Face Anti-Spoofing on OULU-NPU (test)
0.9999AUC-ROCCASO-PAD
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CASO-PADBackbone=involution-augmented MobileNetV3, Input Resolution=256x256, Groups=120, Reduction factor=4, Kernel size=52026.02 | 0.9999 | 99.68 | 0.991 | 44 | 23 | 44 | 65 | |
| I-FASProtocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 0.9268 | — | — | — | — | 14.86 | — | |
| TAR-FASProtocol=Protocol 2, Training Dataset=CelebA-Spoof, Base Model=InternVL-3-8B2026.03 | 0.9263 | — | — | — | — | 14.45 | — | |
| FLIPProtocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 0.873 | — | — | — | — | 20.57 | — | |
| ViTAFProtocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 0.8128 | — | — | — | — | 26.73 | — | |
| ViT-LProtocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 0.7807 | — | — | — | — | 29.42 | — |