Retraining the cellfinder classification network

In this tutorial, you will use the cellfinder plugin for napari to retrain the cell classification network.

The aim of this tutorial is to get up and running retraining cellfinder using a very small training dataset. When it comes to using cellfinder with your own data, it is very important to retrain the network properly. You also may want to use the command line tool instead of using napari. For more information, please see the full cellfinder re-training documentation.

Before running this tutorial, please run the cell detection tutorial first.

Note

You will need napari installed on your computer - please follow napari’s installation instructions to do so (including their recommendation to use a conda environment).

1. Open napari.

2. Open the curation widget by selecting Plugins > cellfinder-napari > Curation in the napari menu bar near the top left of the window.

3. Load some sample data File > Open sample > Sample data (cellfinder-napari). This will open the same small two-channel 3D image from the detection tutorial.

4. Click Files > Open File(s) and select the cells.xml from the detection tutorial. If napari prompts you to choose a plugin, choose brainglobe-napari-io.

5. Set the signal image to Signal

6. Set the background image to Background

7. Click Add training data layers

The curation widget appears on the right-hand side of the window alongside the raw data with detected cells overlaid

8. Highlight the Cells layer and then choose the Select Points tool by clicking the arrow in the layer controls widget, or pressing the 3 key on your keyboard.

9. Select some cells in the image by drawing a box and click Mark as cell(s)

10. Select some different cells in the image and click Mark as non cell(s)

11. Click Save training data

12. Create and then select a new directory on your computer (e.g. cellfinder-retraining) and click Choose.

13. Close the curation widget by clicking the x at the top left of the widget.

14. Open the retraining widget by selecting Plugins > cellfinder-napari > Train network in the napari menu bar near the top left of the window.

15. Select the data for retraining by clicking Select files next to YAML files and choose the training.yml file is inside the cellfinder-retraining directory created earlier

16. Choose a directory to save the training output, e.g., create a trained_network directory

17. Set Epochs 2

The training widget appears on the right-hand side of the window.

18. Click run

19. The training will then run, watch the terminal for updates. Once complete, there will be trained models (files ending in .h5) in the trained_network directory that can be used for cell detection.

Hint

For more information about how to use the curation and training plugins, please see the full using cellfinder in napari guide.