Open jiangwenchao opened 9 years ago
As far as I can figure out, cxxnet is so-called 2D convolution. You can see this in the source code convolution_layer-inl.hpp Inside the Forward() function, it seems that wmat and the input dot channel after channel.
Any way, I think you can try Mshadow or Theano to make any of your kind of nets.
I want to implement 3D CNN with the CxxNet, but I am not sure whether the convolution is 3D convolution or multi-channel 2D convolution? More specially, If I have an image with MNK where M is the number of row, N is the number of column and K is number of channel. The filter is with M'N'K'. The first way to do convolution is just 3D convolution. The second way is process each channel seperately, which is filtering the first channel with the filter in the first channel, and so on. Which way does the CxxNet choose? Thank you very much.
Do you know any other deep learning tookbox supporting 3D convolution?