IOPaint/lama_cleaner/model/base.py

414 lines
13 KiB
Python
Raw Normal View History

2022-04-15 18:11:51 +02:00
import abc
2022-07-14 10:49:03 +02:00
from typing import Optional
2022-04-15 18:11:51 +02:00
import cv2
import torch
import numpy as np
2022-04-15 18:11:51 +02:00
from loguru import logger
2023-05-13 07:45:27 +02:00
from lama_cleaner.helper import (
boxes_from_mask,
resize_max_size,
pad_img_to_modulo,
switch_mps_device,
)
2023-12-01 03:15:35 +01:00
from lama_cleaner.model.helper.g_diffuser_bot import expand_image
2023-11-15 01:50:35 +01:00
from lama_cleaner.model.utils import get_scheduler
from lama_cleaner.schema import Config, HDStrategy, SDSampler
2022-04-15 18:11:51 +02:00
class InpaintModel:
2023-02-11 06:30:09 +01:00
name = "base"
2022-07-14 10:49:03 +02:00
min_size: Optional[int] = None
2022-04-15 18:11:51 +02:00
pad_mod = 8
2022-07-14 10:49:03 +02:00
pad_to_square = False
2023-12-01 03:15:35 +01:00
is_erase_model = False
2022-04-15 18:11:51 +02:00
2022-09-15 16:21:27 +02:00
def __init__(self, device, **kwargs):
2022-04-15 18:11:51 +02:00
"""
Args:
device:
"""
2023-02-11 06:30:09 +01:00
device = switch_mps_device(self.name, device)
2022-04-15 18:11:51 +02:00
self.device = device
2022-09-15 16:21:27 +02:00
self.init_model(device, **kwargs)
2022-04-15 18:11:51 +02:00
@abc.abstractmethod
2022-09-15 16:21:27 +02:00
def init_model(self, device, **kwargs):
2022-04-15 18:11:51 +02:00
...
2022-04-17 17:31:12 +02:00
@staticmethod
@abc.abstractmethod
def is_downloaded() -> bool:
...
2022-04-15 18:11:51 +02:00
@abc.abstractmethod
def forward(self, image, mask, config: Config):
2022-07-14 10:49:03 +02:00
"""Input images and output images have same size
images: [H, W, C] RGB
2022-09-15 16:21:27 +02:00
masks: [H, W, 1] 255 masks 区域
2022-04-15 18:11:51 +02:00
return: BGR IMAGE
"""
...
2023-11-16 14:12:06 +01:00
@staticmethod
def download():
...
2022-04-15 18:11:51 +02:00
def _pad_forward(self, image, mask, config: Config):
origin_height, origin_width = image.shape[:2]
2022-09-15 16:21:27 +02:00
pad_image = pad_img_to_modulo(
image, mod=self.pad_mod, square=self.pad_to_square, min_size=self.min_size
)
pad_mask = pad_img_to_modulo(
mask, mod=self.pad_mod, square=self.pad_to_square, min_size=self.min_size
)
2022-07-14 10:49:03 +02:00
logger.info(f"final forward pad size: {pad_image.shape}")
result = self.forward(pad_image, pad_mask, config)
2022-04-15 18:11:51 +02:00
result = result[0:origin_height, 0:origin_width, :]
2022-11-25 02:29:20 +01:00
result, image, mask = self.forward_post_process(result, image, mask, config)
2023-11-15 02:10:13 +01:00
if config.sd_prevent_unmasked_area:
mask = mask[:, :, np.newaxis]
result = result * (mask / 255) + image[:, :, ::-1] * (1 - (mask / 255))
2022-04-15 18:11:51 +02:00
return result
2022-11-25 02:29:20 +01:00
def forward_post_process(self, result, image, mask, config):
return result, image, mask
2022-04-15 18:11:51 +02:00
@torch.no_grad()
def __call__(self, image, mask, config: Config):
"""
2022-07-14 10:49:03 +02:00
images: [H, W, C] RGB, not normalized
masks: [H, W]
2022-04-15 18:11:51 +02:00
return: BGR IMAGE
"""
inpaint_result = None
logger.info(f"hd_strategy: {config.hd_strategy}")
