Video Question Answering on PVRBench
57Lighting AccuracyGemini-1.5-Pro
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Gemini-1.5-ProSize=–, #Fr=322026.03 | 57 | 52 | 54 | 55 | 55 | |
| ROVASize=72B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 57 | 53 | 56 | 56 | 56 | |
| GPT-4oSize=–, #Fr=322026.03 | 54 | 47 | 50 | 52 | 51 | |
| ROVASize=13B, #Fr=32, Backbone=Qwen2-VL2026.03 | 53 | 49 | 52 | 54 | 52 | |
| ROVASize=7B, #Fr=32, Backbone=Embodied-R2026.03 | 52 | 46 | 49 | 51 | 50 | |
| Video-R1Size=72B, #Fr=322026.03 | 51 | 45 | 49 | 49 | 49 | |
| ROVASize=7B, #Fr=32, Backbone=Qwen2.5-VL2026.03 | 48 | 43 | 47 | 49 | 47 | |
| Qwen2.5-VLSize=72B, #Fr=322026.03 | 48 | 41 | 44 | 47 | 45 | |
| Claude-3.5-Son.Size=–, #Fr=322026.03 | 45 | 41 | 44 | 45 | 44 | |
| Embodied-RSize=7B, #Fr=322026.03 | 45 | 38 | 42 | 43 | 42 | |
| Video-R1Size=7B, #Fr=322026.03 | 43 | 37 | 42 | 41 | 41 | |
| ROVASize=8B, #Fr=32, Backbone=InternVL2.52026.03 | 43 | 36 | 41 | 40 | 40 | |
| Qwen2-VLSize=13B, #Fr=322026.03 | 43 | 35 | 39 | 42 | 40 | |
| LLaVA-Video-RSize=7B, #Fr=322026.03 | 40 | 34 | 38 | 38 | 38 | |
| VideoChat-RSize=7B, #Fr=162026.03 | 36 | 31 | 36 | 35 | 35 | |
| Qwen2.5-VLSize=7B, #Fr=322026.03 | 35 | 28 | 34 | 34 | 33 | |
| MiniCPM-V 2.6Size=8B, #Fr=642026.03 | 34 | 28 | 31 | 32 | 31 | |
| LLaVA-VideoSize=7B, #Fr=322026.03 | 32 | 29 | 30 | 32 | 31 | |
| InternVL2.5Size=8B, #Fr=322026.03 | 31 | 26 | 32 | 33 | 31 | |
| VideoLLaMA2Size=7B, #Fr=162026.03 | 28 | 25 | 27 | 29 | 27 | |
| VideoChat2Size=7B, #Fr=162026.03 | 26 | 23 | 25 | 27 | 25 |