Language-guided Segmentation on Various-LangSeg
83Explicit Semantic gIoUSeg-Agent-7B
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Seg-Agent-7BTraining Protocol=training-free, Size=7B2026.05 | 83 | 83.7 | 41 | 42.1 | 75.2 | 66.7 | 70.6 | 68.5 | |
| LISA-13BTraining Protocol=training-based, Size=13B2026.05 | 82.8 | 83.9 | 35.3 | 40.3 | 54.4 | 47.9 | 65.8 | 67.9 | |
| Seg-Agent-3BTraining Protocol=training-free, Size=3B2026.05 | 82.3 | 81.5 | 40.8 | 31.5 | 66.1 | 62.6 | 68.8 | 60.9 | |
| SAM4MLLMTraining Protocol=training-based2026.05 | 82.1 | 83.5 | 32 | 31.8 | 45.2 | 37.1 | 62.9 | 62.3 | |
| LISA-7BTraining Protocol=training-based, Size=7B2026.05 | 81.9 | 83.1 | 32.4 | 30.8 | 46.8 | 36.7 | 63.2 | 64.2 | |
| Seg-Zero-7BTraining Protocol=training-based, Size=7B2026.05 | 81.8 | 81 | 41.4 | 43.5 | 74.5 | 67.1 | 70 | 68.9 | |
| PixelLLM-7BTraining Protocol=training-based, Size=7B2026.05 | 81.5 | 83.5 | 31.2 | 31.5 | 45.2 | 40.1 | 62.3 | 62.7 | |
| Qwen2.5-VL-7B + SAM2-LTraining Protocol=training-free2026.05 | 80.8 | 79.7 | 39.5 | 38.3 | 70.1 | 58.9 | 68.2 | 65.4 | |
| Qwen2.5-VL-3B + SAM2-LTraining Protocol=training-free2026.05 | 80.1 | 77.4 | 33.3 | 33.7 | 61 | 49.5 | 64.7 | 61.3 | |
| OVSeg*Training Protocol=training-based2026.05 | 77.1 | 76 | 23.2 | 23 | 25 | 22.1 | 54.4 | 53.2 | |
| ReLA*Training Protocol=training-based2026.05 | 76.8 | 77.7 | 20.1 | 22.1 | 25.2 | 21.2 | 53.4 | 53.8 |