Closed Ali-Parsa closed 3 years ago
This choice was made to keep the state-model simple. Since people/pedestrians where the main object class for tracking in the motchallenge the aspect ratio can be approximated with a constant.
Naturally, this isn't true 100% of the time. I.e. someone may transition from standing to lying or the camera's pose may change or a long truck is turning. If this situation is common to your use-case, you should probably consider extending the estimation model to suit.
Closing due to inactivity and answer provided.
Hi Why did you consider the aspect ration constant in your paper and implementation?