PhD Dissertation Proposal by WU Xiongwei | Deep Learning for Real-World Object Detection

About The Talk

The success of deep learning in recent years has helped boost the performance of object detection, a fundamental problem in computer vision, where we have to identify the categories and location of an object in a given image. Despite significant progress, most existing detectors are designed to detect objects in the academic context and have little applicability to real-world scenarios. In this dissertation, we investigate and develop object detection algorithms which improve the real-world applicability and overcome the limitation of existing methods. Our research mainly focuses on three aspects: face detection, high quality detection and few-shot detection. We provide a comprehensive survey of object detection based on deep learning and analyze the limitations of current detection algorithms in the context of the identified real-world problems. We then propose three novel frameworks: FANet, BPN and Meta R-CNN which address these limitations and improve the real-world applicability of object detection algorithms.

Speaker Biography

Xiongwei Wu is a PhD candidate at School of Information Systems, Singapore Management University. He is advised by Associate Professor Steven Hoi and his research direction is deep learning based detection problem such as generic object detection, face detection and few-shot detection.