Niftynet meeting 14th may 2018

Attending: Zach Felix Dzhoshkun Tom Va. Wenqi Carole

IO design

• Demonstration of new input format for use with tf.dataset - one modality numpy array can be used with Keras
• Need for Metadata integration (subject info ex diagnostic, age, genetic status...): either with csv list or as parsed text file - Matching by Subject id
• Preprocessing needed at both image and patch level
• Need for new look at selective sampler. Too slow for the moment
• Loading of very large files: With jpeg and tiff, possible to load only part of the image but need to keep track the place of the sampling. Possible to read png with open-cv

FAQ

• Need for an FAQ list + List of things that can go wrong for various reasons.
• To be populated and written

Migrating to GitHub

• Closing Cmiclab issue list - Deadline for stop on CMICLab and transfer to GitHub thereon
• Update contributors' guide for GitHub workflow
• Need to change md for contributors
• Issue list to be closed on CMICLab
• Pb of migration: No migration infrastructure available. Need to be done manually.
• Squashing of commits after merge requests? How to take steps back?

Demos

• Image classification WIP
• 2D image segmentation demo WIP

New version

• v0.3.0 releasing soon

Compatibility

• NN doesn't work with Tensorflow 1.8