VAE Encoder Stride Chain

Problem

List the spatial resolutions produced by the downsampling blocks of a VAE encoder, each of which halves the previous size with floor division.

The signature is vae_encoder_strides(input_size, num_downsamples). Return the list of resolutions after each downsample.

Hint

Stable Diffusion's VAE downsamples by a factor of 8 in total.

case 1

in: [256,3]

expected: [128,64,32]

case 2

in: [512,4]

expected: [256,128,64,32]

case 3

in: [224,2]

expected: [112,56]

case 4

in: [32,1]

expected: [16]

Starter codePython

def vae_encoder_strides(input_size, num_downsamples):
    # Your code here
    pass

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