orgaSeg: From image annotations to instance classifications

The purpose of this project is to segment individual organoids from microscopy data and classify them into morphological groups.
Due to the nature of the data, the project is divided into two parts: segmentation and classification.
To obtain ground-truth data to train the supervised segmentation and classification models, an annotation using QuPath is expected.
Alternatively, data folders with individual images, their crops containing one object at a time with a file name containing (class_name) and the masks for the whole images can be provided as well. 
It is worth to mention, that the segmentation step uses a model from an album catalog, and therefore can be replaced with any other model (e.g. nnUNet, CSBDeep, DeepCell, etc.). We just assumed star-convex polygons/blob-like structures as a start and provide a python script for StarDist2D. If you want to change it, install another album solution from the catalog and run it via CLI or album-gui.

Pre-release development version

This Version is meant as a development version of the raw code base.
It requires a very specific folder structure and each script to be run from its location as working directory. This is just code, without proper requirement statements, etc., and therefore not fully reproducible.

Outlook

The next and usable version will be in the form of an album catalog, where each script/step is runnable by using the album GUI from album.solutions. Then, this should also install all necessary requirements in a conda environemnt on its own, to make it even more applicable. 
This will automatically lead to a conversion into an easy-to-use pipeline and deployment as an album catalog, so the project is not relying on the specific directory structure on the working directories of each script execution.