Viscy workflow questions and experience from DL@MBL

[edyoshikun]

I am documenting here some of the questions that arose during the use of Viscy and the experience of running this during the DL@MBL course:

I list them as tasks (please cross them out if you think these have been solved) and/or open new issues if these are high priority.

• [x] Capability to select normalization strategy between std or irq.
• [x] Tensorboard validation shows a (Z-stack) rather than the multiple batches
• [ ] Difficulty determining what config file to use and what parameters to change.
• [ ] Ability to do tiled predictions.
• [x] HCS DataLoader outputting the wrong shape #47
• [x] Tool to easily crop and/or rechunk ome-zarr stores. Current mantis chunking of (ZYX)<500MBs make IO bottlenecks. https://github.com/czbiohub-sf/shrimPy/pull/145
• [x] Unclear preprocessing steps, config.yml, and order of CLI calls. Mostly solved by #43 #45.