This book brings together three tutorial introductions to topics that shape modern computer vision: point cloud deep learning, deep photometric stereo, and automated machine learning. Each chapter is written by a researcher active in the field and explains its subject from first principles, so that readers can reach current work without first assembling the background from scattered papers. Two of the chapters approach three-dimensional structure from opposite directions, one asking how shape can be recovered from shading and the other how three-dimensional data should be represented and learned from; the third steps back to ask how the design of learning models can itself be automated.

The chapters began as tutorial lectures organized by the Special Interest Group on Computer Vision and Image Media of the Information Processing Society of Japan, and they retain the character of those lectures. Each is self-contained and assumes no knowledge of the others, so the book can be read straight through or consulted for a single topic. Three further tutorials appear in the companion volume.

Graduate students and newcomers entering any of these areas will find a clear and structured introduction, while researchers and engineers working in one part of computer vision can use the book to gain a working understanding of other areas of computer vision.

Shohei Nobuhara is a Professor of Computer Vision at Kyoto Institute of Technology, Japan. His research interests include computer vision and pattern recognition, with a particular focus on 3D human behavior analysis, 3D animal shape and motion capture, and physics-based vision. He has built several generations of multi-view capture studios and used them to reconstruct the three-dimensional shape and motion of people in action. He also works with unconventional imaging devices, including time-of-flight, polarization, and event cameras as well as catadioptric systems, and on the calibration problems that these devices bring with them.