Re. Receptive field,. I’ve googled it and here is what I’ve found.
The receptive field in Convolutional Neural Networks (CNN) is the region of the input space that affects a particular unit of the network.
Here is a more in-depth explanation
The image below shows how the receptive field is extended when two convolutions 3x3 stride 2x2 are overlayed on each other, the second convolution has receptive filed of 7x7 even though it is stil 3x3 convolution. It is caused by the stride that is larger than 1.The pooling layer seems to be a cheap way of extending the receptive field for the layers that come after it.
