fbiying87/Demographic-Bias-Visualization

Demographic_Bias_Visualization

##The implementatioon of "Towards Explaining Demographic Bias through the Eyes of Face Recognition Models", IJCB 2022.

The pipeline describes the framework of our proposed explainability tools.

Visualization outputs:

Configuration Environment

• Python 3.6
• Pytorch
• torchvision

Dataset

RFW, BFW The activation mappings are stored in the data, which can be downloaded from here. Please download the data and unzip it in the same folder in the following structure.

data 
└───r50
   └───RFW
       │   file111.npy
   └───BFW
       │   file111.npy
   └───gender
       │   file111.npy
└───r100
    └───RFW
       │   file111.npy
   └───BFW
       │   file111.npy
   └───gender
       │   file111.npy

Run

To get the visualization for ethnicity please run:

$ python main.py

To get the visualization for gender please run:

$ python ActivationMapping_Gender.py

To get the fairness metric, please run:

$ python global_threshold.py

Citation

Please cite our paper if the code is helpful to your research.