A collection of code commonly used to process face technology.
Note that this is not the most optimal or state-of-art demo.
Compute the average face of a set of face images or face datasets
VGGFace2 | Celeb-A |
---|---|
Face alignment
Support loss:
- softmax
- arcface
- cosface
- focal loss
we also support evidential deep learning for face recognition
AUC is often used as an evaluation index in face verification.
+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+
| 1e-06 | 1e-05 | 0.0001 | 0.001 | 0.01 | 0.1 | 0.2 | 0.4 | 0.6 | 0.8 | 1 |
+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+
| 0.9251 | 0.9251 | 0.9251 | 0.9331 | 0.9421 | 0.9570 | 0.9710 | 0.9780 | 0.9870 | 0.9940 | 1.0000 |
+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+--------+
ROC AUC: 0.9800934433250592
Visualization the model results.
image | Grad-CAM | Part Score | Edge |
---|---|---|---|