Table 2.
The achieved verification performance of different experimental settings by ResNet-100, ResNet-50 and MobileFaceNet models along with EUM trained with triplet loss and EUM trained with SRT loss. The result is reported using MRF2 dataset. The FMR100_ThUMR-UMP are equal to 0.1711, 0.2038 and 0.2351 for ResNet-100, ResNet-50 and MobileFaceNet, respectively. The FMR1000_ThUMR-UMP are equal to 0.2316, 0.2639 and 0.3041 for ResNet-100, ResNet-50 and MobileFaceNet, respectively. The lowest EER and the lowest average error of FMR100 and FMR1000 at the defined threshold for each of the evaluation cases and each of the evaluated models are marked in bold. One can notice the significant improvement in the verification performance induced by our proposed approach (SRT) in most evaluation cases.
| FMR100_ThUMR-UMP |
FMR1000_ThUMR-UMP |
||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MRF2 | Setting | EER% | FMR100% | FMR1000% | FMR% | FNMR% | Avg.% | FMR% | FNMR% | Avg.% | G-mean | I-mean | FDR |
| UMR-UMP | 0.0000 | 0.0000 | 0.0000 | 1.0000 | 0.0000 | 0.5000 | 0.1000 | 0.0000 | 0.0500 | 0.7605 | 0.0019 | 46.4218 | |
| UMR-MP | 4.0515 | 6.7568 | 7.0946 | 0.9079 | 6.7568 | 3.8323 | 0.1127 | 7.0946 | 3.6036 | 0.4454 | -0.0000 | 9.3458 | |
| UMR-MP(T) | 4.0515 | 6.7568 | 9.4595 | 0.7820 | 6.7568 | 3.7694 | 0.0530 | 11.1486 | 5.6008 | 0.3677 | -0.0012 | 8.3377 | |
| UMR-MP(SRT) | 3.3757 | 5.4054 | 7.0946 | 0.9145 | 5.7432 | 3.3289 | 0.1127 | 7.0946 | 3.6036 | 0.4587 | -0.0003 | 9.8264 | |
| MR-MP | 3.7522 | 3.7559 | 8.4507 | 4.3648 | 3.4429 | 3.9039 | 1.0079 | 3.7559 | 2.3819 | 0.6757 | 0.0183 | 6.4714 | |
| MR-MP(T) | 4.3817 | 9.0767 | 21.5962 | 20.6247 | 2.5039 | 11.5643 | 9.3461 | 3.1299 | 6.2380 | 0.6947 | 0.0834 | 5.8089 | |
| ResNet-100 | MR-MP(SRT) | 3.4416 | 4.3818 | 8.4507 | 3.8651 | 3.1299 | 3.4975 | 0.8247 | 4.3818 | 2.6033 | 0.6738 | 0.0099 | 6.4496 |
| UMR-UMP | 0.0000 | 0.0000 | 0.0000 | 1.0000 | 0.0000 | 0.5000 | 0.1000 | 0.0000 | 0.0500 | 0.7477 | 0.0038 | 37.9345 | |
| UMR-MP | 4.3895 | 6.7568 | 10.4730 | 0.7025 | 8.4459 | 4.5742 | 0.0795 | 10.8108 | 5.4452 | 0.4263 | 0.0005 | 8.2432 | |
| UMR-MP(T) | 6.4169 | 7.7703 | 12.1622 | 0.4241 | 8.7838 | 4.6040 | 0.0000 | 17.9054 | 8.9527 | 0.3567 | -0.0066 | 6.8853 | |
| UMR-MP(SRT) | 4.7274 | 7.4324 | 9.4595 | 0.8748 | 7.4324 | 4.1536 | 0.1193 | 9.1216 | 4.6205 | 0.4553 | 0.0014 | 8.4507 | |
| MR-MP | 6.8831 | 10.0156 | 13.7715 | 4.2316 | 7.8247 | 6.0281 | 1.1662 | 9.7027 | 5.4344 | 0.6496 | 0.0301 | 4.7924 | |
| MR-MP(T) | 6.8831 | 9.7027 | 14.0845 | 97.8759 | 0.0000 | 48.9379 | 90.7622 | 0.0000 | 45.3811 | 0.7759 | 0.3663 | 4.8791 | |
| ResNet-50 | MR-MP(SRT) | 6.2578 | 9.0767 | 11.8936 | 2.9738 | 8.1377 | 5.5557 | 0.8413 | 9.3897 | 5.1155 | 0.6488 | 0.0144 | 4.9381 |
| UMR-UMP | 0.0106 | 0.0000 | 0.0000 | 1.0000 | 0.0000 | 0.5000 | 0.1000 | 0.0000 | 0.0500 | 0.7318 | 0.0078 | 26.4276 | |
| UMR-MP | 6.4169 | 16.8919 | 24.3243 | 0.9874 | 16.8919 | 8.9397 | 0.0663 | 27.3649 | 13.7156 | 0.3803 | -0.0019 | 4.6457 | |
| UMR-MP(T) | 7.7685 | 15.8784 | 34.4595 | 0.6759 | 18.9189 | 9.7974 | 0.0596 | 37.1622 | 18.6109 | 0.3304 | -0.0027 | 4.2067 | |
| UMR-MP(SRT) | 6.079 | 12.5000 | 21.9595 | 0.9675 | 13.1757 | 7.0716 | 0.0928 | 22.2973 | 11.1950 | 0.4157 | -0.0018 | 5.2918 | |
| MR-MP | 8.4777 | 18.1534 | 28.7950 | 6.5056 | 10.3286 | 8.4171 | 1.9908 | 14.0845 | 8.0377 | 0.6087 | 0.0509 | 3.2505 | |
| MR-MP(T) | 8.7634 | 17.5274 | 26.2911 | 95.9683 | 0.0000 | 47.9842 | 84.9896 | 0.0000 | 42.4948 | 0.7638 | 0.3966 | 3.5408 | |
| MobileFaceNet | MR-MP(SRT) | 7.8232 | 15.0235 | 22.5352 | 3.9733 | 9.0767 | 6.525 | 1.1745 | 14.3975 | 7.7860 | 0.6087 | 0.0241 | 3.5815 |