Generalized Referring Expression Segmentation on gRefCOCO v1 (test B)
72.4gIoUARGenSeg
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ARGenSegSize=8B2026.01 | 72.4 | 70.4 | — | |
| HiMTokSize=8B2026.01 | 71.7 | 72 | — | |
| Qwen25VL-SAMTok (rl)Size=3B, GRPO Training=true2026.01 | 71.3 | 69.8 | 72 | |
| MLLMSegSize=8B2026.01 | 69.7 | 68.5 | 65.5 | |
| Qwen25VL-SAMTok (ft)Size=7B, Finetuned on GRES=true2026.01 | 69.6 | 67.7 | 68.6 | |
| Qwen3VL-SAMTokSize=4B2026.01 | 69.1 | 68.3 | 60.8 | |
| Qwen25VL-SAMTokSize=7B2026.01 | 68.5 | 68.2 | 59.8 | |
| Qwen25VL-SAMTok (ft)Size=3B, Finetuned on GRES=true2026.01 | 68.3 | 66.5 | 67.2 | |
| Qwen25VL-SAMTokSize=3B2026.01 | 65.9 | 65.4 | 55.3 | |
| GSVALLM Backbone=Llama2-13B, Finetuning=true2023.12 | 65.45 | 63.2 | 62.47 | |
| SAM4MLLMSize=8B2026.01 | 65.3 | 63.4 | 60 | |
| GSVALLM Backbone=Vicuna-13B, Finetuning=true2023.12 | 63.83 | 61.28 | 63.11 | |
| GSVALLM Backbone=Vicuna-7B, Finetuning=true2023.12 | 62.23 | 60.47 | 60.56 | |
| LISALLM Backbone=Llama2-13B, Finetuning=true2023.12 | 62.11 | 62.29 | 56.34 | |
| GSVALLM Backbone=Llama2-13B, Finetuning=false2023.12 | 62.06 | 60.77 | 58.3 | |
| LISALLM Backbone=Vicuna-13B, Finetuning=true2023.12 | 61.84 | 62.24 | 56.15 | |
| GSVALLM Backbone=Vicuna-7B, Finetuning=false2023.12 | 61.34 | 60.26 | 58.42 | |
| ReLA2023.12 | 61.02 | 59.88 | 58.4 | |
| LISALLM Backbone=Vicuna-7B, Finetuning=true2023.12 | 58.84 | 60.63 | 51.91 | |
| LISASize=7B2026.01 | 58.8 | 60.6 | 51.9 | |
| GSVALLM Backbone=Vicuna-13B, Finetuning=false2023.12 | 58.06 | 57.28 | 52.22 | |
| LAVT2023.12 | 55.83 | 55.04 | 48.46 | |
| CRIS2023.12 | 51.79 | 51.04 | — | |
| VLT2023.12 | 50.88 | 50.52 | 47.82 | |
| LTS2023.12 | 50.42 | 49.96 | — | |
| MattNet2023.12 | 46.14 | 45.33 | 41.32 | |
| LISALLM Backbone=Llama2-13B, Finetuning=false2023.12 | 40.49 | 45.41 | 5.73 | |
| LISALLM Backbone=Vicuna-7B, Finetuning=false2023.12 | 39.65 | 44.79 | 5 | |
| LISALLM Backbone=Vicuna-13B, Finetuning=false2023.12 | 39.49 | 45.35 | 4.41 |