Some clips might have faces going partially out of frame, where maybe only the bottom half of the face is visible for a little while.
Are those frames still usable for anything?
If I had to guess: Usable for convert (it can still generate the face and map it, and it ought to show up properly in the final product), but not usable for training (unless it knows to ignore the landmarks that are out of the frame and can still use the partial face to train the model, as long as there are full faces in there too).
I figure this is different from obstructions, since "out-of-frame landmark" doesn't require state-of-the-art NN technology to figure out, like with obstructions.
So... for the A video, is it worth at least sliding a bounding box around until the in-frame landmarks are aligned? If so, should I be including those in the alignments used for training, or excluding them, to avoid muddying the waters?