Multi-view Spatial Reasoning on BLINK Multi-View
54.1AccuracyGPT-4o-0513
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4o-05132025.11 | 54.1 | |
| SpatialThinker-7B (Ours)Training Data=SpatialThinkerVQA2025.11 | 52.6 | |
| Claude-3.5-Sonnet-06202025.11 | 51.9 | |
| Qwen2.5-VL-7B + Vanilla GRPOTraining Data=SpatialThinkerVQA, Training Protocol=Vanilla GRPO2025.11 | 51.9 | |
| VLAA-Thinker-7B2025.11 | 51.1 | |
| SpaceThinker2025.11 | 50.4 | |
| SpaceOm2025.11 | 48.9 | |
| Qwen2.5-VL-3B + Vanilla GRPOTraining Data=SpatialThinkerVQA, Training Protocol=Vanilla GRPO2025.11 | 45.9 | |
| SpatialThinker-3B (Ours)Training Data=SpatialThinkerVQA2025.11 | 45.1 | |
| Qwen2.5-VL-7B2025.11 | 44.4 | |
| Qwen2.5-VL-7B + SFTTraining Data=SpatialThinkerVQA, Training Protocol=SFT2025.11 | 44.4 | |
| Qwen2.5-VL-3B2025.11 | 42.9 | |
| Qwen2.5-VL-3B + SFTTraining Data=SpatialThinkerVQA, Training Protocol=SFT2025.11 | 42.1 |