About the project
This project showcases how to train a Cellpose model using raw images and associated ROIs in ImageJ format (*.zip).
The general steps are as following:
Load the images and annotations. The should match up with each other, i.e. should be associated with each other through annotations. This can be achieved by naming the image and mask/ROI files the same.
In the case of ROIs, you must convert them to masks. Ensure that the masks have the same size as the raw image.
Annotate each image with its associated masks through the data annotation system (Annotate with data)
Split the raw+mask images into training, test, and validation data sets
Pass all data into the Cellpose training node together with a pretrained model.
You can also:
train from scratch by selecting "No model" as pretrained model within the pretrained model selection
train from multiple pretrained models - just add more into the list (pretrained models) and/or attach multiple loaded pretrained models
train from a pretrained model file - first import the model and then pass it into the pretrained model slot
The data provided with this pipeline was published here:
Cseresnyes Z, Hassan MI, Dahse HM, Voigt K, Figge MT. Quantitative impact of cell membrane fluorescence labeling on phagocytosis measurements in confrontation assays. Frontiers in microbiology. 2020 Jun 5;11:1193.
How to load the project
Download the workflow RO-Crate(s)
Start JIPipe and click Import RO-Crate
Select the download file and follow the instructions
