Video Understanding on MLVU 3-120min (dev)
63AccuracyLLaVA-OneVision-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-OneVision-7BBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=100%2026.02 | 63 | |
| DivPruneBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 62.4 | |
| CaCoVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 62.3 | |
| DivPruneBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 62.2 | |
| Qwen2.5-VL-3BBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=100%2026.02 | 62.1 | |
| CaCoVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 62 | |
| DivPruneBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 61.9 | |
| VisionZipBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 61.5 | |
| VisionZipBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 61.5 | |
| PruneVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 61.4 | |
| VisionZipBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 61.2 | |
| CaCoVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 61.2 | |
| FrameFusionBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 61.1 | |
| PruneVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 60.9 | |
| DivPruneBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 60.7 | |
| CaCoVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 60.6 | |
| FrameFusionBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 60.5 | |
| CaCoVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 60.2 | |
| PruneVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 59.9 | |
| CaCoVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 59.9 | |
| PruneVIDBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 59.7 | |
| PruneVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 59.4 | |
| FastVBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=25%2026.02 | 59 | |
| DivPruneBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 58.8 | |
| VisionZipBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 58.5 | |
| FrameFusionBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 58.5 | |
| FrameFusionBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 58.4 | |
| DivPruneBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 58.4 | |
| FastVBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=20%2026.02 | 58.3 | |
| CaCoVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 58.2 | |
| PruneVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 58.1 | |
| PruneVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 58 | |
| VisionZipBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 57.9 | |
| CaCoVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=10%2026.02 | 57.9 | |
| VisionZipBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 57.7 | |
| DivPruneBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 57.7 | |
| VisionZipBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 57.6 | |
| FrameFusionBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 57.3 | |
| VisionZipBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=10%2026.02 | 57.2 | |
| FastVBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=25%2026.02 | 57 | |
| PruneVIDBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=10%2026.02 | 56.6 | |
| FrameFusionBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 56.5 | |
| DivPruneBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=10%2026.02 | 56.4 | |
| FastVBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=20%2026.02 | 55.6 | |
| FastVBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=15%2026.02 | 55.5 | |
| FrameFusionBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 55.4 | |
| FastVBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=15%2026.02 | 53.7 | |
| FastVBase Model=LLaVA-OneVision-7B, Equivalent Retention Ratio=10%2026.02 | 52.9 | |
| FastVBase Model=Qwen2.5-VL-3B, Equivalent Retention Ratio=10%2026.02 | 50.8 |