| --- |
| tags: [semantic-segmentation, remote-sensing, unet, resnet18, geonusaf] |
| library_name: segmentation-models-pytorch |
| --- |
| |
| # GeoNUSAF - UNet-ResNet18 - block split, fold 1 |
|
|
| Kathmandu Valley land-use segmentation, 6 classes, ignore_index=255. |
| Weights are the **EMA** weights (decay 0.999), not the raw final weights. |
| |
| | field | value | |
| |---|---| |
| | architecture | smp.Unet, encoder resnet18 (ImageNet), decoder [128, 64, 32, 16, 8] | |
| | params | 12.46 M | |
| | arch version | unet-r18-v1 | |
| | split mode | block | |
| | fold | 1 of 3 | |
| | seed | 42 | |
| | input | 512x512, ImageNet norm, effective GSD 0.586 m/px | |
| | regularization | wd 0.01 (norm/bias exempt), ls 0.05, drop 0.1, EMA 0.999 | |
| | classes | Residential, Road, River, Forest, UnusedLand, Agricultural | |
| | best epoch | 62 | |
| | val mIoU | 0.4516 | |
| | val mF1 | 0.5807 | |
| | val OA | 0.7898 | |
| | val kappa | 0.6255 | |
| |
| ## Per-class (validation) |
| | class | IoU | F1 | |
| |---|---|---| |
| | Residential | 0.8310 | 0.9077 | |
| | Road | 0.3965 | 0.5678 | |
| | River | 0.1329 | 0.2346 | |
| | Forest | 0.6495 | 0.7875 | |
| | UnusedLand | 0.1763 | 0.2998 | |
| | Agricultural | 0.5232 | 0.6870 | |
| |
| Checkpoint `best.pt` holds `model_state` (EMA) plus `cfg`, `metrics` and `arch_sig`. |
