Hosted by the Department of Electrical & Systems Engineering (ESE)
Abstract: Implicit neural representations (otherwise called neural fields, neural radiance fields/NeRF) are an emerging type of neural networks that can be used in physics-based computer vision and imaging applications. Their main application has been in the area of traditional image synthesis and 3D reconstruction where they are used to create stunning visuals and graphics. In this talk, I will present applications of implicit neural representations to tomographic imaging problems: namely 4D computed X-ray tomography and synthetic aperture sonar. These problems have applications for medical/scientific imaging and remote sensing. I will discuss how these networks can be used in analysis-by-synthesis optimization without the need for a training dataset or supervised learning to solve inverse imaging problems, and how they can be used to improve image quality. This work will highlight the importance of encoding physics-based image formation models into neural networks for computational imaging in the future.
For inquiries contact ese@wustl.edu.