maby 0.1

U-Net for sub-cellular segmentation from Bright field

Morphology annotation for budding yeasts

Installation

pip install maby

Once installed, use command maby-init to download the pre-trained models and example images. These will be stored in your python environment, so you may need to use special permissions. We recommend using a pythong environment, such as conda.

Training

Once the package is installed, the command maby-train will be accessible from the command line.

maby-train --directory /path/to/image

Evaluating

To evaluate the performance of the trained model on the validation data, use the command maby-evaluate.

It requires a path to a results directory as input.

maby-evaluate --directory /path/to/directory --checkpoint /path/to/model.pt

Visualizing results with napari

We visualize results with napari using the command:

maby-visualize

By default, it loads a network to predict the location of the nucleus, as well as an example time-lapse. Click Run to make a prediction on the example. This may take up to a minute, depending on your hardware.

You can choose which model to run the prediction with in the dropdown menu. The nucleus and vacuole models are pre-trained, but the custom model is loaded with random initial weights.

If you want to predict on a different .tif file, you can choose an image and press Load Image to load it into the viewer. The bright field data will be automatically normalized between -1 and 1.

If you want to use a different model trained with maby-train on your own dataset, locate the checkpoint file and click Load model. Note that this will overwrite which ever model is currently loaded, so we suggest that you first switch to the custom model in the model picking pane.