Multi-task Video Understanding on MLVU (Single detail, Multi detail, Holistic, Avg)
70.9Average ScoreCausalMem
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| CausalMemFrames=0.5/0.2fps, Tokens=12k2026.06 | 70.9 | 74 | 49.5 | 83.6 | |
| ReKVFrames=0.5fps, Tokens=12k2026.06 | 68.5 | — | — | — | |
| LiveVLMFrames=0.5/0.2fps, Tokens=12k2026.06 | 68.1 | — | — | — | |
| StreamingTOMFrames=0.5/0.2fps, Tokens=12k2026.06 | 67.9 | — | — | — | |
| StreamMemFrames=0.5/0.2fps, Tokens=6k2026.06 | 66.9 | — | — | — | |
| LLaVA-OneVisionFrames=32, Tokens=6k2026.06 | 64.7 | 68.6 | 39.1 | 79.3 |