Local Climate Zone Segmentation on WUDAPT LCZ map Swiss cities 100 m (train)
0.13LossTESSERA
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TESSERAArchitecture=Attention U-Net, Predictor set=Details in Table 3, Input Channels=128, Patch Size=128 × 128 px, Spatial Resolution of predictor variables=10 m, No. of Classes=14, Spatial resolution of LCZ reference=100 m, No. of epochs until early stopping=170, Best epoch=148, Loss Function=Dice + Focal, Optimizer=Adam (lr=1×10-3), Study Area=Basel, Bern, Geneva, Lausanne, Zurich, Reference Labels=WUDAPT LCZ map (100 m)2026.06 | 0.13 | 89 | 81 |