70af4845af
new file: inpaint/__main__.py new file: inpaint/api.py new file: inpaint/batch_processing.py new file: inpaint/benchmark.py new file: inpaint/cli.py new file: inpaint/const.py new file: inpaint/download.py new file: inpaint/file_manager/__init__.py new file: inpaint/file_manager/file_manager.py new file: inpaint/file_manager/storage_backends.py new file: inpaint/file_manager/utils.py new file: inpaint/helper.py new file: inpaint/installer.py new file: inpaint/model/__init__.py new file: inpaint/model/anytext/__init__.py new file: inpaint/model/anytext/anytext_model.py new file: inpaint/model/anytext/anytext_pipeline.py new file: inpaint/model/anytext/anytext_sd15.yaml new file: inpaint/model/anytext/cldm/__init__.py new file: inpaint/model/anytext/cldm/cldm.py new file: inpaint/model/anytext/cldm/ddim_hacked.py new file: inpaint/model/anytext/cldm/embedding_manager.py new file: inpaint/model/anytext/cldm/hack.py new file: inpaint/model/anytext/cldm/model.py new file: inpaint/model/anytext/cldm/recognizer.py new file: inpaint/model/anytext/ldm/__init__.py new file: inpaint/model/anytext/ldm/models/__init__.py new file: inpaint/model/anytext/ldm/models/autoencoder.py new file: inpaint/model/anytext/ldm/models/diffusion/__init__.py new file: inpaint/model/anytext/ldm/models/diffusion/ddim.py new file: inpaint/model/anytext/ldm/models/diffusion/ddpm.py new file: inpaint/model/anytext/ldm/models/diffusion/dpm_solver/__init__.py new file: inpaint/model/anytext/ldm/models/diffusion/dpm_solver/dpm_solver.py new file: inpaint/model/anytext/ldm/models/diffusion/dpm_solver/sampler.py new file: inpaint/model/anytext/ldm/models/diffusion/plms.py new file: inpaint/model/anytext/ldm/models/diffusion/sampling_util.py new file: inpaint/model/anytext/ldm/modules/__init__.py new file: inpaint/model/anytext/ldm/modules/attention.py new file: inpaint/model/anytext/ldm/modules/diffusionmodules/__init__.py new file: inpaint/model/anytext/ldm/modules/diffusionmodules/model.py new file: inpaint/model/anytext/ldm/modules/diffusionmodules/openaimodel.py new file: inpaint/model/anytext/ldm/modules/diffusionmodules/upscaling.py new file: inpaint/model/anytext/ldm/modules/diffusionmodules/util.py new file: inpaint/model/anytext/ldm/modules/distributions/__init__.py new file: inpaint/model/anytext/ldm/modules/distributions/distributions.py new file: inpaint/model/anytext/ldm/modules/ema.py new file: inpaint/model/anytext/ldm/modules/encoders/__init__.py new file: inpaint/model/anytext/ldm/modules/encoders/modules.py new file: inpaint/model/anytext/ldm/util.py new file: inpaint/model/anytext/main.py new file: inpaint/model/anytext/ocr_recog/RNN.py new file: inpaint/model/anytext/ocr_recog/RecCTCHead.py new file: inpaint/model/anytext/ocr_recog/RecModel.py new file: inpaint/model/anytext/ocr_recog/RecMv1_enhance.py new file: inpaint/model/anytext/ocr_recog/RecSVTR.py new file: inpaint/model/anytext/ocr_recog/__init__.py new file: inpaint/model/anytext/ocr_recog/common.py new file: inpaint/model/anytext/ocr_recog/en_dict.txt new file: inpaint/model/anytext/ocr_recog/ppocr_keys_v1.txt new file: inpaint/model/anytext/utils.py new file: inpaint/model/base.py new file: inpaint/model/brushnet/__init__.py new file: inpaint/model/brushnet/brushnet.py new file: inpaint/model/brushnet/brushnet_unet_forward.py new file: inpaint/model/brushnet/brushnet_wrapper.py new file: inpaint/model/brushnet/pipeline_brushnet.py new file: inpaint/model/brushnet/unet_2d_blocks.py new file: inpaint/model/controlnet.py new file: inpaint/model/ddim_sampler.py new file: