Spatial VQA on CV-Bench-2D Relation (Level 2)
96.9AccuracyRoboRefer-8B-SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| RoboRefer-8B-SFTInput=RGB-D2026.03 | 96.9 | |
| HiSpatial-3BInput=RGB-XYZ2026.03 | 95.69 | |
| Qwen-3-VL-30B-A3BInput=RGB2026.03 | 95.38 | |
| HiSpatial-3B-RGBInput=RGB2026.03 | 95.08 | |
| Gemini-2.5-ProInput=RGB2026.03 | 93.54 | |
| Qwen-3-VL-8BInput=RGB2026.03 | 92.92 | |
| SpatialRGPT-8BInput=RGB-D2026.03 | 91 | |
| InternVL3.5-14BInput=RGB2026.03 | 88.15 | |
| InternVL3.5-8BInput=RGB2026.03 | 88 | |
| GPT-4oInput=RGB2026.03 | 84.62 | |
| Claude-3.7-SonnetInput=RGB2026.03 | 74.15 | |
| PaliGemma2-3BInput=RGB2026.03 | 72 | |
| SpatialBot-3BInput=RGB-D2026.03 | 69.38 | |
| SpaceLLaVA-13BInput=RGB2026.03 | 63.69 |