search

IanTaehoonYoo / semantic-segmentation-pytorch

Pytorch implementation of FCN, UNet, PSPNet, and various encoder models.
MIT License
86 stars 19 forks source link

how to make the network work well with num_class=2 #20

Open zhou-rui1 opened 3 years ago

zhou-rui1 commented 3 years ago

Hi, originally it work well with num_class=256...

but when I edited img = img.astype('float32') / 255.0, seg = seg.astype('float32') / 255.0 in segmentation_dataset.py, and target = target//255 in trainer.py to adapt the train to num_class=2only(for binary seg),

then the network can not train well and predict nothing (black), did I do something wrong? image

you are so kind and sorry to bother you again...