As someone who’s done pretty detailed testing on LL3 with both the official LL firmware and Photonvision (PV) firmware, I can confidently say that the two are not that different.
LL3 testing
LL2+ running on PV and LL OS comparison
2024 PV testing (I’ll add LL OS results to this once they exist)
You need to look at both FPS and range, and ideally, pose stability as well. LL can get a really high FPS with a super high black level, but your detection distance will like 6ft (or simply impossible).
LL OS running on LL3:
- amazing pose stability in multitag, poor absolute accuracy
- difficult to tune, with gain, black level, and exposure affecting detection speed and range. This is probably why your results look so different.
- best settings are 960x720 40fps, Downscale 2, which yields 26fps at 27ft of max range. 640x480 90fps, Downscale 2 yields 50fps at 15ft range
PV FW running on LL3:
- great absolute accuracy with CalibDB calibration (or the mrcal integration they have now), poor pose stability at long range. Good range stability though, weirdly.
- easy to tune, just set exposure low and brightness high until you see the image. The latter barely affects performance, but it improves range.
- best settings are 1280x960 Decimate 2, yields 12fps 25ft. 640x480 Decimate 2 yields 41fps 16ft.
The two are pretty close overall. I would use the LL OS for its superior pose stability at long range, but you would want to recalibrate the camera to get decent accuracy. If you only care about range and not total XY pose, Photonvision will perform better with calibration.
A good litmus test is to check your maximum range with each firmware. This is easy and allows you to compare results. If you only want short range, stick with the LL OS.