Scale-Invariant Depth Error

Problem

Compute the scale-invariant depth error between two depth maps.

Over pixels where both depths are positive, take the root mean square of log(pred) - log(gt) using natural logarithms. Return the error rounded to 4 decimal places, or 0.0 when no pixel is valid.

Hint

Taking logs turns a global scale factor into a constant shift that the metric forgives.

Test cases

case 1

in: [[[1,2],[3,4]],[[1,2],[3,4]]]

expected: 0

case 2

in: [[[2,2],[2,2]],[[1,1],[1,1]]]

expected: 0.6931

case 3

in: [[[1,4]],[[1,1]]]

expected: 0.9803

case 4

in: [[[0,2]],[[5,1]]]

expected: 0.6931

Starter codePython

def scale_invariant_depth_error(pred, gt):
    # Your code here
    pass

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