Open sersorrel opened 5 years ago
Aye, I have used a similar fix for similar things in the past (processing images from a DSLR). Thanks for reporting the problem. I will dig out my patches for this. We can't just ship this immediately, as it will have a performance impact that needs assessing.
In RoboCon 2018, we ran into an issue with libkoki when using the Raspberry Pi camera module, which can produce significantly higher-resolution images than the webcams SR typically use can. libkoki would intermittently fail this assertion:
https://github.com/srobo/libkoki/blob/5b3c679c1f84b8fae9ab3cc5a93125dc58643a85/src/labelling.c#L277-L278
The fix we used is systemetric/libkoki@d84c77f119ff6f98f5862d0ef1b5200464228227, which just changes
label_t
from a 16-bit uint to a 32-bit uint and bumpsKOKI_LABEL_MAX
accordingly. I had (and still have) no idea if that's the correct thing to do, or if anything else needs changing as a result of that, but it seemed to work for us :)No opinion on whether you should take this change or not, mostly just wanted to document an issue that we ran into.
(before shelling out for better cameras for SR, it's probably worth checking if high-resolution images actually help the marker recognition at all; I don't remember what the conclusion from my year of RoboCon was, but from what I've heard from this year, it's not clear if increasing the resolution actually has any effect apart from making it slower)