Batch-Norm Inference

Problem

Apply batch normalization at inference time to an H x W x C tensor.

mean, var, gamma and beta are per-channel lists. For each pixel and channel compute gamma[c] * (x - mean[c]) / sqrt(var[c] + eps) + beta[c], rounded to 4 decimal places.

Hint

Normalize with the running statistics and then apply the per-channel affine transform.

Test cases

case 1

in: [[[[1,2],[3,4]]],[1,2],[1,4],[1,0.5],[0,1],0.00001]

expected: [[[0,1],[2,1.5]]]

case 2

in: [[[[5,5]]],[5,5],[4,4],[2,2],[1,1],0]

expected: [[[1,1]]]

case 3

in: [[[[0,0],[2,2]]],[1,1],[1,1],[1,3],[0,0],0]

expected: [[[-1,-3],[1,3]]]

Starter codePython

def batchnorm_inference(x, mean, var, gamma, beta, eps):
    # Your code here
    pass

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