This repository has been archived by the owner on Mar 17, 2021. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 403
NiftyNet meeting 3rd August 2018
Wenqi Li edited this page Aug 8, 2018
·
2 revisions
Attendance: Wenqi, Felix, Fernando, Tom Va., Jorge, Carole
- Image reader and sampler decoupled from application with use of Dataset API
- Need for other samplers than full image based sampler
- data too heavy to load completely (selected patch to be loaded)
- reader of csv for metadata handling or patch location definition
- metadata should still allow multi threading - probably to be done indirectly
- Need to always keep in mind that there if there is a sampler, there should be an aggregator
- Need to show how to do customised output (multi-output with different types of aggregator) - example/demo based
Refactoring will involve going to more standardised scipy/Keras naming of functionalities (model.fit, model.compile...) Should improve on readability for user Goal is also to avoid code duplicating across applications (cf work on refactoring of augmentation layers)
- Allow for Hyper parameter search in a systematic manner
- Need to create utilities folder to gather small scripts to take care of preparation of the data
- Going back to basics Each basic application should be fully supported and well organised to enable full deployment (train infer eval).
- Need for evaluation to be at least partly integrated with validation
- Feature visualisation engine - Felix + M
- block-wise convolution - Felix
- 2D Unet demo - Zach
- Compatibility with TF 1.8/1.9 - Wenqi
- CRF demo - Wenqi
- Evaluation docs - Carole
- Config file double compatibility (yaml / ini) - Dzhosh
- Logger test version
- Callbacks - Wenqi to give demos on how to use
- Function to skip iteration when nan are present
- Making a FAQ list https://github.com/NifTK/NiftyNet/wiki/NiftyNet-FAQ
- Wenqi - IO refactoring
- Tom Reader and multi-threading
- Felix - Visualisation
- Carole - Nans and hacks as contrib.
- Evaluation - Separate discussion meeting
New group mailing list?
- Information about release available on the GitHub repo
- GitHub issue list/stackoverflow for general usage discussions
- Team messaging for meetings... on Slack
Setup model zoo uploading workflow and licensing