Video Reasoning Quality Assessment on UrbanVideo and VSI-Bench Clean
3.91Original ScoreGemini-1.5-Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-1.5-ProSize=–, #Fr=322026.03 | 3.91 | |
| GPT-4oSize=–, #Fr=322026.03 | 3.82 | |
| ROVASize=72B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 3.72 | |
| Video-R1Size=72B, #Fr=322026.03 | 3.68 | |
| Claude-3.5-Son.Size=–, #Fr=322026.03 | 3.65 | |
| Qwen2.5-VLSize=72B, #Fr=322026.03 | 3.64 | |
| ROVASize=13B, #Fr=32, Backbone=Qwen2-VL2026.03 | 3.62 | |
| ROVASize=7B, #Fr=32, Backbone=Embodied-R2026.03 | 3.58 | |
| ROVASize=7B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 3.52 | |
| Qwen2-VLSize=13B, #Fr=322026.03 | 3.48 | |
| Embodied-RSize=7B, #Fr=322026.03 | 3.45 | |
| Video-R1Size=7B, #Fr=322026.03 | 3.42 | |
| Qwen2.5-VLSize=7B, #Fr=322026.03 | 3.41 | |
| LLaVA-Video-RSize=7B, #Fr=322026.03 | 3.32 | |
| ROVASize=8B, #Fr=32, Backbone=InternVL2.52026.03 | 3.28 | |
| VideoChat-RSize=7B, #Fr=162026.03 | 3.25 | |
| MiniCPM-V 2.6Size=8B, #Fr=642026.03 | 3.18 | |
| InternVL2.5Size=8B, #Fr=322026.03 | 3.15 | |
| LLaVA-VideoSize=7B, #Fr=322026.03 | 3.12 | |
| VideoLLaMA2Size=7B, #Fr=162026.03 | 3.01 | |
| VideoChat2Size=7B, #Fr=162026.03 | 2.88 |