if config.hd_strategy == HDStrategy.CROP:
if max(image.shape) > config.hd_strategy_crop_trigger_size:
logger.info(f"Run crop strategy")
boxes = boxes_from_mask(mask)
crop_result = []
for box in boxes:
crop_image, crop_box = self._run_box(image, mask, box, config)
crop_result.append((crop_image, crop_box))
inpaint_result = image[:, :, ::-1]
for crop_image, crop_box in crop_result:
x1, y1, x2, y2 = crop_box
inpaint_result[y1:y2, x1:x2, :] = crop_image
elif config.hd_strategy == HDStrategy.RESIZE:
if max(image.shape) > config.hd_strategy_resize_limit:
origin_size = image.shape[:2]
2022-09-15 16:21:27 +02:00
downsize_image = resize_max_size(
image, size_limit=config.hd_strategy_resize_limit
)
downsize_mask = resize_max_size(
mask, size_limit=config.hd_strategy_resize_limit
)
logger.info(
f"Run resize strategy, origin size: {image.shape} forward size: {downsize_image.shape}"
)
inpaint_result = self._pad_forward(
downsize_image, downsize_mask, config
)
2022-04-15 18:11:51 +02:00
# only paste masked area result
2022-09-15 16:21:27 +02:00
inpaint_result = cv2.resize(
inpaint_result,
(origin_size[1], origin_size[0]),
interpolation=cv2.INTER_CUBIC,
)
2022-09-02 05:08:32 +02:00
original_pixel_indices = mask < 127
2022-09-15 16:21:27 +02:00
inpaint_result[original_pixel_indices] = image[:, :, ::-1][
original_pixel_indices
]
2022-04-15 18:11:51 +02:00
if inpaint_result is None:
inpaint_result = self._pad_forward(image, mask, config)
return inpaint_result
2022-09-04 10:00:42 +02:00
def _crop_box(self, image, mask, box, config: Config):
2022-04-15 18:11:51 +02:00
"""
Args:
image: [H, W, C] RGB
mask: [H, W, 1]
box: [left,top,right,bottom]
Returns:
2022-09-04 10:00:42 +02:00
BGR IMAGE, (l, r, r, b)
2022-04-15 18:11:51 +02:00
"""
box_h = box[3] - box[1]
box_w = box[2] - box[0]
cx = (box[0] + box[2]) // 2
cy = (box[1] + box[3]) // 2
img_h, img_w = image.shape[:2]
w = box_w + config.hd_strategy_crop_margin * 2
h = box_h + config.hd_strategy_crop_margin * 2
2022-08-22 17:23:48 +02:00
_l = cx - w // 2
_r = cx + w // 2
_t = cy - h // 2
_b = cy + h // 2
l = max(_l, 0)
r = min(_r, img_w)
t = max(_t, 0)
b = min(_b, img_h)
2022-09-04 10:00:42 +02:00
# try to get more context when crop around image edge
2022-08-22 17:23:48 +02:00
if _l < 0:
r += abs(_l)
if _r > img_w:
2022-09-15 16:21:27 +02:00
l -= _r - img_w
2022-08-22 17:23:48 +02:00
if _t < 0:
b += abs(_t)
if _b > img_h:
2022-09-15 16:21:27 +02:00
t -= _b - img_h
2022-08-22 17:23:48 +02:00
l = max(l, 0)
r = min(r, img_w)
t = max(t, 0)
b = min(b, img_h)
2022-04-15 18:11:51 +02:00
crop_img = image[t:b, l:r, :]
crop_mask = mask[t:b, l:r]
logger.info(f"box size: ({box_h},{box_w}) crop size: {crop_img.shape}")
2022-09-04 10:00:42 +02:00
return crop_img, crop_mask, [l, t, r, b]
def _calculate_cdf(self, histogram):
cdf = histogram.cumsum()
normalized_cdf = cdf / float(cdf.max())
return normalized_cdf
2022-11-25 02:29:20 +01:00
def _calculate_lookup(self, source_cdf, reference_cdf):
lookup_table = np.zeros(256)
lookup_val = 0
for source_index, source_val in enumerate(source_cdf):
for reference_index, reference_val in enumerate(reference_cdf):
if reference_val >= source_val:
lookup_val = reference_index
