Photorealistic Reconstruction from First Principles

Explore a seminar on photorealistic reconstruction from first principles in computational imaging. Delve into the comparison between compressed sensing and deep learning approaches for solving inverse problems in image reconstruction. Learn about a novel method that combines aspects of both approaches to recover optical density and view-dependent color from calibrated photographs. Discover how this technique bridges the gap between compressed sensing and deep learning by using non-neural scene representation, optimization through nonlinear forward models, and memory-efficient compressed representations. Gain insights into the preliminary convergence analysis suggesting faithful reconstruction under the proposed modeling. Presented by Sara Fridovich-Keil, a postdoctoral scholar at Stanford University, this talk offers valuable knowledge for those interested in computer vision, graphics, and advanced computational imaging techniques.

DSI | Photorealistic Reconstruction from First Principles