This repository provides the official PyTorch implementation of the paper Multiscale Sliced Wasserstein Distances as Perceptual Color Difference Measures, accepted at ECCV 2024.

---

Requirements

• Python >= 3.7
• PyTorch >= 1.8

Installation

Clone the repository:

git clone https://github.com/real-hjq/MS-SWD.git
cd MS-SWD

Usage

Python API:

from MS_SWD import MS_SWD

msswd_model = MS_SWD(num_scale=5, num_proj=128)
# X: (N, C, H, W)
# Y: (N, C, H, W)
distance = msswd_model(X, Y)
# distance : (N,)

Command line:

python MS_SWD.py --img1 <img1_path> --img2 <img2_path>

Learned MS-SWD

A learned version of MS-SWD is available in IQA-PyTorch.

import pyiqa
import torch

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
msswd_model = pyiqa.create_metric('msswd', device=device)

Citation

If you find this work useful, please cite:

@inproceedings{he2024ms-swd,
  title={Multiscale Sliced {Wasserstein} Distances as Perceptual Color Difference Measures},
  author={He, Jiaqi and Wang, Zhihua and Wang, Leon and Liu, Tsein-I and Fang, Yuming and Sun, Qilin and Ma, Kede},
  booktitle={European Conference on Computer Vision},
  pages={1--18},
  year={2024},
  url={http://arxiv.org/abs/2407.10181}
}

Acknowledgements

Part of this implementation is adapted from GPDM. The srgb2lab conversion code is adapted from flip_loss.py in ꟻLIP. We sincerely thank the authors for making their excellent work publicly available.