Differentiable SLIC PyTorch

This is a PyTorch implementation of differentiable SLIC which computes superpixels with the soft assignment.
Unlike the original SLIC, the similarity between pixels and centers is computed by the inner product, which corresponds to the clustering step of HCFormer.

Environment

• Python==3.8
• PyTorch==1.12
• NumPy==1.23

How to use

See the docstring of each function and class for details.

Simple usage:

import torch

from diffSLIC import DiffSLIC

slic_fn = DiffSLIC(n_spixels=100, n_iter=5, tau=0.01, candidate_radius=1, stable=True)

rgb_img = torch.arange(30000).reshape(1, 3, 100, 100)
features, spix2pix_assign, pix2spix_assign = slic_fn(rgb_img)

Citation

This repository:

@misc{diffSLIC,
    title = {Differentiable SLIC},
    author = {Suzuki, Teppei},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/perrying/diffSLIC}},
    year = {2022},
}

HCFormer:

@article{suzuki2022clustering,
  title={Clustering as Attention: Unified Image Segmentation with Hierarchical Clustering},
  author={Suzuki, Teppei},
  journal={arXiv preprint arXiv:2205.09949},
  year={2022}
}

and its preliminary work:

@article{suzuki2021implicit,
  title={Implicit Integration of Superpixel Segmentation into Fully Convolutional Networks},
  author={Suzuki, Teppei},
  journal={arXiv preprint arXiv:2103.03435},
  year={2021}
}