Long video understanding on LV-Omni-Bench
43.1AccuracyOmniMem
Evaluation Results
| Method | Links | |
|---|---|---|
| OmniMemBackbone=video-SALMONN 2+ (8B), fine-tuned=true2026.05 | 43.1 | |
| OmniMemBackbone=video-SALMONN 2+ (8B), fine-tuned=false2026.05 | 42.5 | |
| OmniMemBackbone=video-SALMONN 2+ (4B), fine-tuned=true2026.05 | 40.7 | |
| StreamMemBackbone=video-SALMONN 2+ (8B)2026.05 | 40.5 | |
| HERMESBackbone=video-SALMONN 2+ (8B)2026.05 | 40.2 | |
| OmniMemBackbone=video-SALMONN 2+ (4B), fine-tuned=false2026.05 | 39.8 | |
| InfiniPot-VBackbone=video-SALMONN 2+ (8B)2026.05 | 39 | |
| UniformBackbone=video-SALMONN 2+ (8B)2026.05 | 38.8 | |
| InfiniPot-VBackbone=video-SALMONN 2+ (4B)2026.05 | 37.5 | |
| StreamMemBackbone=video-SALMONN 2+ (4B)2026.05 | 37.5 | |
| HERMESBackbone=video-SALMONN 2+ (4B)2026.05 | 37.2 | |
| UniformBackbone=video-SALMONN 2+ (4B)2026.05 | 36.6 | |
| OmniMemBackbone=Qwen-2.5-Omni (7B)2026.05 | 34.3 | |
| InfiniPot-VBackbone=Qwen-2.5-Omni (7B)2026.05 | 32.9 | |
| StreamMemBackbone=Qwen-2.5-Omni (7B)2026.05 | 32.8 | |
| HERMESBackbone=Qwen-2.5-Omni (7B)2026.05 | 32.8 | |
| UniformBackbone=Qwen-2.5-Omni (7B)2026.05 | 32.6 |