Mining bias-target Alignment from Voronoi Cells

The official repository. Please cite as

@InProceedings{Nahon_2023_ICCV,
    author    = {Nahon, R\'emi and Nguyen, Van-Tam and Tartaglione, Enzo},
    title     = {Mining bias-target Alignment from Voronoi Cells},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {4946-4955}
}

This is an example implementation of : Mining bias-target Alignment from Voronoi Cells.

This repository is tailored for Biased-MNIST but can be adapted to other datasets.

Installations

If you want to work with conda :

conda create -n voronoi_cells_bias_alignment python=3.9
conda activate voronoi_cells_bias_alignment
pip3 install -r pip_requirements.txt

If not, simply install the requirement with :

pip3 install -r pip_requirements.txt

Run our method on Biased-MNIST

1. In a terminal, get inside the voronoi_cells_bias_alignment folder

2. Run : python3 main.py

Possible extra arguments

• --dev : to specify the device you want to work on (ex: cpu or cuda:0)

• --rho : to select a level of digit-color correlation in Biased-MNIST (ex : 0.997)

• other extra arguments detailed in utils/configs.py