Time & Locatio: Tuesdays & Thursdays 2PM - 3:30PM PT BWW 1215
Quick Links: Ed Discussion| Self-report Participation Grading | Paper Schedule | Summary Template
Course Description
We live in a world that is 3D and dynamic. How can we perceive the underlying physical world from a set of photographs or video? Although there have been remarkable advances in 2D vision tasks such as image recognition, detection, and segmentation, the 2D nature of these tasks make perception of the underlying 3D world from images a challenge. While 3D vision has always been an integral part of Computer Vision, with the advances in deep learning, there has been a growing new interest and possibility in 3D vision and rapid developments in this area in the past couple of years. The 3D nature of this topic has many potential applications in graphics, robotics, content creation, mixed reality, biometrics, and more.
The goal of this course is to provide a historical and technical context to understand the latest trends and techniques in learning based 3D vision. After taking this course, one should:
Know the most common 3D representations, their pros and cons
Be able to read latest 3D vision papers critically
Learn how to present effectively & engage in discussions
Understand what the interesting next problems are in this domain
Incorporate 3D inductive bias in your own research.
The format of this course will be a mix of lectures, seminar-style discussions, and student presentations. Given the virtual nature of this semester, the course will be heavily discussion oriented. Students will be responsible for paper readings, class presentation, class participation, a few programming assignments, and completing a final project.
Email: kanazawa<at>berkeley.edu
Office Hours: Thursday after class
Note: This 3D reconstruction result is from our ICCV'19 PIFu paper, obtained automatically from a *single* image (including the unseen regions & its color). Yes, the resolution & skirt can be improved and you may have an idea of how to do it after taking this class!
Learning Resources:
As this is a rapidly developing topic, there is no textbook, but if you are looking for technical references in 3D and geometry, these are the classics:
Hartley and Zisserman, Multiple View Geometry in Computer Vision
Ma, Soatto, Košecká, Sastry, An Invitation to 3-D Vision: From Images to Models
Richard Szeliski, Computer Vision: Algorithms and Applications, 2nd Edition, Chapters 11-14
A full lecture version of the course offered by Shubham Tulsiani at CMU.