Populate classId and confidence level for object detection in Java simulation#2372
Populate classId and confidence level for object detection in Java simulation#2372shdalton wants to merge 2 commits intoPhotonVision:mainfrom
Conversation
…puted confidence level in Java (while building the PhotonTrackedTarget)
github-actions
bot
added
the
photonlib
| float conf = tgt.objDetConf; | ||
| if (classId >= 0 && conf < 0) { | ||
| // Simulate confidence using sqrt-scaled area for a more realistic | ||
| // curve. Raw areaPercent/100 is tiny for most targets; sqrt scaling |
There was a problem hiding this comment.
Applies to all 3 languages but can we move this to the target javadoc? So people know "either set confidence yourself, or the sim will guess at it based on area"
There was a problem hiding this comment.
Yes I did implement it for all 3 languages, would you like me to put a function header in all 3 languages for the process function? I apologize I'm not sure exactly what you mean by "move this to the target javadoc". While at it, would you happen to have an idea why my change fails for orangepi5pro (I only have an orangepi5plus here... maybe I can try to duplicate the problem)
There was a problem hiding this comment.
perhaps you meant add some documentation to the AddVisionTargets or even just the VisionTargetSim class
2 participants
Description
compute confidence level based on target area in total image size and populate classId and confidence level in Java (while building the PhotonTrackedTarget)
What changed? Why? (the code + comments should speak for itself on the "how")