sugam24/geonusaf-unet-r18-random-fold1

GeoNUSAF - UNet-ResNet18 - random 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	random
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	55
val mIoU	0.4465
val mF1	0.5887
val OA	0.7561
val kappa	0.6138

Per-class (validation)

class	IoU	F1
Residential	0.8136	0.8973
Road	0.3643	0.5341
River	0.1713	0.2924
Forest	0.6048	0.7537
UnusedLand	0.2840	0.4424
Agricultural	0.4412	0.6122

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