Explore cutting-edge techniques for appearance acquisition in digital 3D content creation through this seminar talk by Sai Bi from UC San Diego. Delve into methods for recovering high-quality texture maps, reconstructing meshes with per-vertex BRDFs, and learning volumetric representations for joint view synthesis and relighting. Gain insights into RGB-D reconstructions, multi-view stereo, SVBRDF prediction, and mobile phone captures with flashlight. Discover the latest advancements in volumetric rendering, neural representations for scenes, and integration with physically-based renderers. Understand the challenges and solutions in reproducing the appearance of real-world objects and scenes for virtual and augmented reality applications.
Appearance Acquisition for Digital 3D Content Creation
Overview
Syllabus
Intro
Geometry + Material
Image-based appearance acquisition
RGBD reconstructions
Projective texture mapping
Previous works
Our approach
Observations
Similarity: coherence
Similarity: completeness + coherence
Consistency
Patch-base energy function
Multi-scale optimization
Comparison against single-view selection
Acquisition setup
Learning-based multi-view stereo
SVBRDF prediction
Geometry reconstruction
Volumetric representations
Relightable reconstructions
Joint view synthesis and relighting
Mobile phone captures with flashlight
Discretized volume rendering
Learning deep reflectance volumes
Loss functions
Comparison to mesh-based methods
Comparison on synthetic data
Environment map rendering
Physically-accurate volume rendering
More results
Integration with a physically-based rendere
Sparse geometry and BRDF acquisition
Neural representations for scenes
Generalizable neural representations
Integration with traditional rendering
Taught by
Andreas Geiger
