@csdashm Those are the tracked wing tips, actually. I'm excited to look at grooming, but will need to segment out those bits to look at together. There's likely some grooming mixed in here -- it's just 10K randomly chosen frames from ~1000 flies doing whatever flies do (just trying to get a sense of the data). This image shows which points we are tracking. Note colors don't match the figure, just an arbitrary colormap.

@kristinmbranson I wonder how well you could train an image generation model based on those key points and your library of images. My initial thought was to try to make a clock out of fly leg positions for the hallway screen (maybe a wing for hours, an extended leg for minutes?) but there are probably a bunch of cute dataviz ideas there.

@csdashm I suspect it would work pretty well! Our flies are not super interesting looking, since they're back-lit. We've been working on trajectory generation (that's what I'm playing with now), and we had considered directly doing video generation instead, but I decided that I didn't really care if our flies looked like flies, just that they moved like flies. I wonder if we have enough data that you could just sample -- like say we find an image where the legs and wings are just where you want them for every minute in the day, and then you just do a look up table!