Explainable Face Recognition (XFR) Project
Explainable face recognition is the problem of providing an interpretable reasoning for the outputs of a face recognition system. This software distribution accompanies the arXiv paper:
J. Williford, B. May and J. Byrne, “Explainable Face Recognition”, ECCV 2020, arXiv:2008.00916
In this paper, we provide the first comprehensive benchmark for explainable face recognition (XFR), and make the following contributions:
- Discriminative saliency maps. We introduce benchmark algorithms for discriminative visualization for both whitebox (convolutional network exposed) and blackbox (matching scores only) face recognition systems.
- Inpainting game protocol. We define a standardized evaluation protocol for fine grained discriminative visualization of faces.
- Inpainting game dataset for facial recognition. We publicly release the inpainting game dataset, algorithms and baseline results to support reproducible research.