When: Thursday 15th of May, 1:00pm AEST
Where: This seminar will be partially presented at the ACFR seminar area, J04 lvl 2 (Rose St Building) and partially online via Zoom. RSVP
Speaker: Jason Lai
Title: Compare the Pair: Change Detection with High-Fidelity Representations
Abstract:
Detecting changes in a scene is an essential capability for autonomous agents operating in complex, unconstrained environments. By understanding relevant scene changes, robots can not only assess evolving conditions, but also replan their actions accordingly. This is essential for deploying them in high-stakes tasks such as inspections and maintenance of critical infrastructure. In this talk, we will examine key challenges in robust scene change detection. View diversity between successive inspection rounds limits methods that rely on precisely aligned image pairs. Change detection algorithms need to also distinguish semantically meaningful changes from visual distractors that fool naïve approaches such as lighting shifts and seasonal variations. Finally, we explore how view dependent radiance field representations can be used to construct a more complete and well-conditioned understanding of scene changes.
Bio:
Jason Lai received his bachelor degree in Mechatronics and Science at the University of Sydney in 2023. Currently, he is doing his PhD at the Australian Centre for Robotics working on robust change detection with radiance field representations. His research interests include computer vision and helping robots understand what they see.