Skip to content
This repository has been archived by the owner on Mar 17, 2021. It is now read-only.

NiftyNet meeting 2nd November 2018

Wenqi Li edited this page Nov 2, 2018 · 2 revisions

Attending: Wenqi, Oeslle, Felix, Tom Va., Dzhoshkun, Jose, Carole, Tom V.

Recent completed progress

New features and functionalities

  • New loss function
  • Smoothing for random scaling / random deformation
  • Moved Model Zoo to GitHub

Recent activities

  • RSNA poster
  • DL Workshop
  • Hackday

Current work in progress

Whole volume validation Hackday

  • Generation of whole volume to temporary folder

  • Validation to read temporary folder

  • Training

  • Validation phase

  • Evaluation phase - Metrics at volume level - Stop training during that time

CSV Reader

  • 1 column -> label (categorical)
  • Multicolumn -> features (float)
  • Multiline -> csv patch sampler Action: Documentation and tests to be written

Pytorch with NiftyNet

  • Pytorch networks NN Preprocessing and Postprocessing but Pytorch network
  • Proof of versatility of NN implementation and the of the modular quality of the framework

Transfer learning

External pull request including tests for selective restoring of variables

Config file transfer to YAML

  • YAML Config file Dan working on it Backward compatibility
  • Ini to be deprecated

Discussions / Interrogations

Model licensing

  • Model licensing: Restriction of data follow down to the models: 1 license per model
  • Need for each model to go back to the data acquisition and make sure of possibility to leave access to potential resulting models
  • Use of Decathlon data?

Model zoo

  • Tensorflow hub Moving Model Zoo towards TFHub?
  • Models usable outside nifty net
Clone this wiki locally