Open ekrismer opened 5 years ago
If we have a 32x32x32x16 input volume (16 channels) shouldn't the filter be 3x3x3x16 to match each channel?
So I believe the correct answer is Undefined: The convolution step is impossible
Actually, resulting output volume size is 30x30x30x16. (I completed the quiz a little while ago, and this was the correct answer).
I believe, when we talk about dimensions, input dimension is 32x32x32. But including 16 filters, the input volume shape is 32x32x32x16. So, a filter of dimension 3x3x3 is applied whose volume shape is 3x3x3x16.
Though it seemed like shape was wrong, but if we talk about dimensions, whole thing was right.
My concept might be wrong. Feel free to correct me.
P.S. Correct the answer.
The answer to question 10 of the 4th week's quiz in "Convolutional Neural Networks" should be 30 30 30 * 32