Identifying reliable pose estimations from Absolute Pose Regression (APR) methods to improve accuracy and robustness, while minimizing computational overhead.
