pietz / unet-keras Goto Github PK

Hello,

I am trying to create a model with residual connections (residual=True). I am using the following command:

from unet import UNet
myModel = UNet((3,256,256),out_ch=3,dropout=0.15,batchnorm=True,residual=True)

but I am getting the following error:

ValueError: `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 3, 256, 256), (None, 64, 256, 256)]

pointing to the last line of conv_block

I am currently using Theano as backend, but I have already tried to pass input shape as channel last (img_shape=(256,256,3)), but got similar error. I even tried to set the axis of the Concatenate, but got no success.

PS: everything works when I use residual=False, but I would really like to use shortcut connections.

def UNet(img_shape, out_ch=1, start_ch=64, depth=4, inc_rate=2., activation='relu', .. if we increase the value of depth like 8,10 ,15 will it automatically increase layers of the model?