Semantic Segmentation on Potsdam (mIoU)
76.17mIoUCross-Scale MAE
Evaluation results
| Method | Links | |
|---|---|---|
| Cross-Scale MAEBackbone=ViT-Base, Scale=100%2024.01 | 76.17 | |
| Scale-MAEBackbone=ViT-Base, Scale=100%2024.01 | 75.07 | |
| U-Net BaselineEvaluation Protocol=trained from scratch, Data Modality=include elevation data2026.05 | 74.918 | |
| OracleTraining Regime=Oracle (fully supervised SegFormer-b0)2025.12 | 74.3 | |
| Supervised BaselineBackbone=ResNet50, Scale=100%2024.01 | 74.11 | |
| Scale-MAEEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 74.058 | |
| SatMAEBackbone=ViT-Base, Scale=100%2024.01 | 73.55 | |
| DOFAEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 73.432 | |
| Scale-MAEBackbone=ViT-Base, Scale=50%2024.01 | 73.33 | |
| mCL-LCBackbone=ResNet50, Scale=100%2024.01 | 73.16 | |
| FLOROEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 73.101 | |
| SeCoBackbone=ResNet50, Scale=100%2024.01 | 72.98 | |
| BYOLBackbone=ResNet50, Scale=100%2024.01 | 72.64 | |
| SimCLRBackbone=ResNet50, Scale=100%2024.01 | 72.33 | |
| Cross-Scale MAEBackbone=ViT-Base, Scale=50%2024.01 | 72.25 | |
| SimSiamBackbone=ResNet50, Scale=100%2024.01 | 71.88 | |
| TerraMind-LEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 71.632 | |
| Cross-Scale MAEBackbone=ViT-Base, Scale=25%2024.01 | 70.82 | |
| GFM-SwinEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 70.632 | |
| RemoteCLIPEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 69.022 | |
| SatMAEBackbone=ViT-Base, Scale=50%2024.01 | 68.79 | |
| Scale-MAEBackbone=ViT-Base, Scale=25%2024.01 | 68.53 | |
| ViT BaselineEvaluation Protocol=trained from scratch, Data Modality=include elevation data2026.05 | 67.083 | |
| CAFe-DINORS Training=No2026.05 | 66.8 | |
| CROMAEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 66.752 | |
| SpectralGPTEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 66.65 | |
| SatlasNetEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 65.965 | |
| SSL4EO-S12-MoCoEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 63.633 | |
| SatMAEBackbone=ViT-Base, Scale=25%2024.01 | 63.25 | |
| SSL4EO-S12-DINOEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 63.024 | |
| SSL4EO-S12-Data2VecEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 62.791 | |
| SSL4EO-S12-MAEEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 61.141 | |
| mCL-LCBackbone=ResNet50, Scale=50%2024.01 | 59.73 | |
| SeCoBackbone=ResNet50, Scale=50%2024.01 | 59.35 | |
| SimSiamBackbone=ResNet50, Scale=50%2024.01 | 59.03 | |
| SegEarth-OV3Training Regime=Training-free2025.12 | 57.8 | |
| SDCI-v1Prompt Setting=Generalized Labels, GFLOPs=2022.72, Structural Branch=DINO-v1 (ViT-B/8)2026.01 | 57.74 | |
| Prithvi 1.0 100MEvaluation Protocol=frozen-encoder, Data Modality=include elevation data2026.05 | 57.603 | |
| SDCI-v2Prompt Setting=Generalized Labels, GFLOPs=1683.29, Structural Branch=DINO-v2 (ViT-B/14)2026.01 | 57.51 | |
| BYOLBackbone=ResNet50, Scale=50%2024.01 | 57.15 | |
| mCL-LCBackbone=ResNet50, Scale=25%2024.01 | 55.89 | |
| SeCoBackbone=ResNet50, Scale=25%2024.01 | 55.71 | |
| BYOLBackbone=ResNet50, Scale=25%2024.01 | 55.09 | |
| SimCLRBackbone=ResNet50, Scale=50%2024.01 | 54.72 | |
| SimSiamBackbone=ResNet50, Scale=25%2024.01 | 54.39 | |
