Semantic Segmentation on SpaceNet v1
80.83macro mIoUDEO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DEOBackbone=Swin, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 80.83 | — | — | |
| DINOv2-BInput Modality=RGB, Backbone=ViT-B, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 79.76 | — | — | |
| DINOv3-BInput Modality=RGB, Backbone=ViT-B, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 79.06 | — | — | |
| DINOv3-LSInput Modality=RGB, Backbone=ViT-L, Pre-trained=satellite imagery, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 78.67 | — | — | |
| GFMInput Modality=RGB, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 77.3 | — | — | |
| SatDiFuserInput Modality=RGB, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 77.17 | — | — | |
| DEOInput Modality=MS, Data Regime=10%2026.02 | 76.46 | — | — | |
| Scale-MAEInput Modality=RGB, Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 76.24 | — | — | |
| DINOv2-BInput Modality=RGB, Data Regime=10%2026.02 | 75.57 | — | — | |
| Cop.-FMInput Modality=Multispectral (MS), Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 75.45 | — | — | |
| DINOv3-LSInput Modality=RGB, Data Regime=10%2026.02 | 74.75 | — | — | |
| DINOv3-BInput Modality=RGB, Data Regime=10%2026.02 | 74.36 | — | — | |
| TerraFMInput Modality=Multispectral (MS), Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 73.15 | — | — | |
| SatDiFuserInput Modality=RGB, Data Regime=10%2026.02 | 73 | — | — | |
| Scale-MAEInput Modality=RGB, Data Regime=10%2026.02 | 72.8 | — | — | |
| GFMInput Modality=RGB, Data Regime=10%2026.02 | 72.57 | — | — | |
| CROMAInput Modality=Multispectral (MS), Segmentation Head=UPerNet, Evaluation Protocol=Frozen backbone2026.02 | 71.82 | — | — | |
| Cop.-FMInput Modality=MS, Data Regime=10%2026.02 | 71.65 | — | — | |
| TerraFMInput Modality=MS, Data Regime=10%2026.02 | 69.24 | — | — | |
| CROMAInput Modality=MS, Data Regime=10%2026.02 | 68.27 | — | — | |
| CACOBackbone=ResNet-50 [101], Params (M)=262024.06 | — | 77.94 | — | |
| CACOBackbone=ResNet-50, Decoder=UperNet2024.03 | — | — | 77.94 | |
| CtxMIMBackbone=Swin-B, Decoder=UperNet2024.03 | — | — | 79.47 | |
| GASSLBackbone=ResNet-50 [101], Params (M)=262024.06 | — | 78.51 | — | |
| GASSLBackbone=ResNet-50, Decoder=UperNet2024.03 | — | — | 78.51 | |
| IMPBackbone=InternImage-XL, Decoder=UperNet2024.03 | — | — | 79.08 | |
| IMP + MTPBackbone=InternImage-XL, Decoder=UperNet2024.03 | — | — | 79.16 | |
| MAEBackbone=ViT-B [2], Params (M)=86, Throughput/Minute=264k, Pre-training=OpticalRS-13 (4M)2024.06 | — | 79.41 | — | |
| MAEBackbone=ViT-B + RVSA, Decoder=UperNet2024.03 | — | — | 79.56 | |
| MAEBackbone=ViT-L + RVSA, Decoder=UperNet2024.03 | — | — | 79.69 | |
| MAE + MTPBackbone=ViT-B + RVSA, Decoder=UperNet2024.03 | — | — | 79.63 | |
| MAE + MTPBackbone=ViT-L + RVSA, Decoder=UperNet2024.03 | — | — | 79.54 | |
| PSANetBackbone=ResNet-50, Decoder=UperNet2024.03 | — | — | 75.61 | |
| SatMAEBackbone=ViT-L [2], Params (M)=307, Throughput/Minute=205k2024.06 | — | 78.07 | — | |
| SatMAEBackbone=ViT-L, Decoder=UperNet2024.03 | — | — | 78.07 | |
| SeCoBackbone=ResNet-50 [101], Params (M)=262024.06 | — | 77.09 | — | |
| SeCoBackbone=ResNet-50, Decoder=UperNet2024.03 | — | — | 77.09 | |
| SelectiveMAEBackbone=ViT-B [2], Params (M)=86, Throughput/Minute=556k, Pre-training=OpticalRS-13 (4M)2024.06 | — | 79.5 | — | |
| SelectiveMAEBackbone=ViT-L [2], Params (M)=307, Throughput/Minute=533k, Pre-training=OpticalRS-13 (4M)2024.06 | — | 79.46 | — | |
| SelectiveMAEBackbone=ViT-B [2], Params (M)=86, Throughput/Minute=556k2024.06 | — | 79.44 | — | |
| SelectiveMAEBackbone=ViT-L [2], Params (M)=307, Throughput/Minute=533k2024.06 | — | 79.48 | — |