https://github.com/Gal4way/TPD

This is the official repository for the paper "Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On". CVPR 2024
https://github.com/Gal4way/TPD

Last synced: 24 days ago
JSON representation

This is the official repository for the paper "Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On". CVPR 2024

• Host: GitHub
• URL: https://github.com/Gal4way/TPD
• Owner: Gal4way
• Created: 2024-03-24T08:12:08.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2024-04-28T12:18:38.000Z (8 months ago)
• Last Synced: 2024-07-31T21:00:25.339Z (4 months ago)
• Language: Python
• Size: 1.08 MB
• Stars: 70
• Watchers: 7
• Forks: 7
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-virtual-try-on - Project/Code

README

        
# [CVPR2024] TPD

This repository is the official implementation of [TPD](https://arxiv.org/abs/2404.01089)

> **Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On**

>

> Xu Yang, Changxing Ding, Zhibin Hong, Junhao Huang, Jin Tao, Xiangmin Xu

![teaser](./assets/Teaser.jpg) 

## TODO List

- [x] Release inference code

- [x] Release model weights

- [x] Release training code

- [x] Release evaluation code

## Environments

```

conda env create -f environment.yml

conda activate TPD

```

## Data Preparation

### Weights

Download the pretrained [checkpoint](https://drive.google.com/file/d/1twsjZ0kQkyFdfLcw8EYmvQmsRIgqnI3o/view?usp=sharing) and save it in the checkpoints folder like: 

```

checkpoints

|-- release

	|-- TPD_240epochs.ckpt

```

### Datasets

Download the VITON-HD dataset from [here](https://github.com/shadow2496/VITON-HD).

You should copy the test folder for validation and the dataset structure should be like: 

```

datasets/VITONHD/

test | train | validation(copied from test)

|-- agnostic-mask

|-- agnostic-v3.2

|-- cloth

|-- cloth_mask

|-- image

|-- image-densepose

|-- image-parse-agnostic-v3.2

|-- image-parse-v3

|-- openpose_img

|-- openpose_json

```

## Inference

Refer to [commands/inference.sh](./commands/inference.sh)

## Training

### Prepare

We utilize the pretrained Paint-by-Example as initialization, and increase it's first conv-layer from 9 to 18 channels (zero initiated).  Please download the [pretrained model](https://github.com/Fantasy-Studio/Paint-by-Example) first and save it in the checkpoints folder. Then run [utils/rm_clip_and_add_channels.py](./utils/rm_clip_and_add_channels.py) to add input channels of the first conv-layer and remove CLIP module. The final checkpoints folder structure is like: 

```

checkpoints

|-- original

	|-- model.ckpt

	|-- mode_prepared.ckpt	

```

### Commands

Refer to [commands/train.sh](./commands/train.sh)

## Evaluation

### Prepare 

LPIPS: https://github.com/richzhang/PerceptualSimilarity

FID: https://github.com/mseitzer/pytorch-fid

Run [utils/generate_GT.py](./utils/generate_GT.py) to generate GT images with 384*512 resolution

### Commands

Refer to  [calculate_metrics/calculate_metrics.sh](./calculate_metrics/calculate_metrics.sh)

## Acknowledgements

Thanks to [Paint-by-Example](https://github.com/Fantasy-Studio/Paint-by-Example), our code is heavily borrowed from it. 

## Citation

```

@misc{yang2024texturepreserving,

      title={Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On}, 

      author={Xu Yang and Changxing Ding and Zhibin Hong and Junhao Huang and Jin Tao and Xiangmin Xu},

      year={2024},

      eprint={2404.01089},

      archivePrefix={arXiv},

      primaryClass={cs.CV}

}

```