Closed akretz closed 3 years ago
Actually, the box representation is inconsistent. In main.js
, Darknet detections use the center point as reference coordinate while MOT17 detections use the top left point. I've closed this pull request until there's consensus on what representation to use.
The velocity vector is computed using the movement of the top left corner. It makes more sense to use the center point of the box instead. If a box doesn't move and just changes its scale over time, the velocity vector should also be close to zero.