Virtual try-on has developed into a game-changing tool that allows people to play with style without ever having to put on clothes in person. However, producing outcomes with great fidelity and consistency in details is a challenge for many of the current methods. Although diffusion models have proven to be capable of producing images that are both high-quality and lifelike, they still struggle to maintain control and consistency in conditional generation scenarios such as virtual try-ons.
For the given code, models are fixed and cannot be uploaded or modified; it only support users uploading their own garments. For a one-piece dress or coat, you only need to upload the image to the ‘top garment’ section and leave the ‘lower garment’ section empty. For building our own outfit anyone model, we need to create a gradio space and upload the files shown below. In models, garments and examples folders sample images needs to be added.
Files
The main code involved in this is as follows.
#app.py
import os
import cv2
import numpy as np
import json
import random
from PIL import Image, ImageDraw, ImageFont
import asyncio
import requests
import base64
import gradio as gr
# from IPython import embed
machine_number = 0
model = os.path.join(os.path.dirname(__file__), "models/eva/Eva_0.png")
MODEL_MAP = {
"AI Model Rouyan_0": ''models/rouyan_new/Rouyan_0.png'',
"AI Model Rouyan_1": ''models/rouyan_new/Rouyan_1.png'',
"AI Model Rouyan_2": ''models/rouyan_new/Rouyan_2.png'',
"AI Model Eva_0": ''models/eva/Eva_0.png'',
"AI Model Eva_1": ''models/eva/Eva_1.png'',
"AI Model Simon_0": ''models/simon_online/Simon_0.png'',
"AI Model Simon_1": ''models/simon_online/Simon_1.png'',
"AI Model Xuanxuan_0": ''models/xiaoxuan_online/Xuanxuan_0.png'',
"AI Model Xuanxuan_1": ''models/xiaoxuan_online/Xuanxuan_1.png'',
"AI Model Xuanxuan_2": ''models/xiaoxuan_online/Xuanxuan_2.png'',
"AI Model Yaqi_0": ''models/yaqi/Yaqi_0.png'',
"AI Model Yaqi_1": ''models/yaqi/Yaqi_1.png'',
"AI Model Yaqi_2": ''models/yaqi/Yaqi_2.png'',
"AI Model Yaqi_3": ''models/yaqi/Yaqi_3.png'',
"AI Model Yifeng_0": ''models/yifeng_online/Yifeng_0.png'',
"AI Model Yifeng_1": ''models/yifeng_online/Yifeng_1.png'',
"AI Model Yifeng_2": ''models/yifeng_online/Yifeng_2.png'',
"AI Model Yifeng_3": ''models/yifeng_online/Yifeng_3.png'',
}
def add_waterprint(img):
h, w, _ = img.shape
img = cv2.putText(img, ''Powered by OutfitAnyone'', (int(0.3*w), h-20), cv2.FONT_HERSHEY_PLAIN, 2, (128, 128, 128), 2, cv2.LINE_AA)
return img
def get_tryon_result(model_name, garment1, garment2, seed=1234):
# model_name = "AI Model " + model_name.split("\\")[-1].split(".")[0] # windows
model_name = "AI Model " + model_name.split("/")[-1].split(".")[0] # linux
print(model_name)
encoded_garment1 = cv2.imencode(''.jpg'', garment1)[1].tobytes()
encoded_garment1 = base64.b64encode(encoded_garment1).decode(''utf-8'')
if garment2 is not None:
encoded_garment2 = cv2.imencode(''.jpg'', garment2)[1].tobytes()
encoded_garment2 = base64.b64encode(encoded_garment2).decode(''utf-8'')
else:
encoded_garment2 = ''
url = os.environ[''OA_IP_ADDRESS'']
headers = {''Content-Type'': ''application/json''}
seed = random.randint(0, 1222222222)
data = {
"garment1": encoded_garment1,
"garment2": encoded_garment2,
"model_name": model_name,
"seed": seed
}
response = requests.post(url, headers=headers, data=json.dumps(data))
print("response code", response.status_code)
if response.status_code == 200:
result = response.json()
result = base64.b64decode(result[''images''][0])
result_np = np.frombuffer(result, np.uint8)
result_img = cv2.imdecode(result_np, cv2.IMREAD_UNCHANGED)
else:
print(''server error!'')
