How to semantic segmentation

Seung Lab

This document describes the procedure for creating a ground truth semantic segmentation.

1. Create and open the project in Omni. (Refer to the How To Vast document on creating an Omni file from a Vast file.)
2. Check the segment ID numbering. If the IDs do not increment with the size ordering, or there seem to be any other errors, notify the project owner.
3. Once you’ve confirmed that they segments are good to go create a google sheet and name it something appropriate to the set you are doing.
4. In that sheet you will be listing the classifiers. (Example.) In this example we hadn’t streamlined the process so the first 500 or so are all the original segment IDs and after that you see that they match up with the sheets numbers minus 1.
5. Go through the segment list and determine to the best of your ability whether the segment should be classified as one of the following: Undefined, Soma, Axon, Dendrite, Glia, Unknown, Not Present. (A suggestion for another useful classifier would be dendritic spine head.)
6. Each of the classifiers has a corresponding number: 0, 1, 2, 3, 5, 6, 99. They are in order with the list above.