Table 2 Equal error rate (EER, %, the lower the better) for BiometricNet+ as a function of the template projection size (r). For comparison, we also report the best EER obtained by the competing method SoftmaxOut [24]
From: Cancelable templates for secure face verification based on deep learning and random projections
Proj. size | LFW | YTF | CALFW | CPLFW | CFP | AgeDB |
|---|---|---|---|---|---|---|
r = 64 | 1.13 | 4.20 | 6.17 | 8.56 | 3.00 | 10.33 |
r = 128 | 0.87 | 3.54 | 5.40 | 6.69 | 1.78 | 8.11 |
r = 256 | 0.76 | 3.31 | 4.22 | 6.19 | 1.33 | 6.67 |
NOCB | 0.56 | 3.02 | 4.00 | 5.72 | 1.03 | 6.13 |
SoftmaxOut [24] | 3.73 | 9.12 | - | - | - | - |