This is the Pytorch implementation of our paper Variational Prototype Learning for Deep Face Recognition which is accepted by CVPR-2021.
Define a new configure file such as configs/example_ms1m.py, and start the training process by:
bash run.sh configs/example_ms1m.py
Results on WebFace600K(subset of WebFace260M), loss is margin-based softmax.
| Backbone | Dataset | VPL? | Mask | Children | African | Caucasian | South Asian | East Asian | MR-All |
|---|---|---|---|---|---|---|---|---|---|
| R50 | WebFace600K | NO | 78.949 | 74.772 | 89.231 | 94.114 | 92.308 | 73.765 | 90.591 |
| R50 | WebFace600K | YES | 78.884 | 75.739 | 89.424 | 94.220 | 92.609 | 74.365 | 90.942 |