Spaces:
Build error
Build error
File size: 24,874 Bytes
699e2a3 9a4d3b8 699e2a3 8a94661 699e2a3 8a94661 699e2a3 8a94661 699e2a3 5150c06 699e2a3 be78530 ecab8cb 5150c06 161799a 5150c06 2afb552 306796b 699e2a3 161799a 898979c 2afb552 898979c d6e08e4 45bccd7 306796b 45bccd7 306796b 59fa012 306796b 898979c 59fa012 161799a 306796b 161799a 306796b 2afb552 306796b 56c2b96 59fa012 45bccd7 59fa012 56c2b96 59fa012 56c2b96 161799a 699e2a3 161799a 306796b 5693b8f 77ec430 161799a 77ec430 161799a 77ec430 161799a 699e2a3 161799a 699e2a3 be78530 699e2a3 be78530 699e2a3 be78530 699e2a3 be78530 699e2a3 be78530 699e2a3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 |
import argparse
import os
os.environ['CUDA_HOME'] = '/usr/local/cuda'
os.environ['PATH'] = os.environ['PATH'] + ':/usr/local/cuda/bin'
from datetime import datetime
import gradio as gr
import spaces
import numpy as np
import torch
from diffusers.image_processor import VaeImageProcessor
from huggingface_hub import snapshot_download
from PIL import Image
torch.jit.script = lambda f: f
from model.cloth_masker import AutoMasker, vis_mask
from model.pipeline import CatVTONPipeline, CatVTONPix2PixPipeline
from model.flux.pipeline_flux_tryon import FluxTryOnPipeline
from utils import init_weight_dtype, resize_and_crop, resize_and_padding
def parse_args():
parser = argparse.ArgumentParser(description="Simple example of a training script.")
parser.add_argument(
"--base_model_path",
type=str,
default="booksforcharlie/stable-diffusion-inpainting",
help=(
"The path to the base model to use for evaluation. This can be a local path or a model identifier from the Model Hub."
),
)
parser.add_argument(
"--p2p_base_model_path",
type=str,
default="timbrooks/instruct-pix2pix",
help=(
"The path to the base model to use for evaluation. This can be a local path or a model identifier from the Model Hub."
),
)
parser.add_argument(
"--resume_path",
type=str,
default="zhengchong/CatVTON",
help=(
"The Path to the checkpoint of trained tryon model."
),
)
parser.add_argument(
"--output_dir",
type=str,
default="resource/demo/output",
help="The output directory where the model predictions will be written.",
)
parser.add_argument(
"--width",
type=int,
default=768,
help=(
"The resolution for input images, all the images in the train/validation dataset will be resized to this"
" resolution"
),
)
parser.add_argument(
"--height",
type=int,
default=1024,
help=(
"The resolution for input images, all the images in the train/validation dataset will be resized to this"
" resolution"
),
)
parser.add_argument(
"--repaint",
action="store_true",
help="Whether to repaint the result image with the original background."
)
parser.add_argument(
"--allow_tf32",
action="store_true",
default=True,
help=(
"Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
" https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
),
)
parser.add_argument(
"--mixed_precision",
type=str,
default="bf16",
choices=["no", "fp16", "bf16"],
help=(
"Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
" 1.10.and an Nvidia Ampere GPU. Default to the value of accelerate config of the current system or the"
" flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
),
)
args = parser.parse_args()
env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
if env_local_rank != -1 and env_local_rank != args.local_rank:
args.local_rank = env_local_rank
return args
def image_grid(imgs, rows, cols):
assert len(imgs) == rows * cols
w, h = imgs[0].size
grid = Image.new("RGB", size=(cols * w, rows * h))
for i, img in enumerate(imgs):
grid.paste(img, box=(i % cols * w, i // cols * h))
return grid
args = parse_args()
# Mask-based CatVTON
catvton_repo = "zhengchong/CatVTON"
repo_path = snapshot_download(repo_id=catvton_repo)
# Pipeline
pipeline = CatVTONPipeline(
base_ckpt=args.base_model_path,
attn_ckpt=repo_path,
attn_ckpt_version="mix",
weight_dtype=init_weight_dtype(args.mixed_precision),
use_tf32=args.allow_tf32,
device='cuda'
)
# AutoMasker
mask_processor = VaeImageProcessor(vae_scale_factor=8, do_normalize=False, do_binarize=True, do_convert_grayscale=True)
automasker = AutoMasker(
densepose_ckpt=os.path.join(repo_path, "DensePose"),
schp_ckpt=os.path.join(repo_path, "SCHP"),
device='cuda',
)
