Why open the colmap course
In 2019, after graduating with a master's degree , he joined UISEE Technology to work on high-precision map algorithms. In his spare time, he started his own journey of sharing technology on station B. The early videos are as follows:
But after posting these few videos, I found that every time I record, there will always be I missed what I wanted to say, and couldn’t express myself well, so I stopped sharing for a while ( another reason is the private message of station B, some people directly asked me to help them deal with the colmap problem immediately in an orderly tone, even chills and anger ).
At the end of August 2020, I made a career decision in my life: I resigned and returned to Chengdu from Shanghai; about three months after I returned to Chengdu, the current person in charge of Pudu Technology’s Chengdu area organized a Chengdu slam exchange conference in first-tier cities Because there are many slammer practitioners, there are countless such exchange meetings, but for Chengdu, there are very few people engaged in 3D vision-related work compared to the first-line, so this is also the first offline slam exchange meeting in Chengdu. So he volunteered to share the knowledge related to 3D reconstruction.
Regarding the work back to Chengdu, it belongs to an enterprise in this industry, and it also comes back with the mentality of building this industry. It is still engaged in work related to 3D reconstruction (I have been doing photogrammetry, and my own philosophy is: in today’s society , If you want to do something, do it vertically, not horizontally, and do one thing to the extreme. Although the road ahead may be narrower, the distance to become an industry expert will be closer ). During the time I returned to Chengdu, I continued to feed videos to station B, but changed the sharing format. I am not participating in it, and only share the video of the results, as follows: In addition, I also shared
an open class "structure from motion", breaking the principle of not revealing your voice before, and it is also the beginning of officially becoming a guest of the 3D Vision Workshop Planet
At the beginning of April 2021, I felt that it was necessary to share the colmap code splitting and explanation publicly on station B. There are two reasons:
(1) There is currently no information on the market explaining sfm+mvs (mainly on sfm), although The openmvg framework has existed for more than 10 years, but there are not many related materials, let alone the information of the SfM SOTA-colmap framework that was born in 2016 (of course, the classic framework of mvs, openmvs, is available in the six brothers of computer vision life . Explanation)
(2) Simply contribute your own strength, although photogrammetry does not need me to promote, I do not have this ability, I just think about how much I can burn my love (I remember that when I was studying for a re-examination, one person provoked The following dozens of professors said: "Only learn photogrammetry, and don't learn anything else", and they have firmly established their own path). To sum up the above two reasons, there is this video on station B:
At the end of April 2021, after communicating with the 3D Vision Workshop , it was decided to open a colmap course (due to the previous "B station incidents", Things that make me feel free can only make some people take it for granted and bully you without a bottom line, so the course that I planned to teach colmap for free in early April was put on hold and became a paid course (a hot pot meal); in addition Later, some people added me to WeChat from station B to ask me about colmap questions. After I answered them, they would send some red envelopes as a thank you. When I learned that they were students, I told them: no money.
The harvest of the course
The joy of sharing is not only about passing on knowledge, but also a process of self-review, summary and improvement. During the preparation of the course, because some formulas have to be deduced, it has become commonplace to go to bed late. Although the whole process is very tiring, it has gained a lot, so I also strongly recommend that if you make something in your own direction, you can use similar The platform of the 3D Vision Workshop can pass on what you know, that is, share knowledge, improve others and yourself, and bring benefits to yourself (in the era of knowledge fees, don’t always think about whoring for nothing), why not not for.
Insufficiency of the curriculum
Of course, if the full score of this course is 100, I will score 85 points myself, because it is not friendly to many beginners, which may have a lot to do with my original intention (transfer photogrammetry, practitioners in this industry). Of course, I think the extra points are: there are many improved solutions and ideas, and some improved and implemented codes
at last
This Sunday (2022.05.22 tentatively scheduled for 8:30 pm) will be the last course live broadcast, explaining the big homework assigned: colmap fusion gps (live broadcast code implementation), although the incremental sfm of colmap is classic, but the disadvantage of incremental sfm: error The accumulated drift source code has not been avoided (although openmvg has already implemented this function, colmap has not implemented it), there are many ways to solve the dirt in large scenes, such as: (1) use high-dimensional features (2) use other
sensors
, Such as gps, imu, etc.
Then using gps-constrained sfm (photogrammetry term: gps-assisted aerial triangulation) can not only restore the scale, but also solve the drift problem and enhance the stability of the system.
The colmap framework is as follows: 
Regarding the second course of visual reconstruction colmap, I also listed some days in the catalog. Compared with the first period, it is more comprehensive and hard-core, but it has not been carried out for the following reasons: (1) Comparison of work and
things Busy
(2) I am not ready yet, maybe it will end in the first issue?
(3) It is impossible for the course to cater to everyone’s tastes. Just like the students in the current course, some of the students in the current course think that what I say is good and inspires them a lot, and some of them must think that what I say is not good. But the course is not to satisfy one person, so it becomes one-on-one. There are various projects, and everyone’s needs are almost different. up, to put it bluntly: I have paid so much tuition for four years in college, and the teacher has not taught me it for four years. It is too unrealistic to rely on an online class for a few months to learn), the catalog of the second phase is as follows: the last link,
I Use the ppt on the last page of the master's graduation defense that year - a page called the true temperament by the tutor to continue to encourage myself
