Visual Reasoning on BLINK (test)
78.23Rel DepthLaViT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LaViTParams=3B2026.01 | 78.23 | 32 | 45.52 | 81.82 | |
| LVRParams=7B2026.01 | 76.61 | 28.7 | 42.5 | 89.5 | |
| Qwen2.5-VLParams=32B2026.01 | 75.81 | 30 | 55.97 | 85.31 | |
| Qwen2.5-VLParams=7B2026.01 | 71.77 | 26 | 38.8 | 87.4 | |
| Naive SFTParams=3B2026.01 | 70.16 | 28 | 34.33 | 81.82 | |
| GPT-4oParams=-2026.01 | 64.52 | 30 | 38.8 | 76.9 | |
| LVR_RLParams=3B2026.01 | 64.52 | 30.7 | 42.54 | 77.62 | |
| Qwen2.5-VL (Baseline)Params=3B2026.01 | 61.29 | 24 | 29.85 | 81.12 | |
| PAPOParams=3B2026.01 | 59.68 | 31.33 | 33.6 | 76.92 | |
| DMLRParams=3B2026.01 | 54.84 | 26 | 23.88 | 72.03 |