Recent advances on application of deep learning for recovering object pose

Recovering object pose is of great importance to many higher level tasks such as robotic manipulation, scene understanding and augmented reality to name a few. Following the recent major breakthroughs in many computer vision tasks made by the deep learning, intensive research to experiment with it also in the task of recovering object pose is conducting. This paper aims to review the state-of-the-art progress on deep learning based pose estimation methods. Firstly, we introduce some popular datasets together with their relevant attributes. Secondly, the deep learning based pose estimation methods are summarized and categorized, and detailed descriptions of representative methods are provided, and their pros and cons are examined. Thirdly, evaluation protocol and comparable performance of reviewed approaches are given. Finally, we highlight the advantages of deep learning based pose estimation methods and provide insights for future.