I have been reading a book about CNNs and they have put the following: enter image description here

Here it makes sense the part that explains how each feature map is formed with the inputs convoluted by the kernels. The problem is in the next part when they explain the dimensions of the output after pooling is applied:

enter image description here

For me, it seems that it is using maxpool with an input of 28x28 (perhaps it is 28x28x12 if we consider the conv-2 of the previous figure), resulting in an output of 14x14x12. However, I cannot understand how, after that step, they obtained a feature map of 10x10 (and presumably, it is of dimensions 10x10x12). In this case, what type of pooling method have the authors used? It appears to be a global average pool, but there are no details provided. Additionally, there is no further information about the size of the kernel, stride, or padding used in that part. Any help?




Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.