4 ms·
I'm going to hijack this thread to mention that I'm thinking of open sourcing a SLAM system I've been working on under BSD or MIT. Ive been working on this bec
by 0xfaded 9y ago
I'm going to hijack this thread to mention that I'm thinking of open sourcing a SLAM system I've been working on under BSD or MIT.
Ive been working on this because I'm interested in underwater drones (uuvs), and hope to commercialise this in the form of services.
My current understanding is as follows:
- most open source systems come from universities and are GPL, with the option for licensing. (PTAM, ORB SLAM, DSO, SVO, LSD)
- Optimistically, it would require $1-3 million to build a compedative system privately. The sky is the limit here.
- No one, neither the universities nor private entities, is making serious money by licensing just a SLAM system. The well performing startups are somehow coupled to a product or a service.
- All systems basically solve the same problems, but mix and match different solutions. For example for initial motion estimation, you could rely on constant motion (DSO, ORB SLAM), image alignment (SVO), or solving directly for rotation between cameras (GSLAM).
- As a researcher, to implement a new idea, you either need to implement an entire system from scratch, or modify a GPL'd system. This is great in theory, but since everyone is playing the GPL or pay game they are reluctant to accept contributions.
I'm imagining something like opencv, with implementations of common algorithms accompanied by wiki pages describing high level details. Components could be mixed and matched into different reference SLAM systems targeting different use cases, do you want something that will run on a MAV or a self driving car?
From a commercial point of view, companies could worry more about tweaking things to work for a particular use case without needing to start from scratch. I believe that universities, with their many departments, have more opportunity than anyone else.
My belief is that it's time for a liberally licensed SLAM system. Has anyone come to a similar conclusion or care to comment to the contrary?
Thanks HN
- stealthcat 9y agoCan companies really get away for not licensing properly with GPL? I mean selling your product without buying the commercial license. How to get yourself caught? Different SLAM solves the same problem anyway.
- singron 9y agoIt sounds like they own the copyright and are dual licensing. However they don't accept contributions (without a copyright assignment) since they wouldn't be able to dual license the contribution.
- the_mitsuhiko 9y agoWe (sentry.io) open source everything as either BSD/MIT or Apache2 and we not only get away with it but also think it’s the best way :)
- kureikain 9y agoDo the code to handle paymet system, user management, plan etc open source as well? I'm not really familiar with Python yet and try to find some example of these code. Thank you.
- the_mitsuhiko 9y agoNo. Nobody who runs sentry on prem needs that functionality.
- snovv_crash 9y agoI was thinking something similar. OpenCV has lots of pieces of the puzzle, but the quality is... inconsistent. For instance, I've had repeatable segfaults from doing RANSAC (which of course disappear in the debugger and Valgrind). There have also been some significant feature regressions since the 2.x --> 3.x migration, and the build system breaks every time MSVC upgrades. If you'd like to release something like this I'd be happy to contribute. I have a background in SLAM, embedded and photogrammetry, so this is something I could really get behind.
- B4TMAN 9y agoI was working with Opencv's python wrapper a few months back and I didn't find a lot of options in it. It maybe because the wrapper didn't include most of the things from native C++ code. Is there any proper way to start with Opencv?
- taneq 9y agoIf you have a robust SLAM method which is relatively computationally efficient, and you release it under an unhindered or attribution license, I think a whole bunch of people would use that. I know I would. I've been wanting to spend some time in that field myself but workload hasn't allowed it. Just out of curiosity, have you looked into DTAM (PTAM's successor) or RATSLAM? Both were on my shortlist for a future project.
- cup-of-tea 9y agoYou should definitely release it, but under GPL. Why do you think BSD or MIT are better?
- jlg23 9y ago"less politics" - and shouldn't you start by justifying your recommendation?