nn.Module):
+    def __init__(self, inplace=True):
+        super(GELU, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        return torch.nn.functional.gelu(x)
+
+
+class Swish(nn.Module):
+    def __init__(self, inplace=True):
+        super(Swish, self).__init__()
+        self.inplace = inplace
+
+    def forward(self, x):
+        if self.inplace:
+            x.mul_(torch.sigmoid(x))
+            return x
+        else:
+            return x*torch.sigmoid(x)
+
+
+class Activation(nn.Module):
+    def __init__(self, act_type, inplace=True):
+        super(Activation, self).__init__()
+        act_type = act_type.lower()
+        if act_type == 'relu':
+            self.act = nn.ReLU(inplace=inplace)
+        elif act_type == 'relu6':
+            self.act = nn.ReLU6(inplace=inplace)
+        elif act_type == 'sigmoid':
+            raise NotImplementedError
+        elif act_type == 'hard_sigmoid':
+            self.act = Hsigmoid(inplace)
+        elif act_type == 'hard_swish':
+            self.act = Hswish(inplace=inplace)
+        elif act_type == 'leakyrelu':
+            self.act = nn.LeakyReLU(inplace=inplace)
+        elif act_type == 'gelu':
+            self.act = GELU(inplace=inplace)
+        elif act_type == 'swish':
+            self.act = Swish(inplace=inplace)
+        else:
+            raise NotImplementedError
+
+    def forward(self, inputs):
+        return self.act(inputs)

iopaint/model/anytext/ocr_recog/en_dict.txt

@@ -0,0 +1,95 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+\
+]
+^
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+|
+}
+~
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+ 

iopaint/model/anytext/utils.py

@@ -0,0 +1,151 @@
+import os
+import datetime
+import cv2
+import numpy as np
+from PIL import Image, ImageDraw
+
+
+def save_images(img_list, folder):
+    if not os.path.exists(folder):
+        os.makedirs(folder)
+    now = datetime.datetime.now()
+    date_str = now.strftime("%Y-%m-%d")
+    folder_path = os.path.join(folder, date_str)
+    if not os.path.exists(folder_path):
+        os.makedirs(folder_path)
+    time_str = now.strftime("%H_%M_%S")
+    for idx, img in enumerate(img_list):
+        image_number = idx + 1
+        filename = f"{time_str}_{image_number}.jpg"
+        save_path = os.path.join(folder_path, filename)
+        cv2.imwrite(save_path, img[..., ::-1])
+
+
+def check_channels(image):
+    channels = image.shape[2] if len(image.shape) == 3 else 1
+    if channels == 1:
+        image = cv2.cvtColor(image, cv2.COLOR_GRAY2BGR)
+    elif channels > 3:
+        image = image[:, :, :3]
+    return image
+
+
+def resize_image(img, max_length=768):
+    height, width = img.shape[:2]
+    max_dimension = max(height, width)
+
+    if max_dimension > max_length:
+        scale_factor = max_length / max_dimension
+        new_width = int(round(width * scale_factor))
+        new_height = int(round(height * scale_factor))
+        new_size = (new_width, new_height)
+        img = cv2.resize(img, new_size)
+    height, width = img.shape[:2]
+    img = cv2.resize(img, (width - (width % 64), height - (height % 64)))
+    return img
+
+
+def insert_spaces(string, nSpace):
+    if nSpace == 0:
+        return string
+    new_string = ""
+    for char in string:
+        new_string += char + " " * nSpace
+    return new_string[:-nSpace]
+
+
+def draw_glyph(font, text):
+    g_size = 50
+    W, H = (512, 80)
+    new_font = font.font_variant(size=g_size)
+    img = Image.new(mode="1", size=(W, H), color=0)
+    draw = ImageDraw.Draw(img)
+    left, top, right, bottom = new_font.getbbox(text)
+    text_width = max(right - left, 5)
+    text_height = max(bottom - top, 5)
+    ratio = min(W * 0.9 / text_width, H * 0.9 / text_height)
+    new_font = font.font_variant(size=int(g_size * ratio))
+
+    text_width, text_height = new_font.getsize(text)
+    offset_x, offset_y = new_font.getoffset(text)
+    x = (img.width - text_width) // 2
+    y = (img.height - text_height) // 2 - offset_y // 2
+    draw.text((x, y), text, font=new_font, fill="white")
+    img = np.expand_dims(np.array(img), axis=2).astype(np.float64)
+    return img
+
+
+def draw_glyph2(
+    font, text, polygon, vertAng=10, scale=1, width=512, height=512, add_space=True
+):
+    enlarge_polygon = polygon * scale
+    rect = cv2.minAreaRect(enlarge_polygon)
+    box = cv2.boxPoints(rect)
+    box = np.int0(box)
+    w, h = rect[1]
+    angle = rect[2]
+    if angle < -45:
+        angle += 90
+    angle = -angle
+    if w < h:
+        angle += 90
+
+    vert = False
+    if abs(angle) % 90 < vertAng or abs(90 - abs(angle) % 90) % 90 < vertAng:
+        _w = max(box[:, 0]) - min(box[:, 0])
+        _h = max(box[:, 1]) - min(box[:, 1])
+        if _h >= _w:
+            vert = True
+            angle = 0
+
+    img = np.zeros((height * scale, width * scale, 3), np.uint8)
+    img = Image.fromarray(img)
+
+    # infer font size
+    image4ratio = Image.new("RGB", img.size, "white")
+    draw = ImageDraw.Draw(image4ratio)
+    _, _, _tw, _th = draw.textbbox(xy=(0, 0), text=text, font=font)
+    text_w = min(w, h) * (_tw / _th)
+    if text_w <= max(w, h):
+        # add space
+        if len(text) > 1 and not vert and add_space:
+            for i in range(1, 100):
+                text_space = insert_spaces(text, i)
+                _, _, _tw2, _th2 = draw.textbbox(xy=(0, 0), text=text_space, font=font)
+                if min(w, h) * (_tw2 / _th2) > max(w, h):
+                    break
+            text = insert_spaces(text, i - 1)
+        font_size = min(w, h) * 0.80
+    else:
+        shrink = 0.75 if vert else 0.85
+        font_size = min(w, h) / (text_w / max(w, h)) * shrink
+    new_font = font.font_variant(size=int(font_size))
+
+    left, top, right, bottom = new_font.getbbox(text)
+    text_width = right - left
+    text_height = bottom - top
+
+    layer = Image.new("RGBA", img.size, (0, 0, 0, 0))
+    draw = ImageDraw.Draw(layer)
+    if not vert:
+        draw.text(
+            (rect[0][0] - text_width // 2, rect[0][1] - text_height // 2 - top),
+            text,
+            font=new_font,
+            fill=(255, 255, 255, 255),
+        )
+    else:
+        x_s = min(box[:, 0]) + _w // 2 - text_height // 2
+        y_s = min(box[:, 1])
+        for c in text:
+            draw.text((x_s, y_s), c, font=new_font, fill=(255, 255, 255, 255))
+            _, _t, _, _b = new_font.getbbox(c)
+            y_s += _b
+
+    rotated_layer = layer.rotate(angle, expand=1, center=(rect[0][0], rect[0][1]))
+
+    x_offset = int((img.width - rotated_layer.width) / 2)
+    y_offset = int((img.height - rotated_layer.height) / 2)
+    img.paste(rotated_layer, (x_offset, y_offset), rotated_layer)
+    img = np.expand_dims(np.array(img.convert("1")), axis=2).astype(np.float64)
+    return img