Multi-level feature fusion and joint refinement for simultaneous object pose estimation and camera localization

Junyi Wang, Yue Qi

Abstract: Highlights•We design a unified framework to simultaneously estimate the object and camera poses.•We design two networks for localization and object pose estimation.•We assign a confidence value of the predicted pose to enhance pose robustness.•We design an object-level BA to jointly optimize camera and object poses.•We conduct experiments to show the outstanding performance of the two tasks.