I’m trying to set up photonvision for the limelight 2 using the 2023.4.2 image.
The thresholding range is very wide for both hue, saturation, and brightness and photonvision still struggles to pick up the bright green (as indicated by the black rectangles on the reflective tape in the processed feed). Has anyone encountered this issue, and how might it be fixed?
Sorry, not a solution:
I’ve been having the exact same issue, pretty much waiting for someone to post about this. (idk why I haven’t myself, nor made an issue on their github). The weird thing to me is that it works just fine locally on windows.
I never solved it, so I switched back to using limelight software a few days ago. It only took about 5 mins. They both accomplish the same task, limelight software seems to be more stable (ex: this), and fps seemed to be better in my experience, though I’ve seen people say performance is the same or better using PhotonVison.
If this ever gets solved I wouldn’t mind switching back to it.
Yeah, I think there is a fundamental code bug on our end in this case, coz I’ve seen another report about Very Green targets blowing out our HSV thresholding GPU shader somehow. I had trouble debugging it though, still looking for the actual problem. Good tips in the meantime tho!
thanks for the tips, though I do think it’s a bug, as the green is pretty consistent and the threshold is really wide. I think I’ll just go with as what one of the other people said, just switch to the limelight software.
Pardon my ignorance - I have an older version of PhotonVision and have used it mostly for AprilTags. But why don’t your yellow threshold bars’ endpoint circles match the values printed to the left and right of the colored threshold areas? The ranges of the digital values are relatively small and apparently don’t include the full field of green values.