-
Notifications
You must be signed in to change notification settings - Fork 76
3rd December 2020
agrimwood edited this page Dec 3, 2020
·
3 revisions
Alex Grimwood AG
Stefano Blumberg SB
Ester Bonmati EB
Zac Baum ZB
Adria Casamitjana AC
Yunguan Fu YF
Yipeng Hu YH
Bryan Ko BK
Zhe Min ZM
Nina Montana Brown NB
Mark Pinnock MP
Matt Clarkson MC
Qianye Yang QY
Remi Delauny RD
Shaheer Saeed SS
Informal discussion
- YH: will focus on demo optimisation over Dec/Jan holiday period
- SB: still looking for opportunities to enable conversin of DeepReg from TF to PT, low priority in face of PhD commitments
- NB: needs to handover issue 526-support multiple loss functions due to work commitments
- task will be delayed to 2021 in absence of further assistance
- AG: suggests arranging offline discussions to methodically approach demo optimisation ticket
- YH: reccommends opening individual tickets for each demo as a jumping off point for discussing changes
- YH: the main priority is to optimise performance, so that models are useful to others' applications
- YF: ivesitgating customised functionality (e.g loss), suggests a class object that can be passed to train function etc.
- SB: suggests reviewing mmdetection codebase for inspiration
- NB: should 3D slicer integration be a priority to entice new users?
- YH&ZB: improved development of base applications and models is possibly required before rollout to 3D slicer
- EB: Has anyone already suggested providing another demo to show how DeepReg can be used for finetuning/transfer learning?
- YH: agrees, but first step should be to establish some optimised models
- NB: users have raised issues identifying issues with restarting training, which is indicative of frustrated efforts to perform transfer learning
- EB: agrees to set up a ticket
tables: for google docs --> markdown, use tablesgenerator.org
formatting: for google docs --> markdown, use stackedit