Avoid having any part of the body outside the frame of the video – especially joints, including hands and feet, and the head. If that happens, our AI will adjust its output to ‘force’ the body back into the frame. As a consequence, in your results, you will likely see violent snaps and non-sensical movement of those limbs or other body parts that are outside the frame.
Simply put, the more we see in the video, the more accurate we get. So you want to maximize the visibility of the actor’s body in the video, especially with respect to the specific joints and limbs that represent the motion you want to capture. Always choose your angle with that in mind. Here are some situations to consider:
- Body obscured by body rotation: Our AI reconstructs the position of joints and entire limbs that are obscured by the actor’s own body. If you are not happy with your results, adjust your camera angle to test a different body rotation relative to the camera. Always try to maximize the visibility of body parts that engage in the most ‘unexpected’ (statistically implausible) pose or motion.
- Body obscured by other objects: Avoid having any part of the body obscured by other objects. Our AI reconstructs the position of joints and entire limbs that are obscured that way, but the longer that state persists, and the more of the body is affected, the more likely it becomes that the AI will produce results that are statistically plausible within the context of the actor’s movements, rather than loyal to what the actor actually does.