FixMatch Consistency Loss

Problem

Compute the FixMatch consistency loss between strong and weak augmentations.

For pixels whose weak-view maximum probability is at least threshold, compare the strong-view probabilities with the one-hot argmax of the weak view (smallest index on ties). Return the MSE over all selected elements rounded to 4 decimal places, or 0.0 when no pixel passes the threshold.

Test cases

case 1

in: [[[[0.9,0.1],[0.2,0.8]]],[[[0.8,0.2],[0.6,0.4]]],0.7]

expected: 0.01

case 2

in: [[[[0.9,0.1],[0.2,0.8]]],[[[0.3,0.7],[0.6,0.4]]],0.6]

expected: 0.725

case 3

in: [[[[0.5,0.5]]],[[[0.5,0.5]]],0.9]

expected: 0

Starter codePython

def fixmatch_loss(strong, weak, threshold):
    # Your code here
    pass

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