Research map / Video / Video segmentation and tracking
Leveraging Real Talking Faces via Self-Supervision for Robust 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 the two-stage method rather than training CSN on forgery data alone (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 82.1 -> 97.1; 83.1 -> 97.1 | Helped |
| Add stage 2 auxiliary video-target prediction to finetuning (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 95.0 -> 97.1; 95.2 -> 97.1 | Helped |
| Use logit adjustment (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 95.7 -> 97.1; 96.4 -> 97.1 | Helped |
| Use time masking (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 96.1 -> 97.1; 95.9 -> 97.1 | Helped |
| Use random erasing (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 96.3 -> 97.1; 96.3 -> 97.1 | Helped |
| Use dense rather than global representations (ablation) | RealForensics stage 1 · FaceShifter and DeeperForensics | accuracy: 87.9 -> 97.1; 89.1 -> 97.1 | Helped |
| Omit negatives in dense representation learning (ablation) | RealForensics stage 1 · FaceShifter and DeeperForensics | accuracy: 96.4 -> 97.1; 96.8 -> 97.1 | Helped |
| Add a predictor to non-contrastive representation learning (ablation) | RealForensics stage 1 · FaceShifter and DeeperForensics | accuracy | Mixed |
| Use the full face rather than mouth crops (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: 95.5 -> 97.1; 95.0 -> 97.1 | Helped |
| Use audiovisual rather than visual-only representation learning (ablation) | RealForensics stage 1 · FaceShifter and DeeperForensics | accuracy: 92.9 -> 97.1; 89.7 -> 97.1 | Helped |
| Use a transformer rather than an MLP predictor (ablation) | RealForensics stage 1 · FaceShifter and DeeperForensics | accuracy: 91.8 -> 97.1; 92.9 -> 97.1 | Helped |
| Train with longer video clips (ablation) | RealForensics · FaceShifter and DeeperForensics | accuracy: FaceShifter: 87.9 (5 frames) -> 97.1 (25 frames), 96.7 (30 frames); DeeperForensics: 88.2 (5 frames) -> 97.4 (30 frames) | Mixed |
Closest papers by meaning
- SpeechForensics: Audio-Visual Speech Representation Learning for Face Forgery Detection (NeurIPS 2024 · similarity 0.91)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.89)
- A Hitchhiker's Guide to Fine-Grained Face Forgery Detection Using Common Sense Reasoning (NeurIPS 2024 · similarity 0.88)
- Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector (CVPR 2025 · similarity 0.88)
- Face Forgery Video Detection via Temporal Forgery Cue Unraveling (CVPR 2025 · similarity 0.88)
- Forensics Adapter: Adapting CLIP for Generalizable Face Forgery Detection (CVPR 2025 · 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.