Research map / Continual learning / Domain and prompt-based
Rehearsal-Free Domain Continual Face Anti-Spoofing: Generalize More and Forget Less
Rizhao Cai, Yawen Cui, Zhi Li, Zitong Yu, Haoliang Li, Yongjian Hu, Alex Kot
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace linear ViT-Adapter with convolutional (DCDCA-family) adapters (ablation) | ViT-Adapter · DCL-FAS Protocol-1 | mAGA: 58.72 -> 78.70 | Improved |
| use ViT-Adapter instead of ResNet18 (ablation) | finetuning · DCL-FAS Protocol-1 | mAA: 83.02 -> 86.72 | Improved |
| add learnable dynamic central-difference ratio (DCDCA) over fixed theta=0.7 (CDCA) (ablation) | ViT-CDCA · DCL-FAS Protocol-1 | mAGA: 77.01 -> 78.70 | Improved |
| add learnable dynamic central-difference ratio (DCDCA) over fixed theta=0.7 (CDCA) (ablation) | ViT-CDCA · DCL-FAS Protocol-2 | mAGA: 74.48 -> 78.09 | Improved |
| add central difference (CDCA) over plain convolution adapter (ConvA) (ablation) | ViT-ConvA · DCL-FAS Protocol-1 | mAA: 93.29 -> 93.06 | No clear effect |
| add PPCR instead of plain finetuning (ablation) | ViT-DCDCA · DCL-FAS Protocol-1 | mAGA: 78.70 -> 82.06 | Improved |
| add PPCR instead of plain finetuning (ablation) | ViT-DCDCA · DCL-FAS Protocol-1 | mABT: -3.59 -> -3.29 | Improved |
| add PPCR instead of plain finetuning (ablation) | ViT-ConvA · DCL-FAS Protocol-1 | mAA: 93.29 -> 93.04 | No clear effect |
| add proxy prototypes to contrastive regularization (PPCR vs CR) (ablation) | ViT-DCDCA · DCL-FAS Protocol-1 | mABT: -3.57 -> -3.48 | Improved |
| add proxy prototypes to contrastive regularization (PPCR vs CR) (ablation) | ViT-DCDCA · DCL-FAS Protocol-2 | mAGA: 80.31 -> 80.08 | No clear effect |
| PPCR instead of EWC/LWF regularization | ViT-DCDCA · DCL-FAS Protocol-1 | mAGA: 78.41 (EWC), 78.68 (LWF) -> 82.06 | Improved |
| PPCR instead of EWC/LWF regularization | ViT-DCDCA · DCL-FAS Protocol-1 | mABT: -0.163 (EWC), -0.97 (LWF) vs -3.48 | Hurt |
Closest papers by meaning
- Rehearsal-Free and Efficient Continual Learning for Cross-Domain Face Anti-Spoofing (TPAMI 2025 · similarity 0.95)
- Multi-Domain Learning for Updating Face Anti-Spoofing Models (ECCV 2022 · similarity 0.90)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.89)
- Domain-Hallucinated Updating for Multi-Domain Face Anti-spoofing (AAAI 2024 · similarity 0.88)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.88)
- CFPL-FAS: Class Free Prompt Learning for Generalizable Face Anti-spoofing (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
