GraPix: Exploring Graph Modularity Optimization for Unsupervised Pixel Clustering

This Paper is published in the 'International Conference on Pattern Recognition (ICPR) 2024.

Watch the GraPix presentation on YouTube.

Unsupervised Segmentation Loss

Data directory organization

data
|
suim
|
|── imgs
|   ├── train
|   |   |── name_1.jpg
|   |   └── name_2.jpg
|   └── val
|       |── name_11.jpg
|       └── name_22.jpg
└── labels
    ├── train
    |   |── name_1.png
    |   └── name_2.png
    └── val
        |── name_12.png
        └── name_22.png

Setup and Activate Conda Environment

Step 1: cd GraPix-master folder
Step 2: conda env create -f GraPix.yml
Step 3: conda activate GraPix

Training GraPix

Step 1: cd src/
Step 2: python main.py
NOTE: Edit dataset path in 'config/train.yaml' before training. Also, after the GraPix training set required booling variable to 'true' and change checkpoint path in yaml file for corresponding Additional Unsupervised Training.

Evaluating Grapix

Step 1: Modify the eval.yaml file of the Config folder accordingly. Change the 'experiment_name' and 'model_paths'.
Step 2: Run cmd 'python eval_segmentation.py'
Note: To reproduce paper results for the GraPix stage, comment out lines 64 to 70 in the 'main.py' file. 

Results

Cite Us

Kumar, S., Sur, A., Baruah, R.D. (2025). GraPix: Exploring Graph Modularity Optimization for Unsupervised Pixel Clustering. In: Antonacopoulos, A., Chaudhuri, S., Chellappa, R., Liu, CL., Bhattacharya, S., Pal, U. (eds) Pattern Recognition. ICPR 2024. Lecture Notes in Computer Science, vol 15310. Springer, Cham. https://doi.org/10.1007/978-3-031-78192-6_13 

Acknowledgement

We thank the authors of the STEGO for providing details on their implementation.