Typically the image data has 3 dimensions. We can still do the convolution for multi-dimensional data using matrix multiplication. All we need to do is to concatenate the channels together and the weights together.
chii
Because the operations over the input channels here are similar to the image manipulations mentioned in DSL lecture last week, is it possible to reuse some of those DSLs in this kind of application?
Typically the image data has 3 dimensions. We can still do the convolution for multi-dimensional data using matrix multiplication. All we need to do is to concatenate the channels together and the weights together.