Loading…
Two independent measurements of one shuttle. The cyan tube is the multi-view reconstruction from ten cameras — the reference. The orange tube is Shuttle3D reading one of those ten views and nothing else, the same monocular pipeline the paper evaluates on broadcast. The red rods are the gap between them, drawn to scale in metres.
Mode A puts the calibrated camera into three.js — same focal length, same principal point, same pose — and draws both tracks onto the video. Nothing is fitted per frame. The lens is a 92° wide angle, so the video is remapped once into the pinhole the browser can draw; the painted court and the net cord then land on their own pixels to 0.0 px, and the cord was never part of any fit.
Mode B is the same scene from anywhere. The floor is that camera's median background rectified through the ground homography, and each player is that frame's pixels cut out and stood up at the foot position the multi-view bodies give. It is a billboard, not a model.
Keys: A/B mode, Space play/pause, ←/→ step one frame.