GraPL

Segmentation via Graph Cuts at the Patch Level

GraPL is implemented inside the GraPL module directory. Our evaluation pipeline is also implemented in this under evaluate.py.

Usage

1. Start by installing the necessary dependencies using the provided conda environment file env.yml.
2. Unzip datasets/BSDS500.zip. This file contains all of the necessary test images and their ground truths.
3. Use the included notebooks to segment images. You can use GraPL_testbench.ipynb to test the GraPL codebase one segmentation at a time and get a sense of typical segmentation performance. Alternatively, our major experiments can be reproduced using the various experiment_*.ipynb notebooks. BSDS500 is included in this codebase for easy reproducibility.