# # Flux-based CatVTON
# access_token = os.getenv("HUGGING_FACE_HUB_TOKEN")
# flux_repo = "black-forest-labs/FLUX.1-Fill-dev"
# pipeline_flux = FluxTryOnPipeline.from_pretrained(flux_repo, use_auth_token=access_token)
# pipeline_flux.load_lora_weights(
# os.path.join(repo_path, "flux-lora"),
# weight_name='pytorch_lora_weights.safetensors'
# )
# pipeline_flux.to("cuda", init_weight_dtype(args.mixed_precision))
@spaces.GPU(duration=120)
def submit_function(
person_image,
cloth_image,
cloth_type,
num_inference_steps,
guidance_scale,
seed,
show_type
):
person_image, mask = person_image["background"], person_image["layers"][0]
mask = Image.open(mask).convert("L")
if len(np.unique(np.array(mask))) == 1:
mask = None
else:
mask = np.array(mask)
mask[mask > 0] = 255
mask = Image.fromarray(mask)
tmp_folder = args.output_dir
date_str = datetime.now().strftime("%Y%m%d%H%M%S")
result_save_path = os.path.join(tmp_folder, date_str[:8], date_str[8:] + ".png")
if not os.path.exists(os.path.join(tmp_folder, date_str[:8])):
os.makedirs(os.path.join(tmp_folder, date_str[:8]))
generator = None
if seed != -1:
generator = torch.Generator(device='cuda').manual_seed(seed)
person_image = Image.open(person_image).convert("RGB")
cloth_image = Image.open(cloth_image).convert("RGB")
person_image = resize_and_crop(person_image, (args.width, args.height))
cloth_image = resize_and_padding(cloth_image, (args.width, args.height))
# Process mask
if mask is not None:
mask = resize_and_crop(mask, (args.width, args.height))
else:
mask = automasker(
person_image,
cloth_type
)['mask']
mask = mask_processor.blur(mask, blur_factor=9)
# Inference
# try:
result_image = pipeline(
image=person_image,
condition_image=cloth_image,
mask=mask,
num_inference_steps=num_inference_steps,
guidance_scale=guidance_scale,
generator=generator
)[0]
# except Exception as e:
# raise gr.Error(
# "An error occurred. Please try again later: {}".format(e)
# )
# Post-process
masked_person = vis_mask(person_image, mask)
save_result_image = image_grid([person_image, masked_person, cloth_image, result_image], 1, 4)
save_result_image.save(result_save_path)
if show_type == "result only":
return result_image
else:
width, height = person_image.size
if show_type == "input & result":
condition_width = width // 2
conditions = image_grid([person_image, cloth_image], 2, 1)
else:
condition_width = width // 3
conditions = image_grid([person_image, masked_person , cloth_image], 3, 1)
conditions = conditions.resize((condition_width, height), Image.NEAREST)
new_result_image = Image.new("RGB", (width + condition_width + 5, height))
new_result_image.paste(conditions, (0, 0))
new_result_image.paste(result_image, (condition_width + 5, 0))
return new_result_image
# @spaces.GPU(duration=120)
# def submit_function_flux(
# person_image,
# cloth_image,
# cloth_type,
# num_inference_steps,
# guidance_scale,
# seed,
# show_type
# ):
# # Process image editor input
# person_image, mask = person_image["background"], person_image["layers"][0]
# mask = Image.open(mask).convert("L")
# if len(np.unique(np.array(mask))) == 1:
# mask = None
# else:
# mask = np.array(mask)
# mask[mask > 0] = 255
# mask = Image.fromarray(mask)
# # Set random seed
# generator = None
# if seed != -1:
# generator = torch.Generator(device='cuda').manual_seed(seed)
# # Process input images
# person_image = Image.open(person_image).convert("RGB")
# cloth_image = Image.open(cloth_image).convert("RGB")
# # Adjust image sizes
# person_image = resize_and_crop(person_image, (args.width, args.height))
# cloth_image = resize_and_padding(cloth_image, (args.width, args.height))
# # Process mask
# if mask is not None:
# mask = resize_and_crop(mask, (args.width, args.height))
# else:
# mask = automasker(
# person_image,
# cloth_type
# )['mask']
# mask = mask_processor.blur(mask, blur_factor=9)
# # Inference
# result_image = pipeline_flux(
# image=person_image,
# condition_image=cloth_image,
# mask_image=mask,
# width=args.width,
# height=args.height,
# num_inference_steps=num_inference_steps,
# guidance_scale=guidance_scale,
# generator=generator
# ).images[0]
# # Post-processing
# masked_person = vis_mask(person_image, mask)
# # Return result based on show type
# if show_type == "result only":
# return result_image
# else:
# width, height = person_image.size
# if show_type == "input & result":
# condition_width = width // 2
# conditions = image_grid([person_image, cloth_image], 2, 1)
# else:
# condition_width = width // 3
# conditions = image_grid([person_image, masked_person, cloth_image], 3, 1)
# conditions = conditions.resize((condition_width, height), Image.NEAREST)
# new_result_image = Image.new("RGB", (width + condition_width + 5, height))
# new_result_image.paste(conditions, (0, 0))
# new_result_image.paste(result_image, (condition_width + 5, 0))
# return new_result_image
def person_example_fn(image_path):
