TryOffAnyone: AI tool for extracting garments from a person as a tiled garment display image

General Introduction

TryOffAnyone is a breakthrough AI image processing tool designed to solve the challenges of apparel display in the e-commerce space. It is able to intelligently convert clothing photos of real people in a wearing state into a flat display effect image, a technology based on the latest Latent Diffusion Models (LDMs). Developed and open-sourced by researchers, the project's core innovation lies in its ability to accurately identify and extract clothing areas in people's photos, and convert them into professional lay-flat display effects through complex AI algorithms. This technological breakthrough not only significantly reduces the cost of e-commerce platforms to produce product images, but also provides a more convenient and standardized solution for clothing product display. The project is open source on GitHub, supports rapid deployment and use in Python environment, and provides detailed usage documentation and sample code.

Experience: https://huggingface.co/spaces/1aurent/TryOffAnyone

Function List

• Support online image URL direct processing function
• Intelligent recognition and extraction of garment areas in images
• Automatically converts worn garments to lay-flat effect.
• Built-in professional background removal and image optimization processing
• Supports batch test function for VITON-HD dataset
• Provide detailed calculations of model evaluation metrics
• Integration of multiple image quality assessment methods (SSIM, LPIPS, FID, KID)
• Support for customizing image sizes and processing parameters
• Provide rapid deployment capability for pre-trained models
• Supports GPU-accelerated processing

Using Help

1. Environment configuration and installation

The first thing you need to do is make sure your system meets the following requirements:

• Python 3.x environment
• CUDA supported GPUs (recommended for accelerated processing)
• Git Version Control Tools

Installation Steps:

# 1. Clone the project code
git clone https://github.com/ixarchakos/try-off-anyone.git
cd try-off-anyone
# 2. Install dependencies
python3 -m pip install -r requirements.txt
# 3. Download pre-trained models
# Visit https://huggingface.co/ixarchakos/tryOffAnyone
# Place the downloaded model file in the "try-off-anyone/ckpt/" directory.

2. Basic methods of use

2.1 Processing a single online image

python3 main.py --inference --url="your image URL address"

The processed images will be saved in the "try-off-anyone/data/" directory.

2.2 Description of parameter configuration

• --seed: random seed setting (default: 36)
• --steps: number of processing steps (default: 50)
• --scale: scale (default: 2.5)
• --width: output image width (default: 384)
• --height: output image height (default: 512)
• --gpu_id: specify GPU device ID (default: 0)

3. Use of advanced functions

3.1 Batch Processing of the VITON-HD Dataset

1. Download VITON-HD raw data set
2. Download the clothing image mask file:
   • Visit: https://huggingface.co/ixarchakos/tryOffAnyone/blob/main/masks.zip
   • Extract to the "try-off-anyone/data/zalando-hd-resized/test/" directory

Perform batch processing:

python3 main.py --test

3.2 Quality assessment

The system automatically calculates the following indicators:

• SSIM (structural similarity)
• LPIPS (perceived similarity)
• FID (Fréchet Inception Distance)
• KID (Kernel Inception Distance)

4. Notes and recommendations

• Input images are recommended to use clear front-facing photos of clothing
• The current version mainly supports top-loading processing
• It is recommended to use GPU for processing to get better performance
• Monitor system resource usage when processing large numbers of images
• Regularly update models and dependency packages for best results

5. Resolution of common problems

1. If you encounter CUDA-related errors, check:
   • Is the GPU driver installed correctly
   • Does the CUDA version match the PyTorch version
2. Image processing quality issues:
   • Adjusting the --steps parameter to add processing steps
   • Adjust the --scale parameter appropriately to improve the effect
3. Insufficient memory problem:
   • Reduce batch size
   • Reduce input image size