Spaces:
Runtime error
Runtime error
Upload app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,355 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import argparse
|
| 2 |
+
import os
|
| 3 |
+
os.environ['CUDA_HOME'] = '/usr/local/cuda'
|
| 4 |
+
os.environ['PATH'] = os.environ['PATH'] + ':/usr/local/cuda/bin'
|
| 5 |
+
from datetime import datetime
|
| 6 |
+
|
| 7 |
+
import gradio as gr
|
| 8 |
+
import spaces
|
| 9 |
+
import numpy as np
|
| 10 |
+
import torch
|
| 11 |
+
from diffusers.image_processor import VaeImageProcessor
|
| 12 |
+
from huggingface_hub import snapshot_download
|
| 13 |
+
from PIL import Image
|
| 14 |
+
torch.jit.script = lambda f: f
|
| 15 |
+
from model.cloth_masker import AutoMasker, vis_mask
|
| 16 |
+
from model.pipeline import CatVTONPipeline
|
| 17 |
+
from utils import init_weight_dtype, resize_and_crop, resize_and_padding
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
def parse_args():
|
| 21 |
+
parser = argparse.ArgumentParser(description="Simple example of a training script.")
|
| 22 |
+
parser.add_argument(
|
| 23 |
+
"--base_model_path",
|
| 24 |
+
type=str,
|
| 25 |
+
default="booksforcharlie/stable-diffusion-inpainting",
|
| 26 |
+
help=(
|
| 27 |
+
"The path to the base model to use for evaluation. This can be a local path or a model identifier from the Model Hub."
|
| 28 |
+
),
|
| 29 |
+
)
|
| 30 |
+
parser.add_argument(
|
| 31 |
+
"--resume_path",
|
| 32 |
+
type=str,
|
| 33 |
+
default="zhengchong/CatVTON",
|
| 34 |
+
help=(
|
| 35 |
+
"The Path to the checkpoint of trained tryon model."
|
| 36 |
+
),
|
| 37 |
+
)
|
| 38 |
+
parser.add_argument(
|
| 39 |
+
"--output_dir",
|
| 40 |
+
type=str,
|
| 41 |
+
default="resource/demo/output",
|
| 42 |
+
help="The output directory where the model predictions will be written.",
|
| 43 |
+
)
|
| 44 |
+
|
| 45 |
+
parser.add_argument(
|
| 46 |
+
"--width",
|
| 47 |
+
type=int,
|
| 48 |
+
default=768,
|
| 49 |
+
help=(
|
| 50 |
+
"The resolution for input images, all the images in the train/validation dataset will be resized to this"
|
| 51 |
+
" resolution"
|
| 52 |
+
),
|
| 53 |
+
)
|
| 54 |
+
parser.add_argument(
|
| 55 |
+
"--height",
|
| 56 |
+
type=int,
|
| 57 |
+
default=1024,
|
| 58 |
+
help=(
|
| 59 |
+
"The resolution for input images, all the images in the train/validation dataset will be resized to this"
|
| 60 |
+
" resolution"
|
| 61 |
+
),
|
| 62 |
+
)
|
| 63 |
+
parser.add_argument(
|
| 64 |
+
"--repaint",
|
| 65 |
+
action="store_true",
|
| 66 |
+
help="Whether to repaint the result image with the original background."
