5. Ground truth annotation workflow with OMERO and Fiji

Requirements:

Data uploaded to OMERO (OMERO upload)
Fiji plugin for OMERO installed (1. Export OMERO images to FIJI)

1. Create tag in OMERO

Create a tag in the group where the images are located. Suggestions to name the tag:

training_1 (we can later create new tags training_2, ... to identify iterations in the training)
training_%user% (replace %user% by the person the image is assigned to annotate)
training_%project% (to identify a specific project, because multiple datasets can exist within a group)

Tags can be renamed after creation. In the case more than one "training tag" exists, we recommend placing them inside a "tag set" in OMERO

2. Assign tag to selected images

To make a good training set, randomly select images from all your datasets and assign the training tag to them.

The training set needs to cover as much variability that exists in your data, in order to give a chance to your model to generalize classification to future data.

3. Open images from Fiji

Following the instructions in 1. Export OMERO images to FIJI, connect to OMERO.
Browse the data by tags, and find your images under the "training tag":

Right click on images → "View in ImageJ"
Make your ROIs and add them to the ROIManager (ctrl+T)

4. Export ROIs to OMERO

When finished with all ROIs of an image, export the ROIs to OMERO (2. Export FIJI ROIs to OMERO)

5. Generate the training dataset

Images and ROIs living inside OMERO, we need to export them to a local folder (training a model can be resource intensive, it's more efficient to avoid loading the same image over and over from the server)
Install the OMERO macro extension (3. Reading data from OMERO in a Fiji macro)
You can create your own macro to loop over every images having the tag (find the tag ID), read their ROIs and export images and masks locally.

Macro example:

save_masks_tagid.ijm

How to find your group id: