SPS Webinar: Differentiable Uncalibrated Imaging
Based on the IEEE Xplore® article with the same title published in the IEEE Transactions on Computational Imaging, December 2023
The original article is open access and freely available to all for download by clicking here.
About this topic:
The presenters will present their paper 'Differentiable Uncalibrated Imaging' where they propose a differentiable imaging framework to address uncertainty in measurement coordinates, such as sensor locations and projection angles.
Their approach leverages implicit neural networks and spline-based neural fields to perform joint calibration and image reconstruction. This makes it possible, for example, to use pre-trained neural networks to configurations different from those for which they were originally trained.
They will begin by presenting the general optimization framework underlying their method. Next, they will illustrate their approach to 2D and 3D computed tomography, and show that it produces improved reconstructions compared to baselines that do not account for the lack of calibration.
Finally, they will explore how this framework extends to the challenging problem of 3D cryo-tomography reconstruction, tackling scenarios with misaligned and noisy input projections.
About the presenters:
Sidharth Gupta received the B.A. and M.Eng. degrees from the University of Cambridge, Cambridge, U.K., in 2014 and the Ph.D. degree in electrical and computer engineering at the University of Illinois at Urbana-Champaign (UIUC), USA, in 2022, where he was advised by Ivan Dokmanić.
He is currently a Machine Learning Scientist at Amazon, London, UK. His primary focus was on designing algorithms and techniques for solving image reconstruction problems. Particularly interesting applications that he looked at were imaging through paint, seismic imaging and xray medical imaging. He has also experimentally demonstrated that some of the imaging methods that he developed can enable optical computing which is to do with using light to perform mathematical operations.
Valentin Debarnot received the Ph.D. degree from the University de Toulouse, France, working on using machine learning to accurately describe the optics of fluorescence microscopes.
He currently holds a junior professorship at CREATIS, University of Lyon, France. From 2021 to 2024, he was a postdoc at the University of Basel, Switzerland, where he worked on machine learning for inverse problems. His main interests include the development of machine learning techniques for image reconstruction and analysis. He is also interested in numerical and theoretical aspects, as well as the precise modeling of acquisition systems. Among various applications, he has been particularly involved in 3D cryo-tomography, 2D tomography and fluorescence imaging (blind-deblurring, super-resolution).
Dr. Debarnot, among various applications, has been particularly involved in 3D cryo-tomography, 2D tomography and fluorescence imaging (blind-deblurring, super-resolution).
