Chapter 13: 3D Reconstruction
Creating 3D models from images using photometric stereo, 3D scanning, surface representations, and neural approaches like NeRF.
Chapter Overview
The ultimate goal of geometric computer vision: recover the 3D world from 2D images. This chapter brings together everything we've learned to create complete 3D models.
What is this chapter about? We explore different approaches to 3D reconstruction: using lighting variations, active sensors, and surface representations. We also introduce neural approaches that learn to represent 3D scenes.
Why does this matter? 3D reconstruction enables:
- Cultural heritage: Digitizing historical artifacts and architecture
- E-commerce: 3D product visualization for online shopping
- Gaming/VR: Creating realistic 3D environments
- Manufacturing: Quality inspection and reverse engineering
How the topics connect: We start with photometric stereo—recovering shape from lighting variations. 3D scanning uses active sensors for precise measurements. Surface representations show how to store and manipulate 3D shapes. Finally, neural radiance fields (NeRF) represent a paradigm shift—learning implicit 3D representations.
Chapter Roadmap
Click any topic to jump in
Photometric Stereo
Recovering surface normals from varying illumination — capturing fine geometric detail that stereo methods miss.
3D Scanning
Active depth capture via structured light and ToF sensors — direct, precise 3D measurement for real-world objects.
Classical and neural representations
Surface Representations
Storing 3D shapes as meshes, point clouds, or SDFs — the data structures that make 3D geometry computable.
Neural 3D Representations
Encoding scenes in neural network weights via NeRF — implicit functions that enable photorealistic novel view synthesis.
Sign up to unlock this chapter
This chapter is part of PixelBank Premium. Create a free account, then upgrade to read the full lesson — concepts, walkthroughs, and exercises.