Adam Tyson

Update cell classification network

4

For many reasons the existing cell classification network (and associated helper functions) is suboptimal: * brainglobe/cellfinder#304 * brainglobe/cellfinder#303 * brainglobe/cellfinder#301 The use of TensorFlow is also causing headaches: • https://github.com/brainglobe/cellfinder/issues/293...

[Feature] Allow the user to select points from more than one layer at once in the curation interface

2

This would be useful if the user wants to select e.g. cells that have been correctly & incorrectly classified in one go. Not sure if this is possible within napari...

[Feature] Freeze upper layers of the network for retraining.

5

Linked to brainglobe/cellfinder#354. Allowing upper layers to be frozen when retraining should make slight tweaks easier

[Feature] Reduce overfitting when training

2

Users are generating small amounts of data, and the network seems to be over fitting. They're having to manually get existing data to constrain their models. This could be improved...

The default batch size for classification (32) was chosen so it would run on pretty much any good GPU but as @larsrollik has found, this can be increased a lot,...

Although the background channel is useful for the cell classification and registration steps, it is not always practical to double the data collected (and in the case of some fluorophores,...

e.g. when deleting a training data layer, it is still there as `self.training_data_cell.....`, and prevents new layers from being added (`CurationWidget` still "thinks" it exists). How do we delete the...

Error raised when no cell candidates are detected.

1

Unlikely to happen in normal use, but may occur in v. small images.