search

NifTK / NiftyNet

[unmaintained] An open-source convolutional neural networks platform for research in medical image analysis and image-guided therapy
http://niftynet.io
Apache License 2.0
1.36k stars 403 forks source link

Mixup / Mixmatch capability? #419

Closed chazzmoney closed 5 years ago

chazzmoney commented 5 years ago

Hello! I found your paper from MIDL 2018 on using mixup/mixmatch in semantic segmentation. In the paper, it mentioned that the code was viewable here - but I can't seem to find it.

Is it present in one of the branches and I missed it?

Thanks!

ericspod commented 5 years ago

@Zach-ER Is there a link to the code somewhere?

chazzmoney commented 5 years ago

Thanks for looking - will be very helpful!

Zach-ER commented 5 years ago

Hi there,

I have created a branch here: https://github.com/NifTK/NiftyNet/tree/419-add-mixup-mixmatch-from-MIDL-2018

That you can use while it pends merging into the main code. Two provisos:

  1. mixup is designed to be of low computational cost. My current implementation calls the sampler twice per iteration: if this is slow, it might be worth optimising further.
  2. mixup only works with dense losses (e.g. Dice_Dense). You may have to implement your own dense loss, although several are already there.
chazzmoney commented 5 years ago

I really appreciate it - thank you!