3D Vision Reading Group | 3D Vision Reading Group webpage.
Description: website
Search Menu Expand Document (external link) nerf-course.github.io Home Global Models Local/Hierarchical Models Inverse Rendering Fundamentals Neural Radiance Fields Neural Light Fields Image Based Rendering Multi Resolution Fast Training Camera Extrinsics Learnable Appearance This site uses Just the Docs , a documentation theme for Jekyll. NeRF Starter Kit: a conceptual journey Global Models DeepSDF and OccNet are of the pioneering works that attempted at representing 3D scenes using implicit functions with
Left: An occupancy function can be modeled as the decision boundary for classifying inside and outside points. Right: The decision boundary corresponds to a surface in 3D.
The goal here is to learn a function to map a 3D point into a continuous function which models the inside and outside of an object. OccNets trains an MLP to classify the points into inside, outside. The learnt decision boundary function can then be turned into a mesh using their introduced MISE method and a hyperparameter which thresholds the surface. MISE basically first hierarchically zooms in to the surface by gridding. Then runs a marching cube and refines it using first and second order gradients. The
3D Vision Reading Group | 3D Vision Reading Group webpage.