Theoretical doubt

Good afternoon why in the Unet exercise we give as input the first element to f the previous con block (con_#[0]) and not directly all the output?

1 Like

This is explained in the notebook. For each of the conv blocks on the “contracting” or “downsampling” path of U-Net there are two different outputs: the first output feeds into the next downsampling block and the second output is the “skip” connection that will feed into the parallel step in the “expanding” or “upsampling” path that is the “other side of the U”. You’ll see when you build the “conv block” logic that the “skip” output is the same as the “contracting path” output, except that it omits the max pooling layer.