how well would BEAST work for small animal in large arena scenarios?

Hey everyone, 

It is more of a discussion I guess. But I am intrigued by the performance gains for pose estimation with the pretrained model. 

From what I understand from the paper though is that in cases where the animal is in a large arena (animal is 35x35 pixels of a 1920x1080 frame), BEAST doesnt make much sense, right? At least not with default settings. 

Or is my intuition mistaken? 