Research map / Vision-language and multimodal / Human and embodied modeling
Unleashing Vision-Language Semantics for Deepfake Video Detection
Jiawen Zhu, Yunqi Miao, Xueyi Zhang, Jiankang Deng, Guansong Pang
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 forgery localization map for local-level authenticity learning to frozen CLIP (ablation) | frozen CLIP baseline · CDF-v2 (frame-level) | AUROC: 58.3 -> 76.3 | Improved |
| add forgery-aware masks into CLIP self-attention (ablation) | VLAForge (+T1) · CDF-v2 (frame-level) | AUROC: 76.3 -> 82.3 | Improved |
| add VLA attention map for local-level reasoning (ablation) | VLAForge (+T2) · CDF-v2 (frame-level) | AUROC: 82.3 -> 90.8 | Improved |
| inject identity prior token into text prompts (ablation) | VLAForge (+T3) · CDF-v2 (frame-level) | AUROC: 90.8 -> 91.2 | Improved |
| inject identity prior token into text prompts (ablation) | VLAForge (+T3) · SiT full-face (frame-level) | AUROC: 76.8 -> 77.4 | Improved |
| add VLA attention loss L_VLA (ablation) | VLAForge · CDF-v2 (frame-level) | AUROC: 89.1 -> 91.2 | Improved |
| add orthogonality loss L_orth on forgery-aware masks (ablation) | VLAForge · SiT full-face (frame-level) | AUROC: 76.4 -> 77.4 | Improved |
| add orthogonality loss L_orth on forgery-aware masks (ablation) | VLAForge · VQGAN full-face (video-level) | AUROC: 97.9 -> 99.7 | Improved |
| use VLAForge instead of ForAda | CLIP-based DFD · DFDC (frame-level) | AUROC: 84.3 -> 87.0 | Improved |
| use VLAForge instead of ForAda on full-face generated forgeries | CLIP-based DFD · DiT (video-level) | AUROC: 67.1 -> 80.3 | Improved |
| increase number of query tokens q beyond 128 (ablation) | VLAForge · average over five datasets | AUROC | Hurt |
| set fusion weight alpha near 0.5 (ablation) | VLAForge · average over five datasets | AUROC | Improved |
Closest papers by meaning
- Unlocking the Capabilities of Large Vision-Language Models for Generalizable and Explainable Deepfake Detection (ICML 2025 · similarity 0.91)
- Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection (AAAI 2025 · similarity 0.90)
- Rethinking Vision-Language Model in Face Forensics: Multi-Modal Interpretable Forged Face Detector (CVPR 2025 · similarity 0.90)
- A Hitchhiker's Guide to Fine-Grained Face Forgery Detection Using Common Sense Reasoning (NeurIPS 2024 · similarity 0.90)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.90)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.90)
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.
