Referring Expression Segmentation on ReasonSeg (val)
66.3mIoUVisionReasoner
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VisionReasonerTraining Mask Dec.=false, Learning Protocol=Reinforcement Learning2025.12 | 66.3 | — | |
| GETok-R1Training Mask Dec.=false, Learning Protocol=Reinforcement Learning2025.12 | 65.9 | — | |
| GETok-R1-gridTraining Mask Dec.=false, Learning Protocol=Reinforcement Learning2025.12 | 64.2 | — | |
| SAM-R1Training Mask Dec.=false, Learning Protocol=Reinforcement Learning2025.12 | 64 | — | |
| Seg-ZeroTraining Mask Dec.=false, Learning Protocol=Reinforcement Learning2025.12 | 62.6 | — | |
| GETok-SFTTraining Mask Dec.=false, Learning Protocol=Supervised Fine-Tuning2025.12 | 59.2 | — | |
| GETok-SFT-gridTraining Mask Dec.=false, Learning Protocol=Supervised Fine-Tuning2025.12 | 58.1 | — | |
| Qwen2.5-VL-7BTraining Mask Dec.=false, Learning Protocol=Supervised Fine-Tuning2025.12 | 55.4 | — | |
| LISATraining Mask Dec.=true, Learning Protocol=Supervised Fine-Tuning2025.12 | 44.4 | — | |
| SAM3 Agent2026.04 | — | 77 | |
| Vision Banana2026.04 | — | 79.3 |