Mask value matters in prediction?

[yaozr]

great job many thanks...

I however meet some issue with the original Pytorch_unet, I use one channel one class case, with 1024*1024 input images(value 0-255) and mask(value 0-1)
and use scale factor 0.25 hopping it train on 256x256 image...

I however get only black on prediction.
when I change mask to (0-255) and predict thresold 0.999999 it gives some meaningful result.
I know its other project but I hope you might have some experience on it?

thanks again

ZY

[YuzhouPeng]

you can try to convert gray image to three channel, and train them.