W1A1 conv_backward

I got the dA, db correct, but the dW doesn’t match the expectation. I got dW_mean = -0.30435376426377897 but expectation is 1.7269914. I can’t figure out what’s going wrong. Is there any correct solution I can refer to?

My implementation as below:

Mentor Edit: Solution Code Removed

Your code is almost correct. Just one correction: it is not necessary that w and h have the same values. So, [i, w, h, c] and [i, h, w, c] will give different results. Check the instructions again for the correct order.

1 Like

Thanks. In graphic program we usually put width before height. But looks like the CNN tensor put the height before width. Lesson learnt.

1 Like