Visual Grounding on VIEW2SPACE v1
69.34mIoUOurs (Grounded CoT)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ours (Grounded CoT)Category=Fine-tuned, Backbone=Qwen3VL-4B2026.03 | 69.34 | 70.92 | |
| Ours (CoT)Category=Fine-tuned, Backbone=Qwen3VL-4B2026.03 | 49.77 | 52.34 | |
| Ours (Instruct-tuning)Category=Fine-tuned, Backbone=Qwen3VL-4B2026.03 | 47.58 | 50.07 | |
| Qwen3-VL-4BCategory=Open-source MLLMs2026.03 | 16.53 | 18.72 | |
| Ours (Grounded CoT, Direct Answer at Test)Category=Fine-tuned, Backbone=Qwen3VL-4B2026.03 | 16.35 | 11.5 | |
| Qwen3-VL-2BCategory=Open-source MLLMs2026.03 | 11.2 | 5.07 | |
| Qwen3-VL-8BCategory=Open-source MLLMs2026.03 | 10.05 | 7.37 | |
| GPT-5.2Category=Closed-source MLLMs2026.03 | 8.74 | 2.84 | |
| GPT-5Category=Closed-source MLLMs2026.03 | 8.18 | 3.43 | |
| GPT-5-miniCategory=Closed-source MLLMs2026.03 | 7.42 | 2.77 | |
| GPT-4oCategory=Closed-source MLLMs2026.03 | 3.6 | 0.56 | |
| GPT-5-nanoCategory=Closed-source MLLMs2026.03 | 2.89 | 0.87 | |
| DeepSeek-VL2-SmallCategory=Open-source MLLMs2026.03 | 2.86 | 0.19 | |
| Qwen2.5-VL-7BCategory=Open-source MLLMs2026.03 | 2.68 | 0.32 | |
| Gemma-3-12B-itCategory=Open-source MLLMs2026.03 | 2.49 | 0.24 | |
| Random (frequency)Category=Baseline2026.03 | 2.01 | 0.27 | |
| Molmo2-8BCategory=Open-source Spatial Models2026.03 | 1.51 | 0 | |
| Mantis-8B (SigLip)Category=Open-source MLLMs2026.03 | 1.24 | 0.23 | |
| Molmo2-4BCategory=Open-source Spatial Models2026.03 | 1.17 | 0.13 | |
| OpenviewCategory=Open-source Spatial Models2026.03 | 1.08 | 0.3 | |
| Internvl2.5-8BCategory=Open-source MLLMs2026.03 | 0.79 | 0 | |
| Idefics2-8BCategory=Open-source MLLMs2026.03 | 0.39 | 0 | |
| Random (chance)Category=Baseline2026.03 | 0.22 | 0 | |
| MINDCUBE-PlainCategory=Open-source Spatial Models2026.03 | 0.12 | 0 | |
| RoBobrain2.0Category=Open-source Spatial Models2026.03 | 0 | 0 | |
| Spatial-MLLMCategory=Open-source Spatial Models2026.03 | 0 | 0 | |
| MINDCUBE-AugCategory=Open-source Spatial Models2026.03 | 0 | 0 |