return image_path
HEADER = ""
def app_gradio():
with gr.Blocks(title="CatVTON") as demo:
gr.Markdown(HEADER)
with gr.Tab("Virtual Try on"):
with gr.Row():
# define root_path
root_path = "resource/demo/example"
# First column ==============================
with gr.Column(scale=1, min_width=350):
# Person image
image_path = gr.Image(
type="filepath",
interactive=True,
visible=False,
)
person_image = gr.ImageEditor(
interactive=True, label="Person Image", type="filepath"
)
# Mask instruction
with gr.Row():
with gr.Column(scale = 2, min_width=80):
gr.Markdown(
'<span style="color: #808080; font-size: small;">NOTE: The model image must fully show the body parts in the area where you want to try on the clothes <br> Two ways to provide Mask:<br>1. Upload the person image and use the `🖌️` above to draw the Mask (higher priority)<br>2. Select the `Try-On Cloth Type` to generate automatically </span>'
)
with gr.Column(scale = 1, min_width=80):
cloth_type = gr.Radio(
label="Try-On Cloth Type",
choices=["upper", "lower", "overall"],
value="upper",
)
# Model column examples
# Men examples
men_exm = gr.Examples(
examples=[
os.path.join(root_path, "person", "men", _)
for _ in os.listdir(os.path.join(root_path, "person", "men"))
],
examples_per_page=4,
inputs=image_path,
label="Person Examples ①",
)
# Women examples
women_exm = gr.Examples(
examples=[
os.path.join(root_path, "person", "women", _)
for _ in os.listdir(os.path.join(root_path, "person", "women"))
],
examples_per_page=4,
inputs=image_path,
label="Person Examples ②",
)
# Markdown: component display text in Gradio
gr.Markdown(
'<span style="color: #808080; font-size: small;">*Person examples come from the demos of <a href="https://huggingface.co/spaces/levihsu/OOTDiffusion">OOTDiffusion</a> and <a href="https://www.outfitanyone.org">OutfitAnyone</a>. </span>'
)
# Second column ==========================================
with gr.Column(scale=1, min_width=350):
# Clothes image
cloth_image = gr.Image(
interactive=True, label="Clothes Image", type="filepath"
)
# Clothes column examples
# Upper clothes examples
condition_upper_exm = gr.Examples(
examples=[
os.path.join(root_path, "condition", "upper", _)
for _ in os.listdir(os.path.join(root_path, "condition", "upper"))
],
examples_per_page=4,
inputs=cloth_image,
label="Upper clothes",
)
# Lower clothes examples
condition_upper_exm = gr.Examples(
examples=[
os.path.join(root_path, "condition", "lower", _)
for _ in os.listdir(os.path.join(root_path, "condition", "lower"))
],
examples_per_page=4,
inputs=cloth_image,
label="Lower clothes",
)
# Full-body clothes examples
condition_overall_exm = gr.Examples(
examples=[
os.path.join(root_path, "condition", "overall", _)
for _ in os.listdir(os.path.join(root_path, "condition", "overall"))
],
examples_per_page=4,
inputs=cloth_image,
label="Full-body clothes",
)
# Below ===============================================================
with gr.Row():
with gr.Column():
# Result pennal
result_image = gr.Image(interactive=False, label="Result")
# Submit button
submit = gr.Button("Submit")
gr.Markdown(
'<center><span style="color: #FF0000">!!! Click only Once, Wait for Delay !!!</span></center>'
)
# Advance options setting
gr.Markdown(
'<span style="color: #808080; font-size: small;">Advanced options can adjust details:<br>1. `Inference Step` may enhance details;<br>2. `CFG` is highly correlated with saturation;<br>3. `Random seed` may improve pseudo-shadow.</span>'
)
with gr.Accordion("Advanced Options", open=False):
num_inference_steps = gr.Slider(
label="Inference Step", minimum=10, maximum=100, step=5, value=50
)
# Guidence Scale
guidance_scale = gr.Slider(
label="CFG Strenth", minimum=0.0, maximum=7.5, step=0.5, value=2.5
)
# Random Seed
seed = gr.Slider(
label="Seed", minimum=-1, maximum=10000, step=1, value=42
)
show_type = gr.Radio(
