https://groups.google.com/forum/m/#!topic/caffe-users/wcKAufmy3uI 印象笔记上有备份
ND convolution with im2col https://github.com/BVLC/caffe/pull/2049 印象笔记上有备份
https://github.com/iscoe/coca Does caffe support using 3D volumes (ie medical images) as an input?
A: It supports this via the color channels already. I noticed on github there was a discussion
about adding a vol2col analogue of im2col for an explicit 3D convolution.
https://github.com/BVLC/caffe/pull/1486
How does 3D convolution really works in CAFFE, a detailed analysis