SMP Imagenet과 Noisynet 비교

[JaegyeongByun]

model : Unet (encoder : timm-efficientnet-b3) loss : CrossEntropyLoss optimizer : Adam learning rate : 0.0001 weight decay : 1.00E-06 batch size : 16 epoch : 40 random seed : 21 augmentation : noramlize

encoder weight	best epoch	train_loss	val_loss	mIoU (val)	mIoU (LB)
Imagenet	37	0.05110	0.3593	0.4803	0.4950
Noisynet	32	0.05539	0.2969	0.5159	0.4951