Geometry reasoning on Geometry3k (test)
48.11Test ScoreDS (Oracle)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DS (Oracle)Backbone=Qwen2.5-VL-7B-Instruct2026.03 | 48.11 | 782 | |
| DPS (Ours)Backbone=Qwen2.5-VL-7B-Instruct2026.03 | 47.78 | 328 | |
| HRBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 46.52 | 328 | |
| USBackbone=Qwen2.5-VL-7B-Instruct2026.03 | 46.22 | 328 | |
| DPS (Ours)Backbone=Qwen2.5-VL-3B-Instruct2026.03 | 44.47 | 492 | |
| DS (Oracle)Backbone=Qwen2.5-VL-3B-Instruct2026.03 | 44.33 | 1,262 | |
| USBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 40.69 | 492 | |
| HRBackbone=Qwen2.5-VL-3B-Instruct2026.03 | 40.44 | 492 |