Marcus
Brubaker
Department
of
Electrical
Engineering
and
Computer
Science,
York
University
Research
Director,
Borealis
AI
Modelling and synthesizing image noise is an important aspect in many computer vision applications. The long-standing additive white Gaussian and heteroscedastic (signal-dependent) noise models widely used in the literature provide only a coarse approximation of real sensor noise.
This talk introduces Noise Flow, a powerful new noise model based on recent normalizing flow architectures. Noise Flow combines well-established basic parametric noise models (e.g., signal-dependent noise) with the flexibility and expressiveness of normalizing flows. The result is a single, comprehensive, compact noise model containing fewer than 2500 parameters yet able to represent multiple cameras and gain factors. Noise Flow dramatically outperforms existing noise models, with 0.42 nats/pixel improvement over the camera-calibrated noise level functions. Noise Flow represents the first serious attempt to go beyond simple parametric models to one that leverages the power of deep learning and data-driven noise distributions.
This is joint work with Abdelrahman Abdelhamed and Michael Brown (York University) that was presented at ICCV2019.
Depending on time, I will also discuss some other research projects, including theoretical aspects of normalizing flows and the estimation of atomic resolution protein structures from electron cryomicroscopy images.
Light refreshments will be available.
Bio: Marcus Brubaker is an Assistant Professor at York University and the Research Director of the Toronto and Waterloo offices for Borealis AI. He received his Ph.D. in Computer Science from the University of Toronto in 2011 and worked as a postdoctoral researcher at Toyota Technological Institute at Chicago and University of Toronto before joining York University in 2016 and Borealis AI in 2018.
He has been involved in a number of startups including as the co-founder of Structura Biotechnology Inc. His interests span computer vision, machine learning and statistics and he has worked on a range of problems including protein structure determination, human motion estimation, Bayesian inference, ballistic forensics, and vehicle localization for autonomous driving.