Research map / Vision-language and multimodal / Cross-modal learning
Multi-View Slot Attention Using Paraphrased Texts for Face Anti-Spoofing
Jeongmin Yu, Susang Kim, Kisu Lee, Taekyoung Kwon, Won-Yong Shin, Ha Young Kim
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 MVS (multi-view slot attention) to baseline (ablation) | MVP-FAS baseline · Protocol 1 avg. | HTER (%): 8.59 -> 7.48 | Improved |
| add MTPA (multi-text patch alignment) to baseline (ablation) | MVP-FAS baseline · Protocol 1 avg. | HTER (%): 8.59 -> 4.13 | Improved |
| combine MVS and MTPA (ablation) | MVP-FAS baseline · Protocol 1 avg. | HTER (%): 8.59 -> 1.82 | Improved |
| use three paraphrased text pairs instead of one (ablation) | MVP-FAS · Protocol 1 avg. | HTER (%): 3.75 -> 1.82 | Improved |
| increase views beyond M=3 (M=4, 5) (ablation) | MVP-FAS · Protocol 1 avg. | HTER (%): 1.82 -> 1.95 (M=4), 1.97 (M=5) | Hurt |
| use mean of paraphrase embeddings as MTPA anchor instead of each text individually (ablation) | MTPA · Protocol 1 avg. | HTER (%): 4.58 -> 1.82 | Improved |
| use mean multi-text anchor instead of single text pair anchor (ablation) | MTPA · Protocol 1 avg. | HTER (%): 3.98 -> 1.82 | Improved |
| use slot attention instead of cosine similarity (ablation) | MVS · Protocol 1 avg. | HTER (%): 5.26 -> 1.82 | Improved |
| use slot attention instead of cross-attention with texts as queries (ablation) | MVS · Protocol 1 avg. | HTER (%): 5.68 -> 1.82 | Improved |
| broadcast CLS token to patch embeddings (global-aware patch embedding) (ablation) | MVS · Protocol 1 avg. | HTER (%): 5.39 -> 1.82 | Improved |
| single text pair choice among five paraphrase pairs (ablation) | MVS single-view · Protocol 1 | HTER (%): 3.75 to 4.19 avg. | Mixed |
| MVP-FAS vs best prior (BUDoPT) on Protocol 1 average | MVP-FAS · Protocol 1 avg. | HTER (%): 2.62 -> 1.82 | Improved |
Closest papers by meaning
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.89)
- Interpretable Face Anti-Spoofing: Enhancing Generalization with Multimodal Large Language Models (AAAI 2025 · similarity 0.89)
- mmFAS: Multimodal Face Anti-Spoofing Using Multi-Level Alignment and Switch-Attention Fusion (AAAI 2025 · similarity 0.89)
- Harnessing Chain-of-Thought Reasoning in Multimodal Large Language Models for Face Anti-Spoofing (CVPR 2026 · similarity 0.88)
- From Intuition to Investigation: A Tool-Augmented Reasoning MLLM Framework for Generalizable Face Anti-Spoofing (CVPR 2026 · similarity 0.87)
- 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.
