Research map / Label-efficient and robust learning / Domain adaptation and generalization
Adaptive Transformers for Robust Few-Shot Cross-Domain Face Anti-Spoofing
Hsin-Ping Huang, Deqing Sun, Yaojie Liu, Wen-Sheng Chu, Taihong Xiao, Jinwei Yuan, Hartwig Adam, Ming-Hsuan Yang
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 |
|---|---|---|---|
| add CelebA-Spoof as extra source data | 0-shot SSDG · Protocol 1 | AUC: average improvement of 0.76 AUC | Improved |
| replace SSDG backbone with ViT | 0-shot · Protocol 1/2 | AUC: +2.5 on M, +5.0 on C, +3.6 on O, +3.6 on W, +0.7 on C, +14.1 on S, -1.1 on I | Improved |
| add triplet loss or adversarial learning to ViT (ablation) | ViT · Protocol 1/2 | AUC | No clear effect |
| add 5-shot target samples | SSDG · Protocol 1/2 | AUC: average of 2.78 AUC | Improved |
| add 5-shot target samples | ViT · Protocol 1/2 | AUC: average of 0.77 AUC | Improved |
| use adapters (ViTA) instead of SSDG | 5-shot · Protocol 1/2 | AUC: +1.45 M, +5.54 C, +0.51 O, +0.40 W, +0.35 C, +3.60 S, -0.57 I | Improved |
| add FWT layers to adapters (ViTAF vs ViTA) (ablation) | ViTA · Protocol 1 | AUC: +0.46 M, +0.72 C, +0.77 I, +0.29 O | Improved |
| add FWT layers to adapters (ViTAF vs ViTA) (ablation) | ViTA · Protocol 2 | AUC: improvements only for the S domain (+2.28) | Mixed |
| use ensemble adapters with cosine loss instead of single adapter (ViTAF* vs ViTAF) (ablation) | ViTAF · Protocol 1/2 | AUC: +0.32 M, +0.07 C, +0.30 I, +0.41 O, +0.27 W, +0.86 C, +0.90 S | Improved |
| add adapters or FWT to ViT (ViTA, ViTF, ViTAF) (ablation) | ViT · Protocol 1 (average of last eight checkpoints) | AUC | Improved |
| remove adapters from first four layers instead of last four (ablation) | ViTAF* · Protocol 1 | TPR@FPR=1% | Hurt |
Closest papers by meaning
- FLIP: Cross-domain Face Anti-spoofing with Language Guidance (ICCV 2023 · similarity 0.91)
- Towards Data-Centric Face Anti-spoofing: Improving Cross-Domain Generalization via Physics-Based Data Synthesis (IJCV 2024 · similarity 0.87)
- Self-Domain Adaptation for Face Anti-Spoofing (AAAI 2021 · similarity 0.87)
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.86)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.86)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.85)
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.
