Change Referring Segmentation on MTRefSeg-21K NS (val)
61.55mIoUGSVA-7B
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| GSVA-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 61.55 | — | — | 66.93 | 58.42 | 50.5 | 43.17 | 26.53 | |
| FIANetVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 61.32 | — | — | 70.13 | 62.16 | 52.23 | 38.15 | 13.48 | |
| RMSINVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 60.1 | — | — | 68.67 | 60.75 | 51.53 | 37.04 | 12.93 | |
| GLAMM-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 60.08 | — | — | 64.16 | 57.43 | 50.3 | 40.4 | 25.94 | |
| GSVA-13BVision Encoder=CLIP-L, Text Encoder=Vicuna-13B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 59.91 | — | — | 62.38 | 56.24 | 50.3 | 43.17 | 28.71 | |
| MTRefSeg-R1Vision Encoder=Swin-B, Text Encoder=Phi-1.5, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 59.87 | — | — | 66.72 | 61.15 | 53.93 | 43.01 | 23.26 | |
| LISA-13BVision Encoder=CLIP-L, Text Encoder=Vicuna-13B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 58.1 | — | — | 60 | 55.45 | 49.31 | 40.79 | 28.12 | |
| LGCEVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 56.78 | — | — | 64.46 | 56.04 | 45.61 | 32.38 | 11.53 | |
| LAVTVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 56.46 | — | — | 63.46 | 54.89 | 45.31 | 31.78 | 12.03 | |
| LISA-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 55.59 | — | — | 58.02 | 52.67 | 44.55 | 36.83 | 24.36 | |
| RefSegformerVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 55.49 | — | — | 62.86 | 54.04 | 43.81 | 29.62 | 9.22 | |
| SegEarth-R1Vision Encoder=Swin-B, Text Encoder=Phi-1.5, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 54.95 | — | — | 60.37 | 54.11 | 47.44 | 37.07 | 19.04 | |
| RSRefSegVision Encoder=Swin-B, Text Encoder=BERT, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 50.02 | — | — | 54.59 | 46.72 | 37.84 | 26.82 | 14.34 | |
| UniChangeVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 41.51 | — | — | 41.39 | 35.84 | 27.72 | 19.6 | 10.89 | |
| CRIS-101Vision Encoder=ResNet-101, Text Encoder=ResNet-101, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 35.81 | — | — | 36.39 | 26.77 | 14.69 | 5.41 | 0.65 | |
| CRIS-50Vision Encoder=ResNet-50, Text Encoder=ResNet-50, Protocol=Specialist Method, Trained on MTRefSeg-21K2026.05 | 35.24 | — | — | 35.04 | 25.81 | 14.99 | 6.72 | 1.3 | |
| GSVA-13BVision Encoder=CLIP-L, Text Encoder=Vicuna-13B, Protocol=Without Fine-tuning2026.05 | 22.74 | — | — | 18.42 | 14.65 | 9.7 | 5.94 | 0.59 | |
| GeoPixel-7BVision Encoder=CLIP-L, Text Encoder=InternLM-XComposer2.5, Protocol=Fine-tuned on MTRefSeg-21K2026.05 | 17.47 | — | — | 10.18 | 6.37 | 3.61 | 1.9 | 0.55 | |
| GSVA-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Without Fine-tuning2026.05 | 16.6 | — | — | 11.29 | 7.13 | 5.94 | 4.36 | 1.78 | |
| UniGeoSegVision Encoder=Swin-B, Text Encoder=Phi-1.5, Protocol=Without Fine-tuning2026.05 | 14.67 | — | — | 12.63 | 10.48 | 7.67 | 5.36 | 2.71 | |
| GeoPixel-7BVision Encoder=CLIP-L, Text Encoder=InternLM-XComposer2.5, Protocol=Without Fine-tuning2026.05 | 12.64 | — | — | 6.42 | 3.86 | 1.85 | 0.85 | 0.2 | |
| GLAMM-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Without Fine-tuning2026.05 | 10.07 | — | — | 7.92 | 5.15 | 2.97 | 1.19 | 0.99 | |
| SegEarth-R1Vision Encoder=Swin-B, Text Encoder=Phi-1.5, Protocol=Without Fine-tuning2026.05 | 5.77 | — | — | 2.81 | 2.01 | 1.1 | 0.8 | 0.3 | |
| LISA-7BVision Encoder=CLIP-L, Text Encoder=Vicuna-7B, Protocol=Without Fine-tuning2026.05 | 4.88 | — | — | 1.78 | 0.79 | 0.4 | 0.2 | 0 | |
| LISA-13BVision Encoder=CLIP-L, Text Encoder=Vicuna-13B, Protocol=Without Fine-tuning2026.05 | 2.58 | — | — | 0.99 | 0.59 | 0.4 | 0.4 | 0 |