Talk Abstract: Turning a collection of 2D images into a 3D model is a challenging inverse problem, especially when no information is given about the relative position between the observer and the scene. Solving this problem holds the potential of unlocking new capabilities like reconstructing a 3D scene from images captured in the wild, inferring the structures of biomolecules from electron microscopy images or their dynamics from X-ray diffraction patterns. Reconstruction methods are still hindered by 3 challenges: (1) the non-convex nature of the problem, (2) the presence of high levels of noise and (3) the sizes of real datasets, which can contain up to 10M images. In this talk, I will show how computational techniques, partly inspired from machine learning, can overcome these challenges and bring us from 2D observations to 3D models, and beyond when taking motion into account. My focus will be on two real-world applications: reconstructing the structures of proteins in cryo-electron microscopy (cryoAI, cryoFIRE and DRGN-AI ) and training a neural field from unposed images (MELON).

Speaker Biography: Axel Levy is a fourth year PhD student in Electrical Engineering at Stanford University. He is advised by Prof. Mike Dunne (director of LCLS at SLAC National Lab) and Prof. Gordon Wetzstein (head of SCI). His research focuses on solving 3D reconstruction problems in unknown-view setups. Most of his work addresses the problem of 3D molecular reconstruction from cryo-electron microscopy images. Prior to his PhD, Axel graduated from the Ecole Polytechnique (France).