Reasoning Segmentation on FReasonSeg 1.0 (L2)
82.7gIoUCR-Seg
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CR-SegTrain=SFT/LoRA2026.06 | 82.7 | 79.74 | |
| LENSTrain=RL/Full FT2026.06 | 79.05 | 78.67 | |
| Dr. SegTrain=RL/Full FT2026.06 | 78.93 | 74.79 | |
| Seg-Zero-7BTrain=RL/Full FT2026.06 | 77.22 | 73.85 | |
| Dr. Seg (LoRA)Train=RL/LoRA2026.06 | 76.37 | 73.53 | |
| Evol-SAM3Train=Agent2026.06 | 74.72 | 73.13 | |
| SAM3-Agent (reproduced)Train=Agent, Backbone=Qwen3-VL-4B and SAM32026.06 | 72.96 | 68.54 | |
| VisionReasonerTrain=RL/Full FT2026.06 | 72.92 | 68.93 | |
| GSVATrain=SFT/LoRA2026.06 | 72.82 | 70.75 | |
| Evol-SAM3 (reproduced)Train=Agent, Backbone=Qwen3-VL-4B and SAM32026.06 | 65.43 | 60.22 | |
| SAM3-AgentTrain=Agent2026.06 | 64.47 | 63.45 | |
| LISA-7BTrain=SFT/LoRA2026.06 | 54.6 | 47.49 |