What is different between 3D-Caffe and current BVLC caffe?

[John1231983]

Hello author, this is very nice work. I just want to ask you some question about your work

1. Your 3D-caffe can work in the voxel level, instead of the pixel. Hence, all operations must be worked in 3D matrix input, instead of 2D matrix. However, I think that a 3D matrix can created by concatenation of 2D matrices. The BVLC caffe also support convolutionND, poolingND (link1, link2), where N is 3 in your case. So, what is different between your 3D-caffe and current BVLC caffe? I refer this question because I installed many caffe versions in my computer. Is it possible to use current caffe instead of your caffe?
2. You are using DICE as loss function to maximize. It is very good idea but it may be only work in 2 classes. Do you think about more two classes case, such as brain segmentation which has at least 4 classes? Because I want to apply V-Net for brain segmentation which is more complex

Best Regards, John