Referring Temporal Change Segmentation on MTRefSeg-21K v1
65.68mIoUMTRefSeg-R1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MTRefSeg-R1Vision Encoder=Swin-B, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 65.68 | 65.17 | 65.17 | 71.65 | |
| FIANetVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 64.73 | 67.87 | 67.87 | 70.07 | |
| RMSINVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 64.35 | 66.07 | 66.07 | 71.5 | |
| LGCEVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 63.27 | 66.22 | 66.22 | 69.87 | |
| LAVTVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 62.6 | 66.02 | 66.02 | 68.94 | |
| RefSegformerVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 61.37 | 64.87 | 64.87 | 68.08 | |
| SegEarth-R1Vision Encoder=Swin-B, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 59.6 | 59.99 | 59.99 | 64.98 | |
| RSRefSegVision Encoder=Swin-B, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 59.08 | 57.59 | 57.59 | 64.02 | |
| GSVA-7BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 56.62 | 52.56 | 52.56 | 61.07 | |
| GLAMM-7BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 56.37 | 51.92 | 51.92 | 60.34 | |
| GSVA-13BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 55.71 | 53.11 | 53.11 | 59.42 | |
| LISA-13BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 54.09 | 50.84 | 50.84 | 57.16 | |
| LISA-7BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 52.4 | 50.85 | 50.85 | 55.15 | |
| UniChangeVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 51.24 | 45.3 | 45.3 | 53.88 | |
| CRIS-101Vision Encoder=ResNet-101, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 48.71 | — | — | 51.94 | |
| CRIS-50Vision Encoder=ResNet-50, Category=Specialist Methods, Trained on MTRefSeg-21K2026.05 | 48.01 | — | — | 51.23 | |
| GSVA-13BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 17.77 | 18.17 | 18.17 | 12.03 | |
| GeoPixel-7BVision Encoder=CLIP-L, Evaluation Protocol=Fine-tuned on MTRefSeg-21K, Category=LVLMs with Single-Temporal Pretrain2026.05 | 16.4 | 16.22 | 16.22 | 9.86 | |
| GSVA-7BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 14.35 | 14.88 | 14.88 | 7.46 | |
| UniGeoSegVision Encoder=Swin-B, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 14.31 | 8.22 | 8.22 | 12.45 | |
| GLAMM-7BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 11.51 | 9.63 | 9.63 | 8.61 | |
| GeoPixel-7BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 11.23 | 12.44 | 12.44 | 5.68 | |
| SegEarth-R1Vision Encoder=Swin-B, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 6.92 | 4.57 | 4.57 | 4.56 | |
| LISA-7BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 5.54 | 6.97 | 6.97 | 1.52 | |
| LISA-13BVision Encoder=CLIP-L, Evaluation Protocol=Without Fine-tuning, Category=LVLMs with Single-Temporal Pretrain2026.05 | 2.46 | 2.58 | 2.58 | 0.67 |