Unsupervised Learning of Image Segmentation Based on Differentiable Feature Clustering

This pytorch code generates segmentation labels of an input image.

Wonjik Kim*, Asako Kanezaki*, and Masayuki Tanaka. Unsupervised Learning of Image Segmentation Based on Differentiable Feature Clustering. IEEE Transactions on Image Processing, accepted, 2020. (arXiv)

*W. Kim and A. Kanezaki contributed equally to this work.

What is new?

This is an extension of our previous work.

• Better performance with spatial continuity loss
• Option of using scribbles as user input
• Option of using reference image(s)

Requirements

pytorch, opencv2, tqdm

Getting started

Vanilla

$ python demo.py --input ./BSD500/101027.jpg

Vanilla + scribbles

$ python demo.py --input ./PASCAL_VOC_2012/2007_001774.jpg --scribble

Vanilla + reference image(s)

$ python demo_ref.py --input ./BBC/