Research map / Video / Video segmentation and tracking
Your One-Stop Solution for AI-Generated Video Detection
Long Ma, Zihao Xue, Yan Wang, Zhiyuan Yan, Jin Xu, Xiaorui Jiang, Haiyang Yu, Yong Liao, Zhen Bi
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 Effort's fine-tuning strategy on top of the base CLIP-ViT-L/14 (ablation) | CLIP-ViT-L/14 detector · AIGVDBench open and closed source | AUC: 1.8% on open source, 28.79% on closed source | Improved |
| train on a small dataset for more epochs (ForgeLens1) instead of a large dataset for fewer epochs (ForgeLens3) (ablation) | ForgeLens · AIGVDBench open and closed source | AUC: 2.22% on open-source, 8.16% on closed source; avg 89.60 -> 91.82 | Improved |
| move evaluation from open-source to closed-source generators | I3D · AIGVDBench | AUC: 82.99 -> 61.18 | Hurt |
| move evaluation from open-source to closed-source generators | CNNSpot · AIGVDBench | AUC: 85.61 -> 70.60 | Hurt |
| move evaluation from open-source to closed-source generators | Effort · AIGVDBench | AUC: 87.49 -> 94.05 | Improved |
| move evaluation from open-source to closed-source generators | TimeSformer · AIGVDBench | AUC: 81.50 -> 86.50 | Improved |
| evaluate on V2V instead of T2V videos | TimeSformer · AIGVDBench open-source | AUC: 86.84 -> 67.59 | Hurt |
| evaluate on V2V instead of T2V videos | Effort · AIGVDBench open-source | AUC: 90.20 -> 74.74 | Hurt |
| evaluate on I2V instead of T2V videos | I3D · AIGVDBench open-source | AUC: 80.78 -> 89.05 | Improved |
| evaluate on I2V instead of T2V videos | TimeSformer · AIGVDBench open-source | AUC: 86.84 -> 75.46 | Hurt |
| use a CLIP-based image detector (UnivFD) instead of a ResNet-based one (CNNSpot) | image detector · AIGVDBench open-source | AUC: 85.61 -> 85.69 | No clear effect |
| scale the VLM from Qwen2.5-VL-3B to 7B to 32B | Qwen2.5-VL · AIGVDBench open-source | accuracy: 50.35 -> 51.84 -> 51.12 | Mixed |
Closest papers by meaning
- Is Artificial Intelligence Generated Image Detection a Solved Problem? (NeurIPS 2025 · similarity 0.90)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.88)
- A Sanity Check for AI-generated Image Detection (ICLR 2025 · similarity 0.88)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.87)
- AI-Generated Video Detection via Perceptual Straightening (NeurIPS 2025 · similarity 0.87)
- CO-SPY: Combining Semantic and Pixel Features to Detect Synthetic Images by AI (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.