|
| 67 |
+
)
|
| 68 |
+
parser.add_argument(
|
| 69 |
+
"--allow_tf32",
|
| 70 |
+
action="store_true",
|
| 71 |
+
default=True,
|
| 72 |
+
help=(
|
| 73 |
+
"Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
|
| 74 |
+
" https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
|
| 75 |
+
),
|
| 76 |
+
)
|
| 77 |
+
parser.add_argument(
|
| 78 |
+
"--mixed_precision",
|
| 79 |
+
type=str,
|
| 80 |
+
default="bf16",
|
| 81 |
+
choices=["no", "fp16", "bf16"],
|
| 82 |
+
help=(
|
| 83 |
+
"Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
|
| 84 |
+
" 1.10.and an Nvidia Ampere GPU. Default to the value of accelerate config of the current system or the"
|
| 85 |
+
" flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
|
| 86 |
+
),
|
| 87 |
+
)
|
| 88 |
+
|
| 89 |
+
args = parser.parse_args()
|
| 90 |
+
env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
|
| 91 |
+
if env_local_rank != -1 and env_local_rank != args.local_rank:
|
| 92 |
+
args.local_rank = env_local_rank
|
| 93 |
+
|
| 94 |
+
return args
|
| 95 |
+
|
| 96 |
+
def image_grid(imgs, rows, cols):
|
| 97 |
+
assert len(imgs) == rows * cols
|
| 98 |
+
|
| 99 |
+
w, h = imgs[0].size
|
| 100 |
+
grid = Image.new("RGB", size=(cols * w, rows * h))
|
| 101 |
+
|
| 102 |
+
for i, img in enumerate(imgs):
|
| 103 |
+
grid.paste(img, box=(i % cols * w, i // cols * h))
|
| 104 |
+
return grid
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
args = parse_args()
|
| 108 |
+
repo_path = snapshot_download(repo_id=args.resume_path)
|
| 109 |
+
# Pipeline
|
| 110 |
+
pipeline = CatVTONPipeline(
|
| 111 |
+
base_ckpt=args.base_model_path,
|
| 112 |
+
attn_ckpt=repo_path,
|
| 113 |
+
attn_ckpt_version="mix",
|
| 114 |
+
weight_dtype=init_weight_dtype(args.mixed_precision),
|
| 115 |
+
use_tf32=args.allow_tf32,
|
| 116 |
+
device='cuda'
|
| 117 |
+
)
|
| 118 |
+
# AutoMasker
|
| 119 |
+
mask_processor = VaeImageProcessor(vae_scale_factor=8, do_normalize=False, do_binarize=True, do_convert_grayscale=True)
|
| 120 |
+
automasker = AutoMasker(
|
| 121 |
+
densepose_ckpt=os.path.join(repo_path, "DensePose"),
|
| 122 |
+
schp_ckpt=os.path.join(repo_path, "SCHP"),
|
| 123 |
+
device='cuda',
|
| 124 |
+
)
|
| 125 |
+
|
| 126 |
+
@spaces.GPU(duration=120)
|
| 127 |
+
def submit_function(
|
| 128 |
+
person_image,
|
| 129 |
+
cloth_image,
|
| 130 |
+
cloth_type,
|
| 131 |
+
num_inference_steps,
|
| 132 |
+
guidance_scale,
|
| 133 |
+
seed,
|
| 134 |
+
show_type
|
| 135 |
+
):
|
| 136 |
+
person_image, mask = person_image["background"], person_image["layers"][0]
|
| 137 |
+
mask = Image.open(mask).convert("L")
|
| 138 |
+
if len(np.unique(np.array(mask))) == 1:
|
| 139 |
+
mask = None
|
| 140 |
+
else:
|
| 141 |
+
mask = np.array(mask)
|
| 142 |
+
mask[mask > 0] = 255
|
| 143 |
+
mask = Image.fromarray(mask)
|
| 144 |
+
|
| 145 |
+
tmp_folder = args.output_dir
|
| 146 |
+
date_str = datetime.now().strftime("%Y%m%d%H%M%S")
|
| 147 |
+
result_save_path = os.path.join(tmp_folder, date_str[:8], date_str[8:] + ".png")
|
| 148 |
+
if not os.path.exists(os.path.join(tmp_folder, date_str[:8])):
|
| 149 |
+
os.makedirs(os.path.join(tmp_folder, date_str[:8]))
|
| 150 |
+
|
| 151 |
+
generator = None
|
| 152 |
+
if seed != -1:
|
| 153 |
+
generator = torch.Generator(device='cuda').manual_seed(seed)
|
| 154 |
+
|
| 155 |
+
person_image = Image.open(person_image).convert("RGB")
|
| 156 |
+
cloth_image = Image.open(cloth_image).convert("RGB")
|
| 157 |
+
person_image = resize_and_crop(person_image, (args.width, args.height))
|
| 158 |
+
cloth_image = resize_and_padding(cloth_image, (args.width, args.height))
|
| 159 |
+
|
| 160 |