inpaint/model/fcf.py new file: inpaint/model/helper/__init__.py new file: inpaint/model/helper/controlnet_preprocess.py new file: inpaint/model/helper/cpu_text_encoder.py new file: inpaint/model/helper/g_diffuser_bot.py new file: inpaint/model/instruct_pix2pix.py new file: inpaint/model/kandinsky.py new file: inpaint/model/lama.py new file: inpaint/model/ldm.py new file: inpaint/model/manga.py new file: inpaint/model/mat.py new file: inpaint/model/mi_gan.py new file: inpaint/model/opencv2.py new file: inpaint/model/original_sd_configs/__init__.py new file: inpaint/model/original_sd_configs/sd_xl_base.yaml new file: inpaint/model/original_sd_configs/sd_xl_refiner.yaml new file: inpaint/model/original_sd_configs/v1-inference.yaml new file: inpaint/model/original_sd_configs/v2-inference-v.yaml new file: inpaint/model/paint_by_example.py new file: inpaint/model/plms_sampler.py new file: inpaint/model/power_paint/__init__.py new file: inpaint/model/power_paint/pipeline_powerpaint.py new file: inpaint/model/power_paint/power_paint.py new file: inpaint/model/power_paint/power_paint_v2.py new file: inpaint/model/power_paint/powerpaint_tokenizer.py
194 lines
6.7 KiB
Python
194 lines
6.7 KiB
Python
import torch
|
|
import numpy as np
|
|
from tqdm import tqdm
|
|
|
|
from .utils import make_ddim_timesteps, make_ddim_sampling_parameters, noise_like
|
|
|
|
from loguru import logger
|
|
|
|
|
|
class DDIMSampler(object):
|
|
def __init__(self, model, schedule="linear"):
|
|
super().__init__()
|
|
self.model = model
|
|
self.ddpm_num_timesteps = model.num_timesteps
|
|
self.schedule = schedule
|
|
|
|
def register_buffer(self, name, attr):
|
|
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,
|
|
# array([1])
|
|
num_ddpm_timesteps=self.ddpm_num_timesteps,
|
|
verbose=verbose,
|
|
)
|
|
alphas_cumprod = self.model.alphas_cumprod # torch.Size([1000])
|
|
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(1.0 - 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, steps, conditioning, batch_size, shape):
|
|
self.make_schedule(ddim_num_steps=steps, ddim_eta=0, verbose=False)
|
|
# sampling
|
|
C, H, W = shape
|
|
size = (batch_size, C, H, W)
|
|
|
|
# samples: 1,3,128,128
|
|
return self.ddim_sampling(
|
|
conditioning,
|
|
size,
|
|
quantize_denoised=False,
|
|
ddim_use_original_steps=False,
|
|
noise_dropout=0,
|
|
temperature=1.0,
|
|
)
|
|
|
|
@torch.no_grad()
|
|
def ddim_sampling(
|
|
self,
|
|
cond,
|
|
shape,
|
|
ddim_use_original_steps=False,
|
|
quantize_denoised=False,
|
|
temperature=1.0,
|
|
noise_dropout=0.0,
|
|
):
|
|
device = self.model.betas.device
|
|
b = shape[0]
|
|
img = torch.randn(shape, device=device, dtype=cond.dtype)
|
|
timesteps = (
|
|
self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
|
|
)
|
|
|
|
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]
|
|
logger.info(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)
|
|
|
|
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,
|
|
)
|
|
img, _ = outs
|
|
|
|
return img
|
|
|
|
@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,
|
|
):
|
|
b, *_, device = *x.shape, x.device
|
|
e_t = self.model.apply_model(x, t, c)
|
|
|
|
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
|
|
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
|
|
if quantize_denoised: # 没用
|
|
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
|
|
# 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: # 没用
|
|
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
|
x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
|
|
return x_prev, pred_x0
|