Generalized Referring Image Segmentation on gRefCOCO (testB)
62.81gIoUSSP-SAM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SSP-SAMModel architecture=SAM based, Input resolution=3362026.03 | 62.81 | 60.91 | 57.19 | |
| GSVA-Vicuna-7BModel architecture=SAM based, Fine-tuning=true2026.03 | 62.23 | 60.47 | 60.56 | |
| SSP-SAMModel architecture=SAM based, Input resolution=2242026.03 | 61.55 | 58.24 | 61.36 | |
| GSVA-Vicuna-7BModel architecture=SAM based, Fine-tuning=false2026.03 | 61.34 | 60.26 | 58.42 | |
| CGFormerModel architecture=Non-SAM based2026.03 | 61.09 | 60.18 | — | |
| ReLAModel architecture=Non-SAM based2026.03 | 61.02 | 59.88 | 58.4 | |
| LISA-Vicuna-7BModel architecture=SAM based, Fine-tuning=true2026.03 | 58.84 | 60.63 | 51.91 | |
| LAVTModel architecture=Non-SAM based2026.03 | 55.83 | 55.04 | 48.46 | |
| CRISModel architecture=Non-SAM based2026.03 | 51.79 | 51.04 | — | |
| VLTModel architecture=Non-SAM based2026.03 | 50.88 | 50.52 | 47.82 | |
| LTSModel architecture=Non-SAM based2026.03 | 50.42 | 49.96 | — | |
| MattNetModel architecture=Non-SAM based2026.03 | 46.14 | 45.33 | 41.32 | |
| LISA-Vicuna-7BModel architecture=SAM based, Fine-tuning=false2026.03 | 39.65 | 44.79 | 5 |