+
# Process mask
|
| 161 |
+
if mask is not None:
|
| 162 |
+
mask = resize_and_crop(mask, (args.width, args.height))
|
| 163 |
+
else:
|
| 164 |
+
mask = automasker(
|
| 165 |
+
person_image,
|
| 166 |
+
cloth_type
|
| 167 |
+
)['mask']
|
| 168 |
+
mask = mask_processor.blur(mask, blur_factor=9)
|
| 169 |
+
|
| 170 |
+
# Inference
|
| 171 |
+
# try:
|
| 172 |
+
result_image = pipeline(
|
| 173 |
+
image=person_image,
|
| 174 |
+
condition_image=cloth_image,
|
| 175 |
+
mask=mask,
|
| 176 |
+
num_inference_steps=num_inference_steps,
|
| 177 |
+
guidance_scale=guidance_scale,
|
| 178 |
+
generator=generator
|
| 179 |
+
)[0]
|
| 180 |
+
# except Exception as e:
|
| 181 |
+
# raise gr.Error(
|
| 182 |
+
# "An error occurred. Please try again later: {}".format(e)
|
| 183 |
+
# )
|
| 184 |
+
|
| 185 |
+
# Post-process
|
| 186 |
+
masked_person = vis_mask(person_image, mask)
|
| 187 |
+
save_result_image = image_grid([person_image, masked_person, cloth_image, result_image], 1, 4)
|
| 188 |
+
save_result_image.save(result_save_path)
|
| 189 |
+
if show_type == "result only":
|
| 190 |
+
return result_image
|
| 191 |
+
else:
|
| 192 |
+
width, height = person_image.size
|
| 193 |
+
if show_type == "input & result":
|
| 194 |
+
condition_width = width // 2
|
| 195 |
+
conditions = image_grid([person_image, cloth_image], 2, 1)
|
| 196 |
+
else:
|
| 197 |
+
condition_width = width // 3
|
| 198 |
+
conditions = image_grid([person_image, masked_person , cloth_image], 3, 1)
|
| 199 |
+
conditions = conditions.resize((condition_width, height), Image.NEAREST)
|
| 200 |
+
new_result_image = Image.new("RGB", (width + condition_width + 6, height))
|
| 201 |
+
new_result_image.paste(conditions, (0, 0))
|
| 202 |
+
new_result_image.paste(result_image, (condition_width + 6, 0))
|
| 203 |
+
return new_result_image
|
| 204 |
+
|
| 205 |
+
|
| 206 |
+
def person_example_fn(image_path):
|
| 207 |
+
return image_path
|
| 208 |
+
|
| 209 |
+
|
| 210 |
+
# Define the HTML content for the header
|
| 211 |
+
HEADER = """
|
| 212 |
+
<div style="text-align: center;">
|
| 213 |
+
<img src="https://i.ibb.co/9bh36NJ/resource-De-XFIT.png" alt="DeX Logo" style="width: 40%; display: block; margin: 0 auto;">
|
| 214 |
+
<h1 style="color: #101820;"> Virtual Try-On with Diffusion Models </h1>
|
| 215 |
+
</div>
|
| 216 |
+
"""
|
| 217 |
+
|
| 218 |
+
def app_gradio():
|
| 219 |
+
with gr.Blocks(title="DeXFit") as demo:
|
| 220 |
+
gr.Markdown(HEADER)
|
| 221 |
+
with gr.Row():
|
| 222 |
+
with gr.Column(scale=1, min_width=350):
|
| 223 |
+
with gr.Row():
|
| 224 |
+
image_path = gr.Image(
|
| 225 |
+
type="filepath",
|
| 226 |
+
interactive=True,
|
| 227 |
+
visible=False,
|
| 228 |
+
)
|
| 229 |
+
person_image = gr.ImageEditor(
|
| 230 |
+
interactive=True, label="Person Image", type="filepath"
|
| 231 |
+
)
|
| 232 |
+
|
| 233 |
+
with gr.Row():
|
| 234 |
+
with gr.Column(scale=1, min_width=230):
|
| 235 |
+
cloth_image = gr.Image(
|
| 236 |
+
interactive=True, label="Condition Image", type="filepath"
|
| 237 |
+
)
|
| 238 |
+
with gr.Column(scale=1, min_width=120):
|
| 239 |
+
gr.Markdown(
|
| 240 |
+
'<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>'
|
| 241 |
+
)
|
| 242 |
+
cloth_type = gr.Radio(
|
| 243 |
+
label="Try-On Cloth Type",
|
| 244 |
+
choices=["upper", "lower", "overall"],
|
| 245 |
+
value="upper",
|
| 246 |
+
)
|
| 247 |
+
|
| 248 |
+
|
| 249 |
+
submit = gr.Button("Submit")
|
| 250 |
+
gr.Markdown(
|
| 251 |
+
'<center><span style="color: #FF0000">!!! Click only Once, Wait for Delay !!!</span></center>'
|
| 252 |
+
)
|
| 253 |
+
|
| 254 |
+
gr.Markdown(
|
| 255 |
+