break
lookup_table[source_index] = lookup_val
return lookup_table
2022-11-25 02:29:20 +01:00
def _match_histograms(self, source, reference, mask):
transformed_channels = []
for channel in range(source.shape[-1]):
source_channel = source[:, :, channel]
reference_channel = reference[:, :, channel]
2022-11-25 02:29:20 +01:00
# only calculate histograms for non-masked parts
2022-11-25 02:29:20 +01:00
source_histogram, _ = np.histogram(source_channel[mask == 0], 256, [0, 256])
2023-05-13 07:45:27 +02:00
reference_histogram, _ = np.histogram(
reference_channel[mask == 0], 256, [0, 256]
)
2022-11-25 02:29:20 +01:00
source_cdf = self._calculate_cdf(source_histogram)
reference_cdf = self._calculate_cdf(reference_histogram)
2022-11-25 02:29:20 +01:00
lookup = self._calculate_lookup(source_cdf, reference_cdf)
2022-11-25 02:29:20 +01:00
transformed_channels.append(cv2.LUT(source_channel, lookup))
2022-11-25 02:29:20 +01:00
result = cv2.merge(transformed_channels)
result = cv2.convertScaleAbs(result)
2022-11-25 02:29:20 +01:00
return result
2022-12-10 15:06:15 +01:00
def _apply_cropper(self, image, mask, config: Config):
img_h, img_w = image.shape[:2]
l, t, w, h = (
config.croper_x,
config.croper_y,
config.croper_width,
config.croper_height,
)
r = l + w
b = t + h
l = max(l, 0)
r = min(r, img_w)
t = max(t, 0)
b = min(b, img_h)
crop_img = image[t:b, l:r, :]
crop_mask = mask[t:b, l:r]
return crop_img, crop_mask, (l, t, r, b)
2022-09-04 10:00:42 +02:00
def _run_box(self, image, mask, box, config: Config):
"""
Args:
image: [H, W, C] RGB
mask: [H, W, 1]
box: [left,top,right,bottom]
Returns:
BGR IMAGE
"""
crop_img, crop_mask, [l, t, r, b] = self._crop_box(image, mask, box, config)
2022-04-15 18:11:51 +02:00
return self._pad_forward(crop_img, crop_mask, config), [l, t, r, b]
2023-01-27 13:59:22 +01:00
class DiffusionInpaintModel(InpaintModel):
2023-12-01 03:15:35 +01:00
def __init__(self, device, **kwargs):
if kwargs.get("model_id_or_path"):
# 用于自定义 diffusers 模型
self.model_id_or_path = kwargs["model_id_or_path"]
super().__init__(device, **kwargs)
2023-01-27 13:59:22 +01:00
@torch.no_grad()
def __call__(self, image, mask, config: Config):
"""
images: [H, W, C] RGB, not normalized
masks: [H, W]
return: BGR IMAGE
"""
# boxes = boxes_from_mask(mask)
if config.use_croper:
2023-08-30 07:28:31 +02:00
if config.croper_is_outpainting:
inpaint_result = self._do_outpainting(image, config)
else:
crop_img, crop_mask, (l, t, r, b) = self._apply_cropper(
image, mask, config
)
crop_image = self._scaled_pad_forward(crop_img, crop_mask, config)
inpaint_result = image[:, :, ::-1]
inpaint_result[t:b, l:r, :] = crop_image
2023-01-27 13:59:22 +01:00
else:
inpaint_result = self._scaled_pad_forward(image, mask, config)
return inpaint_result
2023-08-30 07:28:31 +02:00
def _do_outpainting(self, image, config: Config):
# cropper 和 image 在同一个坐标系下croper_x/y 可能为负数
# 从 image 中 crop 出 outpainting 区域
image_h, image_w = image.shape[:2]
cropper_l = config.croper_x
cropper_t = config.croper_y
cropper_r = config.croper_x + config.croper_width
cropper_b = config.croper_y + config.croper_height
image_l = 0
image_t = 0
image_r = image_w
image_b = image_h
# 类似求 IOU
l = max(cropper_l, image_l)
