AI-Generated Video Detection on ViF-Bench I2V 1.0 (test)
91.02AccuracyOurs-RL(7B)
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Ours-RL(7B)Model Size=7B, Training Protocol=RL2025.12 | 91.02 | 88.35 | 90.27 | 92.12 | 96.36 | 95.94 | 96.05 | 91.02 | |
| Ours-SFT(7B)Model Size=7B, Training Protocol=SFT2025.12 | 90.11 | 84.65 | 88.76 | 90.3 | 96.67 | 96.25 | 97.37 | 90.11 | |
| AIGVDetTraining Protocol=Binary Classifier2025.12 | 69.08 | 44.88 | 56.76 | 67.58 | 71.52 | 65.31 | 73.68 | 69.08 | |
| DeMambaTraining Protocol=Binary Classifier2025.12 | 64.29 | 96.66 | 73 | 65.76 | 65.76 | 64.06 | 66.78 | 64.29 | |
| Video-LLaMa-3(7B)Model Size=7B, Training Protocol=Off-the-shelf2025.12 | 50.91 | 2.39 | 4.57 | 51.84 | 50 | 50.31 | 49.67 | 50.91 | |
| Qwen-2.5-VL(3B)Model Size=3B, Training Protocol=Off-the-shelf2025.12 | 50.7 | 2.62 | 4.94 | 50.61 | 50.61 | 52.5 | 50.33 | 50.7 | |
| NSG-VDTraining Protocol=Binary Classifier2025.12 | 49.65 | 98.66 | 66.21 | 50.3 | 50 | 49.06 | 48.68 | 49.65 |