Reasoning-driven segmentation on GeoSeg-Bench
3.64Faithfulness (Qwen-8B)GeoSeg
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| GeoSegCategory=Ours2026.03 | 3.64 | 3.48 | 3.66 | 3.72 | 3.28 | 3.28 | 4.35 | 4.12 | 4.2 | |
| LISA-7BCategory=Reasoning segmentation2026.03 | 2.77 | 2.65 | 2.83 | 2.97 | 2.51 | 2.7 | 3.85 | 3.45 | 3.65 | |
| CLIP SurgeryCategory=Generalist segmentation2026.03 | 1.98 | 1.91 | 2.02 | 2.15 | 1.88 | 2.05 | 2.86 | 2.51 | 2.78 | |
| SAM3Category=Generalist segmentation2026.03 | 1.78 | 1.72 | 1.81 | 2.03 | 1.74 | 2 | 2.75 | 2.3 | 2.68 | |
| Grounded SAMCategory=Generalist segmentation2026.03 | 1.71 | 1.65 | 1.75 | 1.92 | 1.64 | 1.86 | 2.63 | 2.17 | 2.5 | |
| LISA++Category=Reasoning segmentation2026.03 | 1.57 | 1.54 | 1.58 | 1.62 | 1.49 | 1.53 | 2.22 | 2.03 | 2.07 | |
| GSVACategory=Reasoning segmentation2026.03 | 1.32 | 1.3 | 1.35 | 1.42 | 1.3 | 1.39 | 1.96 | 1.74 | 1.84 | |
| BAGELCategory=Open-source MLLMs2026.03 | 1.31 | 1.29 | 1.34 | 1.43 | 1.31 | 1.42 | 2.2 | 1.95 | 2.15 | |
| UniWorldCategory=Open-source MLLMs2026.03 | 1.22 | 1.21 | 1.24 | 1.35 | 1.23 | 1.32 | 1.84 | 1.67 | 1.75 | |
| Qwen-EditCategory=Open-source MLLMs2026.03 | 1.21 | 1.2 | 1.22 | 1.32 | 1.21 | 1.26 | 1.75 | 1.61 | 1.66 | |
| OmniGen2Category=Open-source MLLMs2026.03 | 1.15 | 1.14 | 1.17 | 1.21 | 1.15 | 1.22 | 1.65 | 1.54 | 1.63 | |
| HunyuanCategory=Open-source MLLMs2026.03 | 1.06 | 1.06 | 1.06 | 1.1 | 1.07 | 1.11 | 1.53 | 1.46 | 1.5 | |
| PixelLMCategory=Reasoning segmentation2026.03 | 1.04 | 1.04 | 1.04 | 1.05 | 1.03 | 1.04 | 1.45 | 1.4 | 1.44 | |
| GLAMMCategory=Reasoning segmentation2026.03 | 1 | 1 | 1 | 1 | 1 | 1 | 1.38 | 1.3 | 1.38 |