PyTorch implementation of Superpixel Sampling Networks
paper: https://arxiv.org/abs/1807.10174
original code: https://github.com/NVlabs/ssn_superpixels

Note

A pure PyTorch implementation of the core component, differentiable SLIC, is available here (note that it implements the similarity function as the cosine similarity instead of the negative Euclidean distance).

Requirements

• PyTorch >= 1.4
• scikit-image
• matplotlib

Usage

inference

SSN_pix

python inference --image /path/to/image

SSN_deep

python inference --image /path/to/image --weight /path/to/pretrained_weight

training

python train.py --root /path/to/BSDS500

Results

SSN_pix
<img src=https://github.com/perrying/ssn-pytorch/blob/master/SSN_pix_result.png>

SSN_deep
<img src=https://github.com/perrying/ssn-pytorch/blob/master/SSN_deep_result.png>