MAE Masked Token Count

Problem

Count masked and visible tokens for masked autoencoders.

Mask mask_ratio of the num_patches tokens (round half up) and return [masked, visible] counts. The visible tokens are the complement.

Hint

Only the visible tokens pass through the encoder in MAE.

case 1

in: [196,0.75]

expected: [147,49]

case 2

in: [196,0]

expected: [0,196]

case 3

in: [100,0.4]

expected: [40,60]

case 4

in: [0,0.5]

expected: [0,0]

Starter codePython

def mae_token_counts(num_patches, mask_ratio):
    # Your code here
    pass

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