Perspective-aware spatial reasoning on COMFORT Visual Illusions
95.38Directional Accuracy (Left/Right)SymPL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SymPL2026.02 | 95.38 | 100 | 100 | |
| APC-Num2026.02 | 84.31 | 27.88 | 23 | |
| APC-Vis2026.02 | 76.75 | 36.25 | 42.25 | |
| LLaVa-OneVision2026.02 | 72 | 95.88 | 98.75 | |
| LLaVa-NeXT2026.02 | 58.19 | 58.75 | 89.13 | |
| Qwen2.5-VLPrompting=SCAFFOLD2026.02 | 58.06 | 85 | 68.5 | |
| Qwen2.5-VL2026.02 | 57 | 82.63 | 47.25 | |
| Gemini-2.5-Flash2026.02 | 54.63 | 55.5 | 44 | |
| SpatialBot2026.02 | 52.25 | 78.25 | 87 | |
| SD-VLM2026.02 | 50 | 49.13 | 46.75 | |
| Random2026.02 | 49.69 | 51.75 | 49.13 | |
| Qwen2.5-VLPrompting=SoM2026.02 | 46.56 | 44.25 | 46.13 | |
| Molmo2026.02 | 44.56 | 93.88 | 74.5 | |
| SpatialVLM2026.02 | 44 | 92.13 | 80.75 | |
| GPT-52026.02 | 43.63 | 99.75 | 89.13 | |
| SpatialRGPT2026.02 | 42.31 | 64.88 | 61.38 | |
| Qwen2.5-VLPrompting=CoT2026.02 | 40.94 | 97.25 | 23.5 | |
| Cambrian-12026.02 | 35.06 | 79.25 | 87.25 | |
| SAT2026.02 | 29.44 | 49.63 | 47.63 |