Video Understanding on Video-MME (Context Inclusion & Length Segmentation)
62.6Performance (w/o Context)LLaVA-Video-7B-qwen2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVA-Video-7B-qwen2Reduction Ratio=100%, Frames=322026.02 | 62.6 | 62.4 | 51.8 | 100 | |
| FSRReduction Ratio=50%, Frames=322026.02 | 61.9 | 62 | 51.6 | 100.3 | |
| HoloVReduction Ratio=50%, Frames=322026.02 | 61.7 | 61.6 | 51.3 | 99.2 | |
| FSRReduction Ratio=60%, Frames=322026.02 | 61.6 | 61.5 | 52.2 | 99.6 | |
| HoloVReduction Ratio=60%, Frames=322026.02 | 61.4 | 61 | 51.3 | 98.5 | |
| FSRReduction Ratio=70%, Frames=322026.02 | 61.3 | 61.5 | 52 | 98.9 | |
| HoloVReduction Ratio=70%, Frames=322026.02 | 60.6 | 61.2 | 51.2 | 98.2 | |
| HoloVReduction Ratio=80%, Frames=322026.02 | 60.4 | 60.8 | 51.6 | 98 | |
| FSRReduction Ratio=80%, Frames=322026.02 | 60.2 | 60.9 | 52.3 | 98.2 |