Generalized Referring Expression Segmentation on gRefCOCO v1 (val)
76.83cIoUGenSeg-R1-G
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| GenSeg-R1-GTraining Reward=SAM 2-in-the-loop reward2026.02 | 76.83 | — | — | 76.69 | 82.4 | |
| GenSeg-R1-4BSize=4B2026.02 | 75.54 | — | — | 75.58 | 79.05 | |
| Qwen25VL-SAMTok (rl)Size=3B, GRPO Training=true2026.01 | 73.7 | 79.4 | 81.5 | — | — | |
| ARGenSegSize=8B2026.01 | 72.2 | 74.7 | — | — | — | |
| Qwen25VL-SAMTok (ft)Size=7B, Finetuned on GRES=true2026.01 | 72.1 | 78 | 79.5 | — | — | |
| MLLMSegSize=8B2026.01 | 71.6 | 75.1 | 73.2 | — | — | |
| Qwen25VL-SAMTokSize=7B2026.01 | 71.4 | 74.4 | 67.5 | — | — | |
| Qwen25VL-SAMTok (ft)Size=3B, Finetuned on GRES=true2026.01 | 71.2 | 77.1 | 78 | — | — | |
| Qwen3VL-SAMTokSize=4B2026.01 | 71 | 74.7 | 68.8 | — | — | |
| HiMTokSize=8B2026.01 | 70.4 | 72.1 | — | — | — | |
| Seg-R1-7BSize=7B2026.02 | 69.02 | — | — | 69.51 | 0 | |
| Qwen25VL-SAMTokSize=3B2026.01 | 68.2 | 70.5 | 60.5 | — | — | |
| SAM4MLLMSize=8B2026.01 | 67.8 | 71.9 | 66.1 | — | — | |
| GSVALLM Backbone=Llama2-13B, Finetuning=true2023.12 | 66.38 | 70.04 | 66.02 | — | — | |
| Seg-Zero-7BSize=7B2026.02 | 65.9 | — | — | 66.11 | 0 | |
| GSVALLM Backbone=Vicuna-13B, Finetuning=true2023.12 | 64.05 | 68.01 | 65.36 | — | — | |
| LISALLM Backbone=Llama2-13B, Finetuning=true2023.12 | 63.96 | 65.24 | 57.49 | — | — | |
| GSVALLM Backbone=Vicuna-7B, Finetuning=true2023.12 | 63.29 | 66.47 | 62.43 | — | — | |
| LISALLM Backbone=Vicuna-13B, Finetuning=true2023.12 | 62.99 | 63.45 | 55.25 | — | — | |
| ReLA2023.12 | 62.42 | 63.6 | 56.37 | — | — | |
| GSVALLM Backbone=Llama2-13B, Finetuning=false2023.12 | 62.38 | 63.2 | 54.51 | — | — | |
| LISASize=7B2026.01 | 61.8 | 61.6 | 54.7 | — | — | |
| LISALLM Backbone=Vicuna-7B, Finetuning=true2023.12 | 61.76 | 61.63 | 54.67 | — | — | |
| GSVALLM Backbone=Vicuna-7B, Finetuning=false2023.12 | 61.7 | 63.32 | 56.45 | — | — | |
| GSVALLM Backbone=Vicuna-13B, Finetuning=false2023.12 | 60.18 | 62.97 | 58.44 | — | — | |
| LAVT2023.12 | 57.64 | 58.4 | 49.32 | — | — | |
| CRIS2023.12 | 55.34 | 56.27 | — | — | — | |
| VLT2023.12 | 52.51 | 52 | 47.17 | — | — | |
| LTS2023.12 | 52.3 | 52.7 | — | — | — | |
| MattNet2023.12 | 47.51 | 48.24 | 41.15 | — | — | |
| LISALLM Backbone=Vicuna-13B, Finetuning=false2023.12 | 39.85 | 32.73 | 3.66 | — | — | |
| LISALLM Backbone=Llama2-13B, Finetuning=false2023.12 | 39.61 | 33.26 | 3.27 | — | — | |
| LISALLM Backbone=Vicuna-7B, Finetuning=false2023.12 | 38.72 | 32.21 | 2.71 | — | — |