Wednesday, December 7, 2022 - 2:00pm to 3:00pm
Location:In Person Traffic21 Classroom, Gates Hillman 6501
Speaker:HAITHEM TURKI, Ph.D. StudentComputer Science DepartmentCarnegie Mellon University https://haithemturki.com/
Towards City-Scale Neural Reconstruction
Recent advances in neural rendering techniques have lead to significant progress towards photo-realistic novel view synthesis, with Neural Radiance Fields (NeRFs) in particular drawing significant attention and spawing a wide body of follow-up work. However, until recently, the vast majority of NeRF-based approaches explore single-object scenes captured indoors or from synthetic data, thus limiting their practical utility. In this talk, I will first discuss how we scale NeRFs to capture and visualize urban-scale environments from drone footage orders of magnitude larger than prior work. I will then discuss recent efforts to build dynamism and semantic knowledge into our model, culminating in the largest city-scale dynamic representation to date. Presented in Partial Fulfillment of the CSD Speaking Skills Requirement.