Motion Hallucination Assessment on MotionHalluc Original
99.36DH (Original) AccuracyLLaVA-OV-1.5-8B + PPV
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LLaVA-OV-1.5-8B + PPVPPV=true, FPS=4, Decoding Strategy=Greedy2026.06 | 99.36 | 0 | 49.67 | 57 | 74.36 | 60.34 | |
| LLaVA-OV-1.5-8BPPV=false, FPS=4, Decoding Strategy=Greedy2026.06 | 98.39 | 1.92 | 50.16 | 53.5 | 41.45 | 48.37 | |
| Gemini-3-FlashPPV=false, FPS=4, Decoding Strategy=Default2026.06 | 91.45 | 38.67 | 65.06 | 83.5 | 53.05 | 67.21 | |
| Qwen3.5-plusPPV=false, FPS=4, Decoding Strategy=Default2026.06 | 90.06 | 38.14 | 64.1 | 81.72 | 68.03 | 71.28 | |
| Qwen3.5-plus + PPVPPV=true, FPS=4, Decoding Strategy=Default2026.06 | 86.96 | 79.7 | 83.33 | 89.94 | 77.74 | 83.67 | |
| Gemini-3-Flash + PPVPPV=true, FPS=4, Decoding Strategy=Default2026.06 | 81.83 | 83.54 | 82.69 | 89.77 | 80.37 | 84.28 | |
| InternVL3.5-8BPPV=false, FPS=4, Decoding Strategy=Greedy2026.06 | 73.07 | 41.34 | 57.21 | 57.33 | 50.31 | 54.95 | |
| InternVL3.5-8B + PPVPPV=true, FPS=4, Decoding Strategy=Greedy2026.06 | 71.79 | 45.19 | 58.49 | 62.5 | 63.29 | 61.42 | |
| Qwen3-VL-8B + PPVPPV=true, FPS=4, Decoding Strategy=Greedy2026.06 | 71.79 | 45.51 | 58.65 | 64 | 76.58 | 66.41 | |
| Qwen3-VL-8BPPV=false, FPS=4, Decoding Strategy=Greedy2026.06 | 61.21 | 48.39 | 54.8 | 66.16 | 62.97 | 61.31 |