label="Show Type",
choices=["result only", "input & result", "input & mask & result"],
value="result only",
)
# event listener for changes to the image_path input component. Whenever the value of image_path changes (e.g., when a new image is uploaded or selected)
image_path.change(
person_example_fn, inputs=image_path, outputs=person_image
)
# when submit button clicked
submit.click(
submit_function,
[
person_image,
cloth_image,
cloth_type,
num_inference_steps,
guidance_scale,
seed,
show_type,
],
result_image,
)
# with gr.Tab("Mask-based & Flux.1 Fill Dev"):
# with gr.Row():
# with gr.Column(scale=1, min_width=350):
# with gr.Row():
# image_path_flux = gr.Image(
# type="filepath",
# interactive=True,
# visible=False,
# )
# person_image_flux = gr.ImageEditor(
# interactive=True, label="Person Image", type="filepath"
# )
# with gr.Row():
# with gr.Column(scale=1, min_width=230):
# cloth_image_flux = gr.Image(
# interactive=True, label="Condition Image", type="filepath"
# )
# with gr.Column(scale=1, min_width=120):
# gr.Markdown(
# '<span style="color: #808080; font-size: small;">Two ways to provide Mask:<br>1. Upload the person image and use the `🖌️` above to draw the Mask (higher priority)<br>2. Select the `Try-On Cloth Type` to generate automatically </span>'
# )
# cloth_type = gr.Radio(
# label="Try-On Cloth Type",
# choices=["upper", "lower", "overall"],
# value="upper",
# )
# submit_flux = gr.Button("Submit")
# gr.Markdown(
# '<center><span style="color: #FF0000">!!! Click only Once, Wait for Delay !!!</span></center>'
# )
# with gr.Accordion("Advanced Options", open=False):
# num_inference_steps_flux = gr.Slider(
# label="Inference Step", minimum=10, maximum=100, step=5, value=50
# )
# # Guidence Scale
# guidance_scale_flux = gr.Slider(
# label="CFG Strenth", minimum=0.0, maximum=50, step=0.5, value=30
# )
# # Random Seed
# seed_flux = gr.Slider(
# label="Seed", minimum=-1, maximum=10000, step=1, value=42
# )
# show_type = gr.Radio(
# label="Show Type",
# choices=["result only", "input & result", "input & mask & result"],
# value="input & mask & result",
# )
# with gr.Column(scale=2, min_width=500):
# result_image_flux = gr.Image(interactive=False, label="Result")
# with gr.Row():
# # Photo Examples
# root_path = "resource/demo/example"
# with gr.Column():
# gr.Examples(
# examples=[
# os.path.join(root_path, "person", "men", _)
# for _ in os.listdir(os.path.join(root_path, "person", "men"))
# ],
# examples_per_page=4,
# inputs=image_path_flux,
# label="Person Examples ①",
# )
# gr.Examples(
# examples=[
# os.path.join(root_path, "person", "women", _)
# for _ in os.listdir(os.path.join(root_path, "person", "women"))
# ],
# examples_per_page=4,
# inputs=image_path_flux,
# label="Person Examples ②",
# )
# gr.Markdown(
# '<span style="color: #808080; font-size: small;">*Person examples come from the demos of <a href="https://huggingface.co/spaces/levihsu/OOTDiffusion">OOTDiffusion</a> and <a href="https://www.outfitanyone.org">OutfitAnyone</a>. </span>'
# )
# with gr.Column():
# gr.Examples(
# examples=[
# os.path.join(root_path, "condition", "upper", _)
# for _ in os.listdir(os.path.join(root_path, "condition", "upper"))
# ],
# examples_per_page=4,
# inputs=cloth_image_flux,
# label="Condition Upper Examples",
# )
# gr.Examples(
# examples=[
# os.path.join(root_path, "condition", "overall", _)
# for _ in os.listdir(os.path.join(root_path, "condition", "overall"))
# ],
# examples_per_page=4,
# inputs=cloth_image_flux,
# label="Condition Overall Examples",
# )
# condition_person_exm = gr.Examples(
# examples=[
# os.path.join(root_path, "condition", "person", _)
# for _ in os.listdir(os.path.join(root_path, "condition", "person"))
# ],
# examples_per_page=4,
# inputs=cloth_image_flux,
# label="Condition Reference Person Examples",
# )
# gr.Markdown(
# '<span style="color: #808080; font-size: small;">*Condition examples come from the Internet. </span>'
# )
# image_path_flux.change(
# person_example_fn, inputs=image_path_flux, outputs=person_image_flux
# )
# submit_flux.click(
# submit_function_flux,
# [person_image_flux, cloth_image_flux, cloth_type, num_inference_steps_flux, guidance_scale_flux, seed_flux, show_type],
# result_image_flux,
# )
demo.queue().launch(share=True, show_error=True)
if __name__ == "__main__":
app_gradio()
|