Is it possible to design a three-dimensional ConvolutionLayer in version 11.1.1?
It suggest it is possible (especially the last note under details & options). Use the option “Dimensionality” found in the Details & Options section in the documentation of ConvolutionLayer…
That Dimensionality option would be handy for PoolingLayer also.
Just one update given by the Wolfram support that may be useful:
Unfortunately, ConvolutionLayer can currently handle tensors of up to rank 3 right now:
It is not currently set up to handle rank 4 tensors as would be required for RGB-D data of 3D images.