I am trying to do 3d image deconvolution using convolution neural network. But I cannot find many famous 3d CNNs. Can anyone point out some for me?

Background: I am using PyTorch, but any language is OK. What I want to know most is the network structure. I can't find papers on this topic.

Links to research papers would be especially appreciated.


There are many approaches for training CNN on 3d data, but the decision to use a particular architecture is heavily dependant upon the format of your dataset.

If you are using 3d point cloud data, I would suggest you go through PointNet and PointCNN.

But training a CNN on 3d point clouds is very tough.

There is also a way to train CNNs by posing the 3d structure from different viewpoints (Multiview CNNs).

But remember that training CNN on 3d data is really a tough task.

If you plan to use a voxelized input data format, I suggest going through VoxelNet.

Since you are mentioning deconvolution, the most relevant paper I can come across is 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation.

But deconvolution in its own right is a very expensive operation, which acting on 3d data makes it very hard, so I would suggest you check for alternate methods.

  • $\begingroup$ Thanks! And for the last sentence, I understand the 3d convolution is a memory consuming operation. Is it because of that that you say it is hard. And also what is the alternate methods except optimization methods like RLTV and etc. ? Thanks again $\endgroup$
    – Rickyim
    May 30 '18 at 12:59
  • $\begingroup$ yes , deconvolution is more consuming than convolution due to upsampling involved in it . regarding optimization methods i don't know much. PS: can you expand RLTV so that i can check? $\endgroup$ May 30 '18 at 17:02
  • $\begingroup$ why is it difficult to learn on from 3D data? $\endgroup$ Aug 14 '19 at 19:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.