'<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>'
|
| 256 |
+
)
|
| 257 |
+
with gr.Accordion("Advanced Options", open=False):
|
| 258 |
+
num_inference_steps = gr.Slider(
|
| 259 |
+
label="Inference Step", minimum=10, maximum=100, step=5, value=50
|
| 260 |
+
)
|
| 261 |
+
# Guidence Scale
|
| 262 |
+
guidance_scale = gr.Slider(
|
| 263 |
+
label="CFG Strenth", minimum=0.0, maximum=7.5, step=0.5, value=2.5
|
| 264 |
+
)
|
| 265 |
+
# Random Seed
|
| 266 |
+
seed = gr.Slider(
|
| 267 |
+
label="Seed", minimum=-1, maximum=10000, step=1, value=1000
|
| 268 |
+
)
|
| 269 |
+
show_type = gr.Radio(
|
| 270 |
+
label="Show Type",
|
| 271 |
+
choices=["result only", "input & result", "input & mask & result"],
|
| 272 |
+
value="input & mask & result",
|
| 273 |
+
)
|
| 274 |
+
|
| 275 |
+
with gr.Column(scale=2, min_width=500):
|
| 276 |
+
result_image = gr.Image(interactive=False, label="Result")
|
| 277 |
+
with gr.Row():
|
| 278 |
+
# Photo Examples
|
| 279 |
+
root_path = "resource/demo/example"
|
| 280 |
+
with gr.Column():
|
| 281 |
+
men_exm = gr.Examples(
|
| 282 |
+
examples=[
|
| 283 |
+
os.path.join(root_path, "person", "men", _)
|
| 284 |
+
for _ in os.listdir(os.path.join(root_path, "person", "men"))
|
| 285 |
+
],
|
| 286 |
+
examples_per_page=4,
|
| 287 |
+
inputs=image_path,
|
| 288 |
+
label="Person Examples ①",
|
| 289 |
+
)
|
| 290 |
+
women_exm = gr.Examples(
|
| 291 |
+
examples=[
|
| 292 |
+
os.path.join(root_path, "person", "women", _)
|
| 293 |
+
for _ in os.listdir(os.path.join(root_path, "person", "women"))
|
| 294 |
+
],
|
| 295 |
+
examples_per_page=4,
|
| 296 |
+
inputs=image_path,
|
| 297 |
+
label="Person Examples ②",
|
| 298 |
+
)
|
| 299 |
+
gr.Markdown(
|
| 300 |
+
'<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>'
|
| 301 |
+
)
|
| 302 |
+
with gr.Column():
|
| 303 |
+
condition_upper_exm = gr.Examples(
|
| 304 |
+
examples=[
|
| 305 |
+
os.path.join(root_path, "condition", "upper", _)
|
| 306 |
+
for _ in os.listdir(os.path.join(root_path, "condition", "upper"))
|
| 307 |
+
],
|
| 308 |
+
examples_per_page=4,
|
| 309 |
+
inputs=cloth_image,
|
| 310 |
+
label="Condition Upper Examples",
|
| 311 |
+
)
|
| 312 |
+
condition_overall_exm = gr.Examples(
|
| 313 |
+
examples=[
|
| 314 |
+
os.path.join(root_path, "condition", "overall", _)
|
| 315 |
+
for _ in os.listdir(os.path.join(root_path, "condition", "overall"))
|
| 316 |
+
],
|
| 317 |
+
examples_per_page=4,
|
| 318 |
+
inputs=cloth_image,
|
| 319 |
+
label="Condition Overall Examples",
|
| 320 |
+
)
|
| 321 |
+
condition_person_exm = gr.Examples(
|
| 322 |
+
examples=[
|
| 323 |
+
os.path.join(root_path, "condition", "person", _)
|
| 324 |
+
for _ in os.listdir(os.path.join(root_path, "condition", "person"))
|
| 325 |
+
],
|
| 326 |
+
examples_per_page=4,
|
| 327 |
+
inputs=cloth_image,
|
| 328 |
+
label="Condition Reference Person Examples",
|
| 329 |
+
)
|
| 330 |
+
gr.Markdown(
|
| 331 |
+
'<span style="color: #808080; font-size: small;">*Condition examples come from the Internet. </span>'
|
| 332 |
+
)
|
| 333 |
+
|
| 334 |
+
image_path.change(
|
| 335 |
+
person_example_fn, inputs=image_path, outputs=person_image
|
| 336 |
+
)
|
| 337 |
+
|
| 338 |
+
submit.click(
|
| 339 |
+
submit_function,
|
| 340 |
+
[
|
| 341 |
+
person_image,
|
| 342 |
+
cloth_image,
|
| 343 |
+
cloth_type,
|
| 344 |
+
num_inference_steps,
|
| 345 |
+
guidance_scale,
|
| 346 |
+
seed,
|
| 347 |
+
show_type,
|
| 348 |
+
],
|
| 349 |
+
result_image,
|
| 350 |
+
)
|
| 351 |
+
demo.queue().launch(share=True, show_error=True)
|
| 352 |
+
|
| 353 |
+
|
| 354 |
+
if __name__ == "__main__":
|
| 355 |
+
app_gradio()
|