2
$\begingroup$

Let's say that we have a CNN with two convolutional layers (https://www.tensorflow.org/tutorials/layers). My question regards the dimension of the tensor, which is the output of the pooling layer 1.

In the first convolutional layer, we apply $32$ filters to the input image (let's say that the output will be $28\times 28 \times 32)$, so as far as I can understand we will get $32$ separate feature maps, because of the number of filters.

In the next step, we can apply an activation function, which does not change the dimensionality.

The $\max$ pooling layer takes as input a tensor of $28\times 28\times 32$ and the output is going to be a $14 \times 14 \times \color{red}{1}$ tensor (according to the link above).


I cannot understand the unit as the depth, since we apply $32$ filters and we apply the $\max$ pooling layer to every feature map. So, why is the output tensor $14 \times 14 \times 1$?

According to my understanding the input in the second convolutional layer should be a $14 \times 14 \times 32$ tensor. Probably, I am missing something here.

$\endgroup$
  • $\begingroup$ Thanks for the edit, it makes sense now. You should have said that the top part of your question comes from the website. $\endgroup$ – Thomas W Jun 11 '17 at 17:20
2
$\begingroup$

I kept spinning my head around this question because I seem to come along the same conclusion as you do. However, it appears to be a mistake in the documentation.

https://stackoverflow.com/questions/43453712/what-is-output-tensor-of-max-pooling-2d-layer-in-tensorflow

$\endgroup$
  • $\begingroup$ Probably, yes. Otherwise, it doesn't make much of a sense. $\endgroup$ – thanasissdr Jun 11 '17 at 17:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.