final_img = add_waterprint(result_img)
return final_img
with gr.Blocks(css = ".output-image, .input-image, .image-preview {height: 400px !important} ") as demo:
# gr.Markdown("# Outfit Anyone v0.9")
gr.HTML(
"""
<div style="display: flex; justify-content: center; align-items: center; text-align: center;">
<a href="https://github.com/HumanAIGC/OutfitAnyone" style="margin-right: 20px; text-decoration: none; display: flex; align-items: center;">
</a>
<div>
<h1 >Outfit Anyone: Ultra-high quality virtual try-on for Any Clothing and Any Person</h1>
<h4 >v0.9</h4>
<h5 style="margin: 0;">If you like our project, please give us a star on Github to stay updated with the latest developments.</h5>
<div style="display: flex; justify-content: center; align-items: center; text-align: center;>
<a href="https://github.com/HumanAIGC/OutfitAnyone"><img src="https://img.shields.io/badge/Arxiv-0000.00000-red"></a>
<a href=''https://humanaigc.github.io/outfit-anyone/''><img src=''https://img.shields.io/badge/Project_Page-OutfitAnyone-green'' alt=''Project Page''></a>
<a href=''https://github.com/HumanAIGC/OutfitAnyone''><img src=''https://img.shields.io/badge/Github-Repo-blue''></a>
</div>
</div>
</div>
""")
with gr.Row():
with gr.Column():
init_image = gr.Image(sources=''clipboard'', type="filepath", label="model", value=model)
example = gr.Examples(inputs=init_image,
examples_per_page=4,
examples=[os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Rouyan_0'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Rouyan_2'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Eva_0'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Simon_1'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Eva_1'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Simon_0'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Xuanxuan_0'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Xuanxuan_2'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Yaqi_1'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Yifeng_0'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Yifeng_3'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Rouyan_1'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Yifeng_2'')),
os.path.join(os.path.dirname(__file__), MODEL_MAP.get(''AI Model Yaqi_0'')),
])
with gr.Column():
gr.HTML(
"""
<div style="display: flex; justify-content: center; align-items: center; text-align: center;">
<div>
<h3>Models are fixed and cannot be uploaded or modified; we only support users uploading their own garments.</h3>
<h4 style="margin: 0;">For a one-piece dress or coat, you only need to upload the image to the ''top garment'' section and leave the ''lower garment'' section empty.</h4>
</div>
</div>
""")
with gr.Row():
garment_top = gr.Image(sources=''upload'', type="numpy", label="top garment")
example_top = gr.Examples(inputs=garment_top,
examples_per_page=5,
examples=[os.path.join(os.path.dirname(__file__), "garments/top222.JPG"),
os.path.join(os.path.dirname(__file__), "garments/top5.png"),
os.path.join(os.path.dirname(__file__), "garments/top333.png"),
os.path.join(os.path.dirname(__file__), "garments/dress1.png"),
os.path.join(os.path.dirname(__file__), "garments/dress2.png"),
])
garment_down = gr.Image(sources=''upload'', type="numpy", label="lower garment")
example_down = gr.Examples(inputs=garment_down,
examples_per_page=5,
examples=[os.path.join(os.path.dirname(__file__), "garments/bottom1.png"),
os.path.join(os.path.dirname(__file__), "garments/bottom2.PNG"),
os.path.join(os.path.dirname(__file__), "garments/bottom3.JPG"),
os.path.join(os.path.dirname(__file__), "garments/bottom4.PNG"),
os.path.join(os.path.dirname(__file__), "garments/bottom5.png"),
])
run_button = gr.Button(value="Run")
with gr.Column():
gallery = gr.Image()
run_button.click(fn=get_tryon_result,
inputs=[
init_image,
garment_top,
garment_down,
],
outputs=[gallery],
concurrency_limit=2)
# Examples
gr.Markdown("## Examples")
with gr.Row():
reference_image1 = gr.Image(label="model", scale=1, value="examples/basemodel.png")
reference_image2 = gr.Image(label="garment", scale=1, value="examples/garment1.jpg")
reference_image3 = gr.Image(label="result", scale=1, value="examples/result1.png")
gr.Examples(
examples=[
["examples/basemodel.png", "examples/garment1.png", "examples/result1.png"],
["examples/basemodel.png", "examples/garment2.png", "examples/result2.png"],
["examples/basemodel.png", "examples/garment3.png", "examples/result3.png"],
],
inputs=[reference_image1, reference_image2, reference_image3],
label=None,
)
if __name__ == "__main__":
ip = requests.get(''http://ifconfig.me/ip'', timeout=1).text.strip()
print("ip address alibaba", ip)
demo.queue(max_size=10)
demo.launch()
After uploading these files, you can try with any cloth on model for a better shopping experience.
Results
The complete files and app is available at OutfitAnyone — a Hugging Face Space by HumanAIGC.