NiftyNet Dev Meeting 2nd August 2019

Attendance Wenqi Irme Ben Zach Tom Jorge Carole

Round-table / Continued features integration

Irme Tom - Still working on whole volume validation - Current pb with multi threading - one thread works fine - Solution aggregator need to know the number of patches - That information can be retrieved from the aggregator - Need to add small function to know the total number of expected patches
Ben - Working on sparse input - to be continued as a separate branch - Will merge with Wenqi's work on memory management at a later stage
Wenqi - Memory / API - Work in progress - Halfway through

MICCAI educational challenge from Fernando showing Pytorch integration with NN IO
Zach initial demo of TF2.0 to be put as a demo for next NN release
Ben - Working network. Appears to be more sensitive to initialisation than 1.x networks. To try and put it in 1.13 Eager mode - Work in progress with Eric

What do we have?
- What are the strength of NN so far
- What do we want to keep
What do we want?
- What do we want from API / Interactive workflows
- What do we want for IO
What do we know?
- TF2.0 expertise from Zach + Ben
- Pytorch expertise from NN Pytorch API from Tom
What do we need (but outside of our remit
- Identify TF1.x specificities optimised by NVIDIA and not available for Pytorch / TF 2.0
- List 3D operations that could be accelerated (NVIDIA)