For the LeNet 5 architecture in the video Classic Networks, I was a bit confused by the last 2 FC layers:
After consulting some other sources including wikipedia, at first I think at the position of first FC layer in the Video, this should be a convolution layer instead.
[Update]
But looked at the wikipedia page again, I thought that because using a convolution layer with a 5x5 filter (the filter size is the same as the previous feature map’s height x width) will result in an operation that has a pretty much the same effect as using a fully connected layer. So that’s why prof. Andrew just put FC instead of Conv. layer?
This also assumes we don’t count flatten as a layer based on prof. Andrew’s suggestion.
Need your help to verify this.