Augmentation and Occluded Point Advice

[jaredsabre]

Hi. I was hoping someone might be able to provide some advice on how to best train a single model for my setup as I have 4 camera angles of the same animal. Long term goal is to use Anipose with my output but I need to get the tracking in a better spot.

Our setup is a horse stall, approximately 15 feet by 15 feet, with one camera placed in each of the 4 corners about 8 feet high and aiming downward to the center of the stall floor. They are angled downward so that they capture as much of the floor beneath them as possible while ensuring they aim high enough that the horse’s head is always in frame. The points I am trying to track are head (between the ears), shoulder (above the front legs on the back), hind (above the rear legs on the back), and all four hoofs.

After reading the manual about using the horizontal flip augmentation, thought I wasn’t using it, made me wonder if I should be since opposing corner cameras essentially produce flipped frames of each other in terms of the horse’s orientation. Also, at any given point in time, 1-2 cameras will have say 4 of 7 points not visible and out of the frame entirely. there are also other instances where some points are out of frame while others are occluded but I could easily mark where they would be by marking what's in front of it. Is there a best practice for doing this or can it make my models worse if it happens too often?

Thank you for any help in advance!