florian6973 / class_collapse

Class collapse study, for supervised contrastive learning

CLASS COLLAPSE PROJECT

Antoine Poirier, Florent Pollet

Class project based on the study of this paper.

The report can be found here. The trained models for the report can be downloaded here.

Installation

• Make sure you have Python installed (version 3.8 or above).
• Navigate to this folder.
• Run pip install -e .[cuda] -U or pip install -e .[cpu] -U (tested on Windows and Mac).

Use

Class collapse highlight on circle (section 3 of the report)

To run the script about the class circle study, you can just run the file scripts/cc_circle.py with Python.

Using $L_{spread}$ on datasets (section 4 and 5 of the report)

To run the study on embeddings, you can use the console script ccrun, like ccrun +dataset=house +loss=mse.

You can choose a dataset among house and synthetic. You can choose a loss between mse, supcon, nce, spread.

Please feel free to tune other parameters by overriding them (please see class_collapse/config and Hydra documentation).

The output will be in the folder outputs, automatically generated.

Troubleshooting

Please feel free to submit an issue if you have any questions.