https://github.com/ryushinn/texture-synthesis-sliced-wasserstein

An unofficial JAX implementation of "A Sliced Wasserstein Loss for Neural Texture Synthesis" (CVPR 2021).
https://github.com/ryushinn/texture-synthesis-sliced-wasserstein

jax sliced-wasserstein-distance texture-synthesis

Last synced: about 1 month ago
JSON representation

An unofficial JAX implementation of "A Sliced Wasserstein Loss for Neural Texture Synthesis" (CVPR 2021).

• Host: GitHub
• URL: https://github.com/ryushinn/texture-synthesis-sliced-wasserstein
• Owner: ryushinn
• Created: 2024-08-02T13:07:59.000Z (5 months ago)
• Default Branch: main
• Last Pushed: 2024-08-03T18:01:33.000Z (5 months ago)
• Last Synced: 2024-08-04T11:59:08.541Z (5 months ago)
• Topics: jax, sliced-wasserstein-distance, texture-synthesis
• Language: Python
• Homepage:
• Size: 38.5 MB
• Stars: 0
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# A Sliced Wasserstein Loss for Neural Texture Synthesis

> This is an unofficial [**JAX**](https://github.com/google/jax) implementation for [A Sliced Wasserstein Loss for Neural Texture Synthesis (CVPR'21)](https://arxiv.org/abs/2006.07229).

Please see [here](https://github.com/tchambon/A-Sliced-Wasserstein-Loss-for-Neural-Texture-Synthesis) for the author's repository and cite them:

```bib

@InProceedings{Heitz_2021_CVPR,

    author = {Heitz, Eric and Vanhoey, Kenneth and Chambon, Thomas and Belcour, Laurent},

    title = {A Sliced Wasserstein Loss for Neural Texture Synthesis},

    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},

    month = {June},

    year = {2021}

}

```

## Notes

We require these libraries:

```bash

pip install -U "jax[cuda]" equinox optax tqdm pillow

```

The pre-trained VGG weights `vgg19.npy` is ported from the `vgg19.pth` file provided in the official repo.

We re-write the VGG network and Slice Wasserstein Loss in JAX code.

## Run

```bash

python texsyn.py --exemplar_path data/input.png --loss_type sw

```

## Results

Input | Output (Slice) | Output (Gram)

---------|----------|---------

 ![alt text](data/input.png) | ![alt text](data/result_sw.png) | ![alt text](data/result_gram.png)

## Last words

Thanks all efforts put on making all mentioned repositories public.

We appreciate bug reports. I will fix them when I make time around.