Nov 13 (Mon) @ 2:30pm: "Compositional Networks for Detecting and Localizing Activities," A S M Iftekhar, ECE PhD Defense
The development of automated methods capable of detecting and localizing actions is crucial for a variety of applications, ranging from surveillance and autonomous driving to content moderation. This thesis focuses on creating action detection methods that deliver robust performances. At the heart of these methods’ robustness lie two fundamental elements: the detection of atomic actions and the ability for compositional understanding.
Atomic actions are those that are identifiable from a single image or a short sequence of video frames. We have developed innovative methods to detect and localize such actions that achieve state-of-the art performance. The key strength of these methods is the spatial and semantic refinement of visual features, thus enabling the identification of the spatial regions that contain actions. For scalability, we further developed a multi-branch deep network architecture to recognize new composition of objects and actions. Our design ensures that each branch learns decoupled features, allowing the network to transfer previously learned concepts to identify new compositions. This approach outperforms existing methods by a good margin as our extensive experiments on benchmark datasets demonstrate.
A S M Iftekhar is a Ph.D. candidate in the Electrical and Computer Engineering department working with Professor Manjunath in the Vision Research Lab. His research focuses on activity detection and compositional learning, with applications spanning from content moderation to autonomous driving. He received his B.S. in Electrical and Electronics Engineering from the Bangladesh University of Engineering and Technology (BUET) in 2017 and an M.S. in ECE from UCSB in 2020. His Ph.D. journey has included internships at Amazon AWS AI Labs and the autonomous vehicle company Zoox.
Hosted by: Professor B.S. Manjunath
Submitted by: A S M Iftekhar <email@example.com>