Research map / Label-efficient and robust learning / Domain adaptation and generalization
Purify then Guide: Rethinking Domain Generalization for Multimodal Face Anti-Spoofing
Yingjie Ma, Xun Lin, Zitong Yu, Haonan Wang, Ruixin Zhang, Shouhong Ding, Xin Liu, Xiaochen Yuan, Weicheng Xie, Linlin Shen
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 standard multi-head self-attention (ablation) | Dense Adaptor · PS->CW | HTER: 23.26 -> 25.85 | Hurt |
| Replace MHSA with Differential Attention (ablation) | Dense Adaptor · PS->CW | HTER: 25.85 -> 16.49 | Improved |
| Replace DA with MD2A (ablation) | Dense Adaptor · PS->CW | HTER: 16.49 -> 13.47 | Improved |
| Replace DA with MD2A (ablation) | MoE Adaptor · PS->CW | HTER: 12.72 -> 9.70 | Improved |
| Replace MHSA with MD2A (ablation) | MoE Adaptor · PS->CW | HTER: 12.83 -> 9.70 | Improved |
| Use label smoothing in alignment (ablation) | MMDA · PS->CW | HTER: 9.70 -> 9.17 | Improved |
| Use RS2 subspace alignment plus shared classifier instead of smooth alignment (ablation) | MMDA · PS->CW | HTER: 9.17 -> 8.88 | Improved |
| Increase U-DSA depth up to 7 layers (ablation) | MMDA · PS->CW | AUC | Improved |
| Use MMDA over DADM, complete modality | MMDA · Protocol 1 average | HTER: 13.63 -> 4.00 | Improved |
| Use MMDA over DADM, missing modality | MMDA · Protocol 2 average | HTER: 21.66 -> 10.14 | Improved |
| Pair MD2A noise branch with same-domain samples instead of random or cross-domain (ablation) | MD2A · residual variance | Var[R]: 7.48e-7 (random) -> 4.31e-7 | Improved |
| Same-domain pairing for decision variance (ablation) | MD2A · decision stability | Var[z]: 1.662e-3 (random) vs 1.672e-3 (same-domain) | No clear effect |
Closest papers by meaning
- Semantic Freedom Bottleneck for Domain-Generalized Multimodal Face Anti-Spoofing (NeurIPS 2026 · similarity 0.94)
- mmFAS: Multimodal Face Anti-Spoofing Using Multi-Level Alignment and Switch-Attention Fusion (AAAI 2025 · similarity 0.92)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.91)
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.90)
- Interpretable Face Anti-Spoofing: Enhancing Generalization with Multimodal Large Language Models (AAAI 2025 · similarity 0.90)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.89)
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.
