Real-world cross-view spatial reasoning on OmniSpatial
38.9CL ScorePanoramic + VDrop
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Panoramic + VDropVDrop=✓2026.05 | 38.9 | 45.3 | — | |
| ThinkMorph (24K)VDrop=✗2026.05 | 33.7 | 44.6 | — | |
| Point Matching + VDropVDrop=✓2026.05 | 33.7 | 43.3 | — | |
| PanoramicVDrop=✗2026.05 | 32.9 | 43.9 | — | |
| Top-down + VDropVDrop=✓2026.05 | 32.5 | 44.2 | — | |
| Top-downVDrop=✗2026.05 | 31.4 | 39.8 | — | |
| BAGELVDrop=✗2026.05 | 31.1 | 38.9 | — | |
| Point MatchingVDrop=✗2026.05 | 31 | 41.7 | — | |
| No-ThinkVDrop=✗2026.05 | 30.2 | 44 | — | |
| Text CoTVDrop=✗2026.05 | 29.4 | 35.8 | — | |
| BAGEL-Zebra-CoT (182K)VDrop=✗2026.05 | 29 | 43 | — | |
| Qwen3-VL-4BVDrop=✗2026.05 | 28.6 | 40.1 | — | |
| Qwen3-VL-8BVDrop=✗2026.05 | 25.4 | 45.5 | — |