ECE Seminar: Anagh Malik
- Starts: 11:00 am on Thursday, September 19, 2024
- Ends: 12:30 pm on Thursday, September 19, 2024
Title: Multiview Scene Reconstruction Using Transients
Abstract: Current methods for scene reconstruction commonly rely on conventional images or depth maps, which discard the rich information about the time of arrival of photons at the sensor. In contrast in this talk, we propose methods for learning scene representations using videos captured with an ultrafast camera. These images, called transients are completely invisible to the naked eye, but reveal complex light transport phenomena such as a pulse of light propagating through liquid or rays of light reflecting off a mirror. The merger of multiview reconstruction algorithms and transient data gives us a glimpse into exciting new methods that can recover objects from just indirect reflections or learn complex material properties. I will discuss recent progress towards this goal, including techniques for transient-based multiview reconstruction, novel view synthesis of propagating light, and estimating material properties using transients.
Bio: Anagh Malik is a 2nd year Computer Science PhD student at the University of Toronto, supervised by David Lindell. His work lies at the intersection of machine learning, computer vision, and computational imaging. In his PhD, he has mainly tackled the topic of multiview transient imaging, attempting problems such as 3D reconstruction or light-in-flight visualization.
- Location:
- PHO 339