The ability to localize and recognize objects has been at the core of many modern AI applications such as autonomous driving, robotics, smart city, and medical AI. It has also been one of the central modules in understanding and analyzing the contents of social media and cloud. Being a natural capability of human, object detection and instance segmentation have been widely studied as fundamental problems in modern computer vision. Despite the recent significant progress thanks to deep representation learning, these problems still face challenges with considerable gaps to reliable real-world applications. Such gap comes from many aspects including but are not limited to: 1) On-device requirement: the ability to maintain good performance with low latency under limited computation. 2) 3D perception and recognition: the ability to leverage 3D/geometry information and make predictions beyond 2D image planes. 3) Continuous learning: the ability to leverage large quantities of unlabeled or weakly labeled data and push the boundaries of detection performance. As a result, my PhD research focuses on addressing these challenges.

Examining Committee: 

 

                          Chair:               Dr. Larry S. Davis            
                          Dept rep:         Dr. Matthias Zwicker
                          Members:        Dr. Abhinav Shrivastava