2D data support - Githubissues

General issue to track progress towards 2.5D data (i.e. serial 2D sections, where axial sampling is much larger than cell soma diameter) support.

To be done:

[ ] Implement 2D ResNet, adapting existing code. Should be changing input sizes, kernel sizes, and using 2D keras layers.
[ ] Implement a 2D CubeGenerator
[ ] Retrain 3D network with 2D data, e.g. using central N planes of the 3D data detailed here (direct link to data).
[ ] Implement existing 2D cell candidate detection on 2D data.
[ ] Implement 2D filtering of cell candidate pixels from initial detection. Could implement a 2D version of existing 3D code, but a simpler scikit-image/scipy implementation may be quick enough for 2.5D data (e.g. 1/10th data volume, and only 2D processing).
[ ] Support both 2/3D inputs. Do we decide when to use each (based on axial sampling and soma size), or does the user choose? Maybe the software decides, but with an override flag.
[ ] Refactor code to support 2/3D throughout (i.e. single functions, and no code duplication).
[ ] Support upstream software (cellfinder & cellfinder-napari).

brainglobe / cellfinder