Generalized Referring Expression Segmentation on GRES Avg.
76.7gIoUQwen25VL-SAMTok (rl)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen25VL-SAMTok (rl)Size=3B, GRPO Training=true2026.01 | 76.7 | 73.7 | 77.1 | |
| Qwen25VL-SAMTok (ft)Size=7B, Finetuned on GRES=true2026.01 | 75.4 | 72.4 | 74.7 | |
| Qwen25VL-SAMTok (ft)Size=3B, Finetuned on GRES=true2026.01 | 74.3 | 71.1 | 72.9 | |
| MLLMSegSize=8B2026.01 | 73.9 | 72.3 | 70.4 | |
| ARGenSegSize=8B2026.01 | 73.6 | 72.1 | — | |
| Qwen3VL-SAMTokSize=4B2026.01 | 73.6 | 71.7 | 65.3 | |
| Qwen25VL-SAMTokSize=7B2026.01 | 73.6 | 70.8 | 68.2 | |
| HiMTokSize=8B2026.01 | 72.4 | 72.4 | — | |
| SAM4MLLMSize=8B2026.01 | 70.5 | 67.8 | 63.3 | |
| Qwen25VL-SAMTokSize=3B2026.01 | 70.1 | 68.9 | 58.2 | |
| LISASize=7B2026.01 | 62.2 | 63.6 | 52.2 |