Keep segmentation task for U-Nets?

[ziw-liu]

In microDL, using U-Nets for segmentation was explored, leaving branching code paths for both regression (virtual staining) and segmentation tasks (e.g. in preprocessing and model architecture).

@mattersoflight Given the strategy to use virtual staining models with off-the-shelf segmentation tools such as CellPose, is it still useful to keep these under-tested code?

[mattersoflight]

@ziw-liu our strategy is to work towards generalization. Approaches such as CellPose and StarDist generalize a lot better than vanilla U-Net. Having said the above, @edyoshikun will drive the segmentation of zebrafish organs, which may require a U-Net. Let's discuss this offline today.

[mattersoflight]

As we discussed, let's remove segmentation-related code.