-1

I have a folder containing files with 11*11*21 3D data and I am trying to do a binary classification with a 3D convolutional neural network. I only found the tutorial for the 2D convolutional NN on the gluon tutorials and I am not sure what things to change to accommodate 3D data. I am following this tutorial but using my own data, I tried to change the layer to 3D but I am stuck on what else I need to change. https://gluon.mxnet.io/chapter04_convolutional-neural-networks/cnn-gluon.html

I am very new to convnets and 3D convnets so any help would be greatly appreciated.

accAscrub
  • 609
  • 1
  • 6
  • 11

1 Answers1

2

For something as simple as the example tutorial you provided, you simply need to change Conv2D to Conv3D and MaxPool2D to MaxPool3D and your data would have to be in (N, C, D, H, W) layout instead of (N, C, H, W) layout.

Sina Afrooze
  • 960
  • 6
  • 11
  • Thanks very much for the help. I simply uploaded my 3D data into a mxnet dataset and used a dataloader to get batches and fed that to the neural network. Therefore, the input is only (N, C, H, W), how do I make it so that it is (N,C, D, H, W) as you described? Sorry for the stupid questions. – accAscrub Aug 06 '18 at 20:50
  • What are the three dimensions of your data? Doesn't it have width, heigh, and depth? – Sina Afrooze Aug 08 '18 at 16:32