I want accepted bright regions in our UR5e