Video Understanding on VideoMME (Efficiency Metrics)
3.27Wall-time SpeedupHIPPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| HIPPOBackbone Model=LLaVA-OneVision 7&72B2026.01 | 3.27 | 8.93 | — | — | |
| HIPPOBackbone Model=Video-SALMONN2+ 7&72B2026.01 | 2.85 | 11.82 | — | — | |
| HIPPOBackbone Model=Qwen2.5-VL 7&72B2026.01 | 2.71 | 10.29 | — | — | |
| HIPPOBackbone Model=Qwen3-VL 7&32B2026.01 | 2.38 | 4.67 | — | — | |
| SpecVLMBackbone Model=Video-SALMONN2+ 7&72B2026.01 | 2.24 | 3.46 | — | — | |
| SpecVLMBackbone Model=Qwen2.5-VL 7&72B2026.01 | 2.14 | 3.05 | — | — | |
| SpecVLMBackbone Model=LLaVA-OneVision 7&72B2026.01 | 2.1 | 4.32 | — | — | |
| SD-treeBackbone Model=LLaVA-OneVision 7&72B2026.01 | 2.03 | 4.47 | — | — | |
| IVSBTemperature=0, Visual Token Length=25K2026.02 | 2.02 | — | 3.98 | 2.96 | |
| SpecVLMBackbone Model=Qwen3-VL 7&32B2026.01 | 1.89 | 3.87 | — | — | |
| Vanilla SDBackbone Model=LLaVA-OneVision 7&72B2026.01 | 1.88 | 4.01 | — | — | |
| SD-treeBackbone Model=Video-SALMONN2+ 7&72B2026.01 | 1.87 | 3.71 | — | — | |
| SD-treeBackbone Model=Qwen2.5-VL 7&72B2026.01 | 1.73 | 3.32 | — | — | |
| SD-treeBackbone Model=Qwen3-VL 7&32B2026.01 | 1.73 | 3.91 | — | — | |
| IVSBTemperature=1, Visual Token Length=25K2026.02 | 1.55 | — | 2.57 | 1.91 | |
| MSDTemperature=0, Visual Token Length=25K2026.02 | 1.49 | — | 2.68 | 1.77 | |
| Vanilla SDBackbone Model=Qwen2.5-VL 7&72B2026.01 | 1.48 | 2.7 | — | — | |
| Vanilla SDBackbone Model=Video-SALMONN2+ 7&72B2026.01 | 1.36 | 2.9 | — | — | |
| Vanilla SDBackbone Model=Qwen3-VL 7&32B2026.01 | 1.35 | 3.39 | — | — | |
| MSDTemperature=1, Visual Token Length=25K2026.02 | 1.22 | — | 2.05 | 1.35 | |
| SpecVLMTemperature=0, Visual Token Length=25K2026.02 | 1.21 | — | 3.91 | 1.34 |