PhD Dissertation Defense | Deep Learning for Real-World Object Detection

About The Talk

Despite achieving significant progresses, most existing detectors are designed to detect objects in academic contexts but consider little in real-world scenarios. In real-world applications, the scale variance of objects can be significantly higher than objects in academic contexts; In addition, existing methods are designed for achieving localization with relatively low precision, however more precise localization is demanded in real-world scenarios; Existing methods are optimized with huge amount of annotated data, but in certain real-world scenarios, only a few samples are available.

In this dissertation, we aim to explore novel techniques to address these research problems to make object detection algorithms practical for real-world applications. The first problem is scale-invariant detection. Face detection is used as a benchmark here and we propose ``Feature Agglomeration Networks" (FAN) to build a new single stage face detector. The second problem is high-quality detection, and we propose two frameworks, anchor-based detector ``Bidirectional Pyramid Networks'' (BPN) and anchor-free detector KPNet to study it. The final problem is few-shot detection, and we propose two meta-learning based detector: Meta-RCNN and ``Meta Contrastive Detector'' (MCD) to study it. We demonstrate the proposed techniques address the discussed problems and show significant improvement on real-world utility.

Speaker Biography

Xiongwei WU is a Ph.D. candidate in Data Management at School of Information Systems, Singapore Management University. He is advised by Associate Professor Steven HOI. In his Ph.D. study, His research direction focuses on deep learning based object detection problems, including face detection, few-shot detection and generic object detection.