t = max(cropper_t, image_t)
r = min(cropper_r, image_r)
b = min(cropper_b, image_b)
assert (
0 <= l < r and 0 <= t < b
), f"cropper and image not overlap, {l},{t},{r},{b}"
cropped_image = image[t:b, l:r, :]
padding_l = max(0, image_l - cropper_l)
padding_t = max(0, image_t - cropper_t)
padding_r = max(0, cropper_r - image_r)
padding_b = max(0, cropper_b - image_b)
zero_padding_count = [padding_l, padding_t, padding_r, padding_b].count(0)
if zero_padding_count not in [0, 3]:
logger.warning(
f"padding count({zero_padding_count}) not 0 or 3, may result in bad edge outpainting"
)
expanded_image, mask_image = expand_image(
cropped_image,
left=padding_l,
top=padding_t,
right=padding_r,
bottom=padding_b,
softness=config.sd_outpainting_softness,
space=config.sd_outpainting_space,
)
# 最终扩大了的 image, BGR
expanded_cropped_result_image = self._scaled_pad_forward(
expanded_image, mask_image, config
)
# RGB -> BGR
outpainting_image = cv2.copyMakeBorder(
image,
left=padding_l,
top=padding_t,
right=padding_r,
bottom=padding_b,
borderType=cv2.BORDER_CONSTANT,
value=0,
)[:, :, ::-1]
# 把 cropped_result_image 贴到 outpainting_image 上,这一步不需要 blend
paste_t = 0 if config.croper_y < 0 else config.croper_y
paste_l = 0 if config.croper_x < 0 else config.croper_x
outpainting_image[
paste_t : paste_t + expanded_cropped_result_image.shape[0],
paste_l : paste_l + expanded_cropped_result_image.shape[1],
:,
] = expanded_cropped_result_image
return outpainting_image
2023-01-27 13:59:22 +01:00
def _scaled_pad_forward(self, image, mask, config: Config):
longer_side_length = int(config.sd_scale * max(image.shape[:2]))
origin_size = image.shape[:2]
downsize_image = resize_max_size(image, size_limit=longer_side_length)
downsize_mask = resize_max_size(mask, size_limit=longer_side_length)
2023-05-13 07:45:27 +02:00
if config.sd_scale != 1:
logger.info(
f"Resize image to do sd inpainting: {image.shape} -> {downsize_image.shape}"
)
2023-01-27 13:59:22 +01:00
inpaint_result = self._pad_forward(downsize_image, downsize_mask, config)
# only paste masked area result
inpaint_result = cv2.resize(
inpaint_result,
(origin_size[1], origin_size[0]),
interpolation=cv2.INTER_CUBIC,
)
2023-08-30 07:28:31 +02:00
# blend result, copy from g_diffuser_bot
# mask_rgb = 1.0 - np_img_grey_to_rgb(mask / 255.0)
# inpaint_result = np.clip(
# inpaint_result * (1.0 - mask_rgb) + image * mask_rgb, 0.0, 255.0
# )
# original_pixel_indices = mask < 127
# inpaint_result[original_pixel_indices] = image[:, :, ::-1][
# original_pixel_indices
# ]
2023-01-27 13:59:22 +01:00
return inpaint_result
2023-11-15 01:50:35 +01:00
def set_scheduler(self, config: Config):
scheduler_config = self.model.scheduler.config
sd_sampler = config.sd_sampler
if config.sd_lcm_lora:
sd_sampler = SDSampler.lcm
scheduler = get_scheduler(sd_sampler, scheduler_config)
self.model.scheduler = scheduler
2023-11-15 01:58:52 +01:00
def forward_post_process(self, result, image, mask, config):
if config.sd_match_histograms:
result = self._match_histograms(result, image[:, :, ::-1], mask)
if config.sd_mask_blur != 0:
k = 2 * config.sd_mask_blur + 1
mask = cv2.GaussianBlur(mask, (k, k), 0)
return result, image, mask