sugam24/geonusaf-unet-r18-random-fold0

GeoNUSAF - UNet-ResNet18 - random split, fold 0

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	random
fold	0 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	64
val mIoU	0.4533
val mF1	0.5968
val OA	0.7655
val kappa	0.6264

Per-class (validation)

class	IoU	F1
Residential	0.8053	0.8922
Road	0.3970	0.5683
River	0.1575	0.2721
Forest	0.5576	0.7160
UnusedLand	0.3063	0.4689
Agricultural	0.4962	0.6632

Checkpoint best.pt holds model_state (EMA) plus cfg, metrics and arch_sig.