Video Reasoning Quality Assessment on PVRBench
1.72Frame ConsistencyGemini-1.5-Pro
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemini-1.5-ProSize=–, #Fr=322026.03 | 1.72 | 3.61 | 3.48 | 3.58 | 3.41 | 3.52 | |
| GPT-4oSize=–, #Fr=322026.03 | 1.85 | 3.42 | 3.55 | 3.38 | 3.21 | 3.39 | |
| ROVASize=72B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 1.95 | 3.45 | 3.35 | 3.42 | 3.18 | 3.35 | |
| Claude-3.5-Son.Size=–, #Fr=322026.03 | 2.08 | 3.18 | 3.22 | 2.95 | 3.15 | 3.13 | |
| Video-R1Size=72B, #Fr=322026.03 | 2.11 | 3.25 | 3.18 | 3.21 | 2.98 | 3.16 | |
| ROVASize=13B, #Fr=32, Backbone=Qwen2-VL2026.03 | 2.12 | 3.28 | 3.32 | 3.18 | 3.05 | 3.21 | |
| Qwen2.5-VLSize=72B, #Fr=322026.03 | 2.18 | 3.15 | 3.08 | 2.92 | 3.12 | 3.07 | |
| ROVASize=7B, #Fr=32, Backbone=Embodied-R2026.03 | 2.25 | 3.15 | 3.18 | 3.22 | 2.91 | 3.12 | |
| ROVASize=7B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 2.31 | 3.05 | 3.08 | 2.98 | 2.85 | 2.99 | |
| Qwen2-VLSize=13B, #Fr=322026.03 | 2.41 | 2.85 | 2.92 | 2.78 | 2.72 | 2.82 | |
| Embodied-RSize=7B, #Fr=322026.03 | 2.45 | 2.82 | 2.91 | 2.72 | 2.68 | 2.78 | |
| ROVASize=8B, #Fr=32, Backbone=InternVL2.52026.03 | 2.45 | 2.82 | 2.75 | 2.78 | 2.58 | 2.73 | |
| Video-R1Size=7B, #Fr=322026.03 | 2.48 | 2.75 | 2.85 | 2.68 | 2.65 | 2.73 | |
| LLaVA-Video-RSize=7B, #Fr=322026.03 | 2.58 | 2.68 | 2.61 | 2.78 | 2.42 | 2.62 | |
| VideoChat-RSize=7B, #Fr=162026.03 | 2.65 | 2.62 | 2.55 | 2.71 | 2.28 | 2.54 | |
| Qwen2.5-VLSize=7B, #Fr=322026.03 | 2.71 | 2.58 | 2.62 | 2.68 | 2.31 | 2.55 | |
| MiniCPM-V 2.6Size=8B, #Fr=642026.03 | 2.75 | 2.48 | 2.42 | 2.55 | 2.21 | 2.42 | |
| LLaVA-VideoSize=7B, #Fr=322026.03 | 2.78 | 2.45 | 2.35 | 2.52 | 2.25 | 2.39 | |
| InternVL2.5Size=8B, #Fr=322026.03 | 2.85 | 2.38 | 2.28 | 2.42 | 2.18 | 2.32 | |
| VideoLLaMA2Size=7B, #Fr=162026.03 | 2.92 | 2.18 | 2.25 | 2.12 | 2.15 | 2.18 | |
| VideoChat2Size=7B, #Fr=162026.03 | 3.01 | 2.08 | 2.15 | 2.05 | 2.02 | 2.08 |