U-Net Skip Connection Shape

Problem

List the U-Net encoder shapes level by level. Starting at 64 channels, every level halves the spatial dimensions with floor division and doubles the channel count.

The signature is unet_skip_shape(input_h, input_w, depth). Return [[h, w, channels], ...] for levels 0 through depth.

Hint

Skip connections need encoder and decoder shapes to match exactly at every level.

Test cases

case 1

in: [32,32,2]

expected: [[32,32,64],[16,16,128],[8,8,256]]

case 2

in: [16,24,1]

expected: [[16,24,64],[8,12,128]]

case 3

in: [8,8,0]

expected: [[8,8,64]]

case 4

in: [5,5,1]

expected: [[5,5,64],[2,2,128]]

Starter codePython

def unet_skip_shape(input_h, input_w, depth):
    # Your code here
    pass

4 test cases · run your code against them in the editor