sronilsson
commented
1 year ago Hi @DanieleRomeo! I don't know enough about the behaviors and tracking data, but there is nothing, theoretically, stopping you if you're tracking is solid and behavior is salient (e.g. human can tell the behaviours apart).
We've done a few fish projects lately, albeit single fish, and some general points:
Out-of-the-box default SimBA calculates mainly variables related to movements, shapes and distances between animals and their body-parts to build the classifiers. So, out-of-the-box, you will easily be able to capture behaviors like “jolts”, while some other behaviors may be trickier with the default variables. For more complex fish behaviors we’ve had to concentrate further on circular statistics, documented HERE, to get good models going and use these methods as documented HERE, through the SimBA GUI.
So, it’s good to include two body parts for each animal in tracking which allows you to calculate circular statistics, e.g. head and swim bladder or whatever the equivalent is in your species as in image below (not just the perimeter of the fishes). If such body-parts are included, we can do more appropriate stats to capture what the human annotator is looking at. The image is from the side but I worked with it from above as well.
A caveat is that building your own feature extraction class in SimBA, and pick-and-mixing these statistics calculators, requires some moderate python skills which probably keep many people away (I haven’t yet thought of a user-friendly way to do it graphically). If that is the case, I am happy to help to help put together and explain the code.
Lastly, we do not have domain knowledge about fish behavior, and the calculators have been written under guidance from people with such knowledge telling us what they need. If you find something missing, that you want calculated, let us know.
Thanks!
Simon
DanieleRomeo
Hello,
We work with cleaner fish, studying the cleaning behavior in coral reef. In this interaction, the cleaner fish cleans another fish called "client", standing close to him and biting the surface of his body, eating parasites and dead skins.
My study is mainly focus on the "cheating behavior", when the cleaner fish eats the mucus of the client instead of the parasites. This, provoke a clear reaction in the client identify as the "Jolt", a rapid movement of the client away from the cleaner fish.
Do you think Simba could be a good tool to analyse this kind of interaction, considering the interaction time (time spent together, where the cleaner fish cleans the client), dance (are wide longitudinal movement of the cleaner fish to attract the attention of the client), tactile stimulation (where the cleaner fish stays on the dorsal part of the client providing a massage with pelvic fins), and the cheating behavior (considering so far we have only video from frontal view)?
I would like to hear your idea about before proceed and try it!
Thank you so much for your time and for your precious help!
Daniele