card epoch 0
Browse files
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags: [semantic-segmentation, remote-sensing, unet, resnet18, geonusaf]
|
| 3 |
+
library_name: segmentation-models-pytorch
|
| 4 |
+
---
|
| 5 |
+
|
| 6 |
+
# GeoNUSAF - UNet-ResNet18 - block split, fold 1
|
| 7 |
+
|
| 8 |
+
Kathmandu Valley land-use segmentation, 6 classes, ignore_index=255.
|
| 9 |
+
Weights are the **EMA** weights (decay 0.999), not the raw final weights.
|
| 10 |
+
|
| 11 |
+
| field | value |
|
| 12 |
+
|---|---|
|
| 13 |
+
| architecture | smp.Unet, encoder resnet18 (ImageNet), decoder [128, 64, 32, 16, 8] |
|
| 14 |
+
| params | 12.46 M |
|
| 15 |
+
| arch version | unet-r18-v1 |
|
| 16 |
+
| split mode | block |
|
| 17 |
+
| fold | 1 of 3 |
|
| 18 |
+
| seed | 42 |
|
| 19 |
+
| input | 512x512, ImageNet norm, effective GSD 0.586 m/px |
|
| 20 |
+
| regularization | wd 0.01 (norm/bias exempt), ls 0.05, drop 0.1, EMA 0.999 |
|
| 21 |
+
| classes | Residential, Road, River, Forest, UnusedLand, Agricultural |
|
| 22 |
+
| best epoch | 0 |
|
| 23 |
+
| val mIoU | 0.0629 |
|
| 24 |
+
| val mF1 | 0.1128 |
|
| 25 |
+
| val OA | 0.1808 |
|
| 26 |
+
| val kappa | -0.0059 |
|
| 27 |
+
|
| 28 |
+
## Per-class (validation)
|
| 29 |
+
| class | IoU | F1 |
|
| 30 |
+
|---|---|---|
|
| 31 |
+
| Residential | 0.1800 | 0.3051 |
|
| 32 |
+
| Road | 0.0238 | 0.0465 |
|
| 33 |
+
| River | 0.0080 | 0.0158 |
|
| 34 |
+
| Forest | 0.0220 | 0.0430 |
|
| 35 |
+
| UnusedLand | 0.0541 | 0.1027 |
|
| 36 |
+
| Agricultural | 0.0893 | 0.1639 |
|
| 37 |
+
|
| 38 |
+
Checkpoint `best.pt` holds `model_state` (EMA) plus `cfg`, `metrics` and `arch_sig`.
|