Face Anti-Spoofing on CASIA-MFSD
0HTEROurs-CelebA
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Ours-CelebATraining Dataset=CelebA-Spoof (annotated by FAS caption model)2025.06 | 0 | — | — | — | 100 | |
| Ours-100KTraining Dataset=FaceCoT-Gold100K2025.06 | 0 | — | — | — | 100 | |
| Ours-AllTraining Dataset=Full FaceCoT2025.06 | 0 | — | — | — | 100 | |
| TAR-FASTraining Protocol=Protocol 2, Training Dataset=CelebA-Spoof, Base MLLM=InternVL-3-8B2026.03 | 0 | — | — | — | 100 | |
| TAR-FASEvaluation Protocol=Protocol 22026.03 | 0 | — | — | — | 100 | |
| ViT-BTraining Dataset=CelebA-Spoof2025.06 | 0.7 | — | — | — | 99.86 | |
| ViT-LTraining Dataset=CelebA-Spoof2025.06 | 0.93 | — | — | — | 99.95 | |
| ViT-LTraining Protocol=Protocol 2, Training Dataset=CelebA-Spoof, Backbone=ViT-L2026.03 | 0.93 | — | — | — | 99.95 | |
| ViT-LEvaluation Protocol=Protocol 22026.03 | 0.93 | — | — | — | 99.95 | |
| I-FASTraining Dataset=CelebA-Spoof2025.06 | 1.11 | — | — | — | 99.88 | |
| I-FASTraining Protocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 1.11 | — | — | — | 99.88 | |
| I-FASEvaluation Protocol=Protocol 22026.03 | 1.11 | — | — | — | 99.88 | |
| ViTAFTraining Dataset=CelebA-Spoof2025.06 | 3.11 | — | — | — | 99.48 | |
| ViTAFTraining Protocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 3.11 | — | — | — | 99.48 | |
| ViTAFEvaluation Protocol=Protocol 22026.03 | 3.11 | — | — | — | 99.48 | |
| FLIPTraining Dataset=CelebA-Spoof2025.06 | 4.88 | — | — | — | 98.48 | |
| FLIPTraining Protocol=Protocol 2, Training Dataset=CelebA-Spoof2026.03 | 4.88 | — | — | — | 98.48 | |
| FLIPEvaluation Protocol=Protocol 22026.03 | 4.88 | — | — | — | 98.48 | |
| CDCN2020.03 | — | 98.48 | 99.9 | 99.8 | — | |
| CDCN++2020.03 | — | 98.07 | 99.9 | 99.6 | — | |
| DTNlearning paradigm=zero-shot learning2020.03 | — | 90 | 97.3 | 97.5 | — | |
| NN+LBP2020.03 | — | 94.16 | 88.39 | 79.85 | — | |
| OC-SVM_RBF+BSIF2020.03 | — | 70.74 | 60.73 | 95.9 | — | |
| SVM_RBF+LBP2020.03 | — | 91.94 | 91.7 | 84.47 | — |