Research map / Video / Video segmentation and tracking
Lips Don't Lie: A Generalisable and Robust Approach To Face Forgery Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use lipreading-pretrained features and finetune only the temporal network | LipForensics · FF++ HQ cross-manipulation average | AUC (%): 94.9 -> 97.1 | Helped |
| pretrain the detector on lipreading rather than training from scratch (ablation) | LipForensics · FF++ cross-manipulation | AUC (%): 86.7 -> 94.4; 86.7 -> 97.1 | Helped |
| freeze the feature extractor instead of finetuning the whole network (ablation) | LipForensics · FF++ cross-manipulation | AUC (%): 94.4 -> 97.1 | Helped |
| add the MS-TCN to model long-term temporal information (ablation) | LipForensics · FaceShifter and DeeperForensics | Video-level accuracy (%): 67.1 -> 83.2; 74.6 -> 84.6 | Helped |
| add short-term temporal modeling with ResNet-3D/2D (ablation) | LipForensics · FaceShifter | Video-level accuracy (%): 59.3 -> 67.1 | Helped |
| add short-term temporal modeling with ResNet-3D/2D (ablation) | LipForensics · DeeperForensics | Video-level accuracy (%): 75.7 -> 74.6 | Hurt |
| pretrain on lipreading instead of other tested pretraining datasets (ablation) | LipForensics · FaceShifter and DeeperForensics | Video-level accuracy (%): Kinetics: 65.7 / 68.2; LRW: 83.2 / 84.6 | Helped |
| use frozen features rather than finetuning the whole network (ablation) | LipForensics · FaceShifter and DeeperForensics | Video-level accuracy (%): 83.2 -> 87.5; 84.6 -> 90.4 | Helped |
| use mouth crops instead of full-face crops with lipreading pretraining (ablation) | LipForensics · FaceShifter | Video-level accuracy (%): 84.3 -> 87.5 | Helped |
| use mouth crops instead of full-face crops with lipreading pretraining (ablation) | LipForensics · DeeperForensics | Video-level accuracy (%): 90.0 -> 90.4 | Helped |
| train on more diverse forgery methods (ablation) | LipForensics · FaceSwap | AUC (%): 95.1 | Helped |
| increase the number of frames per clip (ablation) | LipForensics · FaceShifter and DeeperForensics | Video-level performance | Helped |
Closest papers by meaning
- Lips Are Lying: Spotting the Temporal Inconsistency between Audio and Visual in Lip-Syncing DeepFakes (NeurIPS 2024 · similarity 0.90)
- SpeechForensics: Audio-Visual Speech Representation Learning for Face Forgery Detection (NeurIPS 2024 · similarity 0.90)
- Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector (CVPR 2025 · similarity 0.86)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.86)
- Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection (CVPR 2025 · similarity 0.86)
- A Hitchhiker's Guide to Fine-Grained Face Forgery Detection Using Common Sense Reasoning (NeurIPS 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.