| SkySense-OTraining Regime=Training on remote sensing segmentation data2025.12 | 54.1 | |
| ResCLIPPrompt Setting=Generalized Labels, GFLOPs=1029.662026.01 | 53.31 | |
| SimCLRBackbone=ResNet50, Scale=25%2024.01 | 53.27 | |
| Supervised BaselineBackbone=ResNet50, Scale=50%2024.01 | 52.12 | |
| SegEarth-OVRS Training=Self-Supervised2026.05 | 52 | |
| CorrCLIPTraining Regime=Training-free2025.12 | 51.9 | |
| ConInfer2026.03 | 49.99 | |
| SCLIPPrompt Setting=Generalized Labels, GFLOPs=2926.292026.01 | 49.93 | |
| VIPBackbone=DINOv3 (ViT-L)2026.05 | 49.8 | |
| ProxyCLIPTraining Regime=Training-free2025.12 | 49 | |
| SegEarth-OVPrompt Setting=Generalized Labels, GFLOPs=1124.282026.01 | 48.8 | |
| SegEarth-OVTraining Regime=Training-free2025.12 | 48.5 | |
| SegEarth-OV2026.03 | 47.94 | |
| SegEarth-OVVenue=Ours2024.10 | 47.1 | |
| SegEarth-OV2026.05 | 47.1 | |
| ClearCLIPPrompt Setting=Generalized Labels, GFLOPs=2953.552026.01 | 46.88 | |
| MaskCLIPPrompt Setting=Generalized Labels, GFLOPs=2944.882026.01 | 46.76 | |
| CLIPerPrompt Setting=Generalized Labels, GFLOPs=4615.632026.01 | 45.22 | |
| Supervised BaselineBackbone=ResNet50, Scale=25%2024.01 | 45.18 | |
| MaskCLIP*type=baseline2026.03 | 44.49 | |
| Trident2026.05 | 44.4 | |
| ProxyCLIP2026.05 | 44.1 | |
| SC-CLIP2026.05 | 43.4 | |
| CorrCLIPPrompt Setting=Generalized Labels, GFLOPs=6424.472026.01 | 43.28 | |
| GSNetRS Training=Yes2026.05 | 43.2 | |
| CorrCLIP2026.05 | 42.6 | |
| ClearCLIPTraining Regime=Training-free2025.12 | 42 | |
| ClearCLIPVenue=ECCV'242024.10 | 40.9 | |
| ClearCLIP2026.03 | 40.2 | |
| SCLIPTraining Regime=Training-free2025.12 | 39.6 | |
| GEMTraining Regime=Training-free2025.12 | 39.1 | |
| RSKT-SegTraining Regime=Training on remote sensing segmentation data2025.12 | 38.4 | |
| GEM2026.03 | 38.24 | |
| GSNetTraining Regime=Training on remote sensing segmentation data2025.12 | 37.9 | |
| SANTraining Regime=Training on remote sensing segmentation data2025.12 | 37.3 | |
| SCLIPVenue=arXiv'232024.10 | 36.6 | |
| GEMVenue=CVPR 242024.10 | 36.5 | |
| SDCI-v2Prompt Setting=Original Labels, GFLOPs=1683.29, Structural Branch=DINO-v2 (ViT-B/14)2026.01 | 36.47 | |
| OVRSTraining Regime=Training on remote sensing segmentation data2025.12 | 36.4 | |
| Cat-SegTraining Regime=Training on remote sensing segmentation data2025.12 | 35.8 | |
| CASSPrompt Setting=Generalized Labels, GFLOPs=26943.792026.01 | 34.28 | |
| MaskCLIPTraining Regime=Training-free2025.12 | 33.9 | |
| NACLIPPrompt Setting=Original Labels, GFLOPs=242.962026.01 | 33.47 | |
| CASSPrompt Setting=Original Labels, GFLOPs=26943.792026.01 | 32.68 | |
| SDCI-v1Prompt Setting=Original Labels, GFLOPs=2022.72, Structural Branch=DINO-v1 (ViT-B/8)2026.01 | 32.37 | |
| NACLIPPrompt Setting=Generalized Labels, GFLOPs=242.962026.01 | 32.03 | |
| SCLIP2026.03 | 31.84 | |
| MaskCLIPVenue=ECCV'222024.10 | 31.7 | |
| MaskCLIP2026.03 | 31.41 | |
| SegEarth-OVPrompt Setting=Original Labels, GFLOPs=1124.282026.01 | 30.63 | |
| SEDTraining Regime=Training on remote sensing segmentation data2025.12 | 29.4 | |
| OVRSRS Training=Yes2026.05 | 28.6 | |
| SCANTraining Regime=Training on remote sensing segmentation data2025.12 | 27.5 | |
| SCLIPPrompt Setting=Original Labels, GFLOPs=2926.292026.01 | 27.13 | |
| ResCLIPPrompt Setting=Original Labels, GFLOPs=1029.662026.01 | 27.02 |