Research map / Vision-language and multimodal / CLIP and zero-shot
FLIP: Cross-domain Face Anti-spoofing with Language Guidance
Koushik Srivatsan, Muzammal Naseer, Karthik Nandakumar
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 |
|---|---|---|---|
| initialize ViT with CLIP weights instead of ImageNet weights (ablation) | ViT for face anti-spoofing · Protocol 1 (zero-shot) | average HTER: 6.00 -> 3.48 | Improved |
| initialize ViT with ImageNet instead of training from scratch (ablation) | ViT for face anti-spoofing · Protocol 1 (zero-shot) | average HTER: 25.86 -> 6.00 | Improved |
| initialize with BeIT instead of ImageNet (ablation) | ViT for face anti-spoofing · Protocol 1 (zero-shot) | average HTER: 6.00 -> 8.70 | Hurt |
| align image features to text prompts (FLIP-IT) over plain finetuning (FLIP-V) (ablation) | FLIP · Protocol 1 (zero-shot) | average HTER: 3.48 -> 3.06 | Improved |
| align image features to text prompts (FLIP-IT) over plain finetuning (FLIP-V) (ablation) | FLIP · Protocol 2 (zero-shot) | average HTER: 9.83 -> 10.2 | Hurt |
| add multimodal contrastive learning (FLIP-MCL) over FLIP-IT (ablation) | FLIP · Protocol 3 (zero-shot) | average HTER: 6.03 -> 4.84 | Improved |
| use ensemble of prompts instead of a single prompt (ablation) | FLIP-IT · Protocol 1 (zero-shot) | average HTER: 3.06 vs single prompts 2.94 to 4.52 | Mixed |
| remove SimCLR loss or MSE loss from joint loss (ablation) | FLIP-MCL · Protocol 1 (zero-shot) | average HTER: 3.47 -> 3.01 | Improved |
| upweight self-supervised losses to (1,5,5) (ablation) | FLIP-MCL · Protocol 1 (zero-shot) | average HTER: 3.01 -> 3.67 | Hurt |
| full finetuning of CLIP-pretrained ViT image encoder | FLIP-V · Protocol 1 (zero-shot) | average HTER: 6.00 -> 3.48 | Improved |
| FLIP versus zero-shot ViT on target domain M | FLIP-MCL · Protocol 1 OCI->M | HTER: 1.58 -> 4.95 | Hurt |
Closest papers by meaning
- FaceShield: Explainable Face Anti-Spoofing with Multimodal Large Language Models (AAAI 2026 · similarity 0.91)
- Interpretable Face Anti-Spoofing: Enhancing Generalization with Multimodal Large Language Models (AAAI 2025 · similarity 0.91)
- Test-Time Domain Generalization for Face Anti-Spoofing (CVPR 2024 · similarity 0.90)
- SLIP: Spoof-Aware One-Class Face Anti-Spoofing with Language Image Pretraining (AAAI 2025 · similarity 0.89)
- Gradient Alignment for Cross-Domain Face Anti-Spoofing (CVPR 2024 · similarity 0.89)
- Harnessing Chain-of-Thought Reasoning in Multimodal Large Language Models for Face Anti-Spoofing (CVPR 2026 · 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.
