Incorporate feedback for the AI@2025

[edyoshikun]

This PR addresses #10 with following :

• Remove Part 2 ( GAN approach for virtual staining). Note: we had already removed this in Adding the backup for part2  #9 .
• Update the README.md to remove the Part 1 and Part2 since we are consolidating this exercise.
• Test and update the dependencies
• Mirror the demo we have in the VCP here.
• Opened a paired PR in VisCy 128.
• Address Ziwen, Dian, and Albert's review

[AlbertDominguez]

Hey @edyoshikun,

It's looking great, and seems to mostly run on the VMs! The main issue I found is that the cell computing SSIM/Dice seems to break for me, due to an unexpected dim given to cellpose. I've suggested the change.

I also had to play around a bit with the paths to the downloaded data. Are they all gonna download the data, or they'll use it from the shared mount?

Here's my other minor feedback/suggestions:

• In the log_batch_jupyter, it'd be nice to add cmap=gray to display the images for coherency with the tensorboard viz.
• In the alert containing several code snippets, It'd be nice to add the line to load phase2fluor model user trains in case there's a crash, laptop sleeps, nb disconnection, etc. (I made the suggestion, feel free to accept/ignore!)

Lmk if anything is unclear 🙂

[adjavon]

It seems good to me aside from a few minor comments! There are some types/organization things that I've been adding as I go along. I'll push those once I've merged them with your other changes :)

[adjavon]

A couple of quick notes:

• The import cell took surprisingly long to run (~45s)
• To use iohub in the terminal (in VSCode) they will also have to set the environment there. We could have them select the interpreter (in VSCode preferences)?
• In "Visualize segmentation comparison: Fluorescence vs Virtual Staining vs Pretrained" it is getting: "Resizing is depricated in v4.0.1+" a few times --> this seems to be a cellpose issue?