3
$\begingroup$

I know that if we input an image of shape [6,128] to a convolutional layer with 5 filters each of shape[3,128] with S=1 and P=0 , then there will be 5 activation maps as output from the layer ..each map has a shape of [4,1]

But what about the number of maps outputted when we use batching ?

I mean if we are using a batch with size of 2 images (each of shape[6,128]) as input to a convolutional layer with 5 filters each of shape[3,128] with S=1 and P=0 also , then there will be 10 activation maps as output from the layer instead of 5 (each image in the batch has 5 maps) ?? or we will have also 5 maps but each of shape[2,4,1] ???

I think there will be 10 maps since if we apply then max pooling , we want to choose the max from each image independently , right? so each image must have separate maps from the maps of the other image..

$\endgroup$
1
$\begingroup$

Yes, separate maps. The number of the image in the batch is another dimension.

Max pool is performed on a per-image basis, other layers operate on an entire batch - e.g. batch normalisation.

| improve this answer | |
$\endgroup$

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.