🐈 CatVTON: Concatenation Is All You Need for Virtual Try-On with Diffusion Models

Setting Up the Environment

1. Clone the repository:

git clone https://github.com/your-repo/catvton.git

2. Create and activate a conda environment:

cd <path_to_your_folder_project>
conda create -n catvton python==3.9.0
conda activate catvton

3. Install dependencies:

pip install -r requirements.txt

4. Download VITON-HD dataset:

VITON-HD dataset kaggle

Inference

1. Data preparation

Once the datasets are downloaded, the folder structures should look like these:

├── VITON-HD
|   ├── test_pairs_paired.txt
│   ├── test
|   |   ├── image
│   │   │   ├── [000006_00.jpg | 000008_00.jpg | ...]
│   │   ├── cloth
│   │   │   ├── [000006_00.jpg | 000008_00.jpg | ...]
│   │   ├── agnostic-mask
│   │   │   ├── [000006_00_mask.png | 000008_00.png | ...]
...

We just use these for inference processing.

2. Inference on VTION-HD

Run the following command, checkpoints will be automatically downloaded from HuggingFace.

$env:CUDA_VISIBLE_DEVICES = "0"
python inference.py `
    --dataset_name vitonhd `
    --data_root_path "C:\Users\ADMIN\CatVTON\VITON-HD" `
    --output_dir "C:\Users\ADMIN\CatVTON\output" `
    --dataloader_num_workers 8 `
    --batch_size 8 `
    --seed 555 `
    --mixed_precision fp16 `
    --allow_tf32 `
    --repaint `
    --eval_pair

INPUT:

OUTPUT:

Fast API