Please join us online for the Vision External Speaker Series Talk.
Abstract: Neural scene representations have had a significant impact on 3D computer vision, seemingly freeing deep learning from the shackles of large and curated 3D datasets. Nonetheless, some of these techniques come with strong assumptions. For example, they may assume that the scene remains motionless throughout the capture process, and that the training and test viewpoints are similar. In this presentation, I will delve into two methodologies addressing these challenges. The first approach (RobustNeRF, a CVPR 2023 highlight) draws inspiration from statistically-robust estimation to devise a loss function that selectively ‘discounts’ areas of data lacking photometric consistency; this is achieved through a simple modification to the loss function. The second technique (to be soon available on arXiv as CertaiNeRF) quantifies the uncertainty inherent in the acquired 3D scene via fast algebraic post-processing. By doing so, it addresses the issue of ‘floaters’ that commonly afflict NeRF models when cameras venture beyond the training view distribution.
Bio: Andrea Tagliasacchi is an associate professor at Simon Fraser University (Vancouver, Canada) where he holds the appointment of “visual computing research chair” within the school of computing science. He is also a part-time (20%) staff research scientist at Google Brain (Toronto), as well as an associate professor (status only) in the computer science department at the University of Toronto. Before joining SFU, he spent four wonderful years as a full-time researcher at Google (mentored by Paul Lalonde, Geoffrey Hinton, and David Fleet). Before joining Google, he was an assistant professor at the University of Victoria (2015-2017), where he held the Industrial Research Chair in 3D Sensing (jointly sponsored by Google and Intel). His alma mater include EPFL (postdoc), SFU (PhD, NSERC Alexander Graham Bell fellow) and Politecnico di Milano (MSc, gold medalist). His research focuses on 3D visual perception, which lies at the intersection of computer vision, computer graphics and machine learning.
