Seminar by Mr. Stanislaw Szymanowicz from Department of Engineering Science, University of Oxford

3D Reconstruction -- recovery of the 3D structure of the world from 2D observations -- is a fundamental task that has applications across Computer Vision, Computer Graphics and Robotics. Photorealistic 3D reconstruction has seen remarkable progress since the popularisation of Neural Radiance Fields, however, it still critically relies on challenging assumptions of capturing many photos per scene and precise knowledge of the associated camera poses. This talk will cover recent progress in relaxing these assumptions and will show a bridge between 3D Reconstruction and 3D Generation by leveraging progress in Diffusion Models, Multi-View Neural Reconstruction and Synthetic Data. Finally, at the end of the talk I will briefly describe how one can controllably animate the reconstructed objects in real-time.

Stan Szymanowicz is a second-year PhD student at Visual Geometry Group at the University of Oxford working with Andrea Vedaldi and Christian Rupprecht. His research interests span 3D Computer Vision, Computer Graphics, Machine Learning and Robotics. Previously, he spent 12 months as a Research Intern at Microsoft Mixed Reality Labs in Cambridge. He holds an MEng with Distinction and the IET `Student of the Year' Award from University of Cambridge where he worked under the supervision of Roberto Cipolla.