MAE Patch Masking Count

Problem

Compute the number of patches and masked patches for masked autoencoders. For a square image and square patch size, num_patches = (image_size // patch_size)^2 and the masked count is round(num_patches * mask_ratio).

The signature is mae_patch_mask_count(image_size, patch_size, mask_ratio). Return [num_patches, num_masked].

Hint

MAE typically masks 75 percent of the patches and encodes only the visible ones.

Test cases

case 1

in: [224,16,0.75]

expected: [196,147]

case 2

in: [32,8,0.5]

expected: [16,8]

case 3

in: [96,16,0.25]

expected: [36,9]

case 4

in: [64,16,0.5]

expected: [16,8]

Starter codePython

def mae_patch_mask_count(image_size, patch_size, mask_ratio):
    # Your code here
    pass

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