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

Start by installing the necessary dependencies using the provided conda environment file env.yml.
Unzip datasets/BSDS500.zip. This file contains all of the necessary test images and their ground truths.
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.

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
.ipynb_checkpoints		.ipynb_checkpoints
.jupyter/desktop-workspaces		.jupyter/desktop-workspaces
GraPL		GraPL
__MACOSX/PascalVOC2012		__MACOSX/PascalVOC2012
datasets		datasets
experiment_results		experiment_results
.DS_Store		.DS_Store
.gitignore		.gitignore
GraPL_testbench.ipynb		GraPL_testbench.ipynb
README.md		README.md
coco_val.ipynb		coco_val.ipynb
env.yml		env.yml
experiment_baselines.ipynb		experiment_baselines.ipynb
experiment_coldstart.ipynb		experiment_coldstart.ipynb
experiment_continuity_ablation.ipynb		experiment_continuity_ablation.ipynb
experiment_d.ipynb		experiment_d.ipynb
experiment_graph_weights.ipynb		experiment_graph_weights.ipynb
experiment_initialization.ipynb		experiment_initialization.ipynb
experiment_lambda_ablation.ipynb		experiment_lambda_ablation.ipynb
naive_voc.ipynb		naive_voc.ipynb
naive_voc.py		naive_voc.py
voc_with_separation.py		voc_with_separation.py