Long-form video understanding on VideoMME
61.7Overall ScoreST-SimDiff
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ST-SimDiffBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 61.7 | 73.9 | 59.7 | 51.4 | |
| NVILA-VideoBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=100% (Upper Bound)2026.05 | 61.5 | — | — | — | |
| FasterVLMBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 60.8 | 74.6 | 57.8 | 50.1 | |
| VisionZipBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 60.5 | 72.8 | 58.1 | 50.6 | |
| ST-SimDiffBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 60.2 | 73.3 | 57.7 | 49.7 | |
| FasterVLMBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 60.1 | 72.6 | 56.9 | 51 | |
| FrameFusionBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 59.4 | 72.7 | 57.2 | 48.3 | |
| VisionZipBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 59.1 | 71.4 | 56.2 | 49.7 | |
| FastVBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 58.9 | 71.9 | 58.3 | 49.3 | |
| FrameFusionBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 58.8 | 72.1 | 55.7 | 48.7 | |
| PruMergeBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 58.2 | 71 | 53.7 | 49.9 | |
| FastVBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=30%2026.05 | 57.9 | 69.4 | 55.3 | 48.8 | |
| PruMergeBase Model=NVILA-Video-8B, Input Frames=64, Token Retain Ratio=50%2026.05 | 57.6 | 71.2 | 54.8 | 46.8 |