This is a real time face recognition demo(1:N in the face database ) fine-tuning by FaceNet using a Chinese Face Training Dataset spiding by PIXEL COMPANY and have a better performance than the pre-trained model provided by the FaceNet's author.
1.you should register your face in the face database named ./FaceDataSet by running RegisterFace.py
2.Run Real time face detection and recognition.py to test the performance.
This is a dataset spiding by PIXEL COMPANY which is not privided in public,including 7984 people(about 38850 samples).The example of one of the persons are showed as belowed:
I use the source train_tripletloss.py provided by the FaceNet's author to fine-tuning the pre-trained model provided by him.
- Input image size(aligned Face image)
160*160*3
- embedding_size
1*1*512
- images_per_person
40
- batch_size
90
- epoch_size
500
- max_nrof_epochs
500
- Images:
0: b1.png
1: b6.png
2: y1.png
3: y6.png
4: k1.png
5: s1.png
6: j1.jpg
7: j2.jpg
8: c1.png
9: c2.png
- Distance Matrix
Distance matrix
0 1 2 3 4 5 6 7 8 9
0 0.0000 0.1829 1.2259 1.2261 1.1326 1.1039 0.9905 0.9857 1.2346 1.4076
1 0.1829 0.0000 1.2482 1.2475 1.1383 1.1056 1.0358 1.0373 1.2421 1.4111
2 1.2259 1.2482 0.0000 0.1146 1.0772 1.1344 1.0892 1.0916 0.9325 1.0117
3 1.2261 1.2475 0.1146 0.0000 1.0658 1.1275 1.1226 1.1223 0.9188 1.0004
4 1.1326 1.1383 1.0772 1.0658 0.0000 0.9561 1.1950 1.1986 1.0749 1.2382
5 1.1039 1.1056 1.1344 1.1275 0.9561 0.0000 1.0651 1.0700 1.1142 1.2993
6 0.9905 1.0358 1.0892 1.1226 1.1950 1.0651 0.0000 0.1443 1.2670 1.2683
7 0.9857 1.0373 1.0916 1.1223 1.1986 1.0700 0.1443 0.0000 1.2678 1.2682
8 1.2346 1.2421 0.9325 0.9188 1.0749 1.1142 1.2670 1.2678 0.0000 0.9273
9 1.4076 1.4111 1.0117 1.0004 1.2382 1.2993 1.2683 1.2682 0.9273 0.0000
ori model
Distance matrix
0 1 2 3 4 5 6 7 8 9
0 0.0000 0.1378 1.0435 1.0528 0.9380 0.8837 1.1112 1.1011 1.0005 0.9732
1 0.1378 0.0000 1.0400 1.0485 0.9404 0.8830 1.1114 1.1042 1.0048 0.9796
2 1.0435 1.0400 0.0000 0.1265 1.0505 1.1035 1.0239 1.0140 0.9753 0.9692
3 1.0528 1.0485 0.1265 0.0000 1.0518 1.1178 1.0379 1.0280 0.9686 0.9720
4 0.9380 0.9404 1.0505 1.0518 0.0000 0.8731 1.1282 1.1159 0.9533 1.0348
5 0.8837 0.8830 1.1035 1.1178 0.8731 0.0000 1.0611 1.0712 1.1097 1.0857
6 1.1112 1.1114 1.0239 1.0379 1.1282 1.0611 0.0000 0.1318 1.0176 1.1431
7 1.1011 1.1042 1.0140 1.0280 1.1159 1.0712 0.1318 0.0000 1.0147 1.1384
8 1.0005 1.0048 0.9753 0.9686 0.9533 1.1097 1.0176 1.0147 0.0000 0.9370
9 0.9732 0.9796 0.9692 0.9720 1.0348 1.0857 1.1431 1.1384 0.9370 0.0000
1.MTCNN
2.FaceNet