Quantitative Spatial Visual Question Answering on QSpatial-Bench
84.17ScanNet ScoreHiSpatial-3B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HiSpatial-3BInput=RGB-XYZ2026.03 | 84.17 | 88.12 | 85.16 | |
| HiSpatial-3B*Input=RGB-XYZ (GT)2026.03 | 84.12 | — | — | |
| HiSpatial-3B-RGBInput=RGB2026.03 | 75.88 | 76.24 | 76.01 | |
| GPT-5Input=RGB2026.03 | 73.53 | 63.37 | 68.45 | |
| GPT-4oInput=RGB2026.03 | 69.41 | 61.06 | 66.3 | |
| InternVL3.5-8BInput=RGB2026.03 | 57.06 | 59.41 | 57.94 | |
| Qwen3VL-8BInput=RGB2026.03 | 55.88 | 35.64 | 48.34 | |
| Gemini-2.5-ProInput=RGB2026.03 | 55.46 | 40.59 | 49.92 | |
| RoboRefer-8B-SFTInput=RGB-D2026.03 | 31.07 | 20.79 | 27.24 | |
| PaliGemma2-3BInput=RGB2026.03 | 30.06 | 36.63 | 32.84 |