Table 5 Size (MB) and number of trainable parameters of the CNN models used in the benchmark.
From: A new periocular dataset collected by mobile devices in unconstrained scenarios
Model | Size (MB) | Trainable parameters |
|---|---|---|
VGG16 | 1088 | 135, 886, 084 |
VGG16-Face | 1088 | 135, 886, 084 |
InceptionResNet | 445 | 55, 246, 372 |
ResNet50V2 | 400 | 49, 786, 436 |
ResNet50 | 198 | 24, 609, 284 |
ResNet50-Face | 198 | 24, 609, 284 |
Xception | 176 | 21, 908, 204 |
DenseNet121 | 64 | 7, 792, 964 |
MobileNetV2 | 26 | 3, 128, 516 |
Multi-task | 37 | 4, 494, 230 |
Siamese | 21 | 2, 551, 808 |
Pairwise | 20 | 2, 349, 479 |
