Reasoning Segmentation on FReasonSeg 1.0 (Overall)
79.85gIoUCR-Seg
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CR-SegTrain=SFT/LoRA2026.06 | 79.85 | 74.56 | |
| Dr. Seg (LoRA)Train=RL/LoRA2026.06 | 76.82 | 72.09 | |
| Dr. SegTrain=RL/Full FT2026.06 | 76.51 | 70.73 | |
| Seg-Zero-7BTrain=RL/Full FT2026.06 | 76 | 69.78 | |
| LENSTrain=RL/Full FT2026.06 | 75.03 | 72.66 | |
| VisionReasonerTrain=RL/Full FT2026.06 | 73.29 | 66.23 | |
| Evol-SAM3Train=Agent2026.06 | 69.84 | 66.96 | |
| SAM3-Agent (reproduced)Train=Agent, Backbone=Qwen3-VL-4B and SAM32026.06 | 68.02 | 63.34 | |
| GSVATrain=SFT/LoRA2026.06 | 67.23 | 62.84 | |
| LISA-7BTrain=SFT/LoRA2026.06 | 61.16 | 55.72 | |
| Evol-SAM3 (reproduced)Train=Agent, Backbone=Qwen3-VL-4B and SAM32026.06 | 58.4 | 50.44 | |
| SAM3-AgentTrain=Agent2026.06 | 57.86 | 54.49 |