Why do you want to do this column?
A major research goal of computer vision is to enable computers to recognize 3D environmental information through 2D images. In recent years, with the explosion of technologies such as SLAM, SfM+MVS, and NeRF and the booming development of related industries, more and more people have joined the field of 3D reconstruction.
However, getting started with 3D vision is difficult. In the process of communicating with my classmates, interning in companies, and sharing in online communities, I learned that quite a few people started working under the condition of "rushing ducks to get on the shelves": as students, they want to move horizontally quickly I want to finish the project and publish an article quickly to graduate; as a practitioner, I want to make a demo for investors, improve the open source code, and pass this month’s KPI... I may be able to cope with it in the short term, but if I want to go further, I may encounter to many questions. I also came into contact with this book under such circumstances. At that time, I found this direction by accident in a horizontal subject that I didn't understand. I tried various ways to modify the network to meet the graduation requirements, but I was not willing to accept it. Just using deep learning to violently fit to get an answer is just a state of "knowing it but not knowing why", so I really hope to understand the "principle".
Later, I found the book "Multiple View Geometry in Computer Vision", which was really a shame. I highly, highly recommend everyone to read the original book. If you have no difficulty reading English, I recommend reading the English version.
Some references
Here are some resources and references for reading this book:
- Original book download channel 1: r5-org
- Original book download channel 2: Oxford VGG
- Matlab code link: Oxford VGG
- Notes from a Zhihu boss: "Multiple View Geometry in Computer Vision" Reading Notes - Electro-optical Phantom Alchemy
- Notes on Github: baoguoma/note-for-MVG
What this column will contain
Starting today, I will organize this must-read book on 3D vision into notes by chapter and form a column. This book has more than 600 pages and a lot of content, so the notes cannot cover everything (otherwise it would be equivalent to copying all the contents). I only organize the important knowledge points and some thoughts, and match them with some code (if possible). I hope that the notes It can be used as supplementary reading material for you.