This seminar is open to the public. No registration is required but seating is limited.
Zan Gojcic
Research Manager, NVIDIA Toronto AI Lab
Tuesday, Jan. 30, 2024
3 p.m. – 4 p.m. (ET)
BA 5187 (DGP Seminar room)
Title: Hybrid Rendering: Combining Volumetric and Surface Rendering to get the Best of Both Worlds
Abstract:
Neural radiance fields have emerged as a powerful 3D representation enabling photorealistic novel-view synthesis and reconstruction. However, the high visual fidelity comes at a high computational cost, as the volume rendering formulation requires a large number of samples along each ray. On the other hand, surface based rendering methods are very efficient to render, but may lack the high visual fidelity, especially in regions that are more fuzzy and volumetric. Recently, several works in the field of text-to-3D have shown how the two can be combined into a two-stage pipeline that yields improved visual fidelity and high resolution details. In this talk, I will allude to further possibilities of combining the two formulations in order to get the best of both worlds.
Bio:
Zan Gojcic is a research manager at the Nvidia Toronto AI Lab, where his team is exploring topics related to neural reconstruction and data driven simulation. Prior to joining Nvidia, Zan received his PhD from ETH Zurich. His thesis titled “Benefiting from local rigidity in 3D point cloud processing” was awarded ETH Medal - an award bestowed upon outstanding doctoral theses. During his PhD, Zan has also visited the geometric computation group led by Leonidas Guibas at Stanford University. His research interests are focused around 3D neural reconstruction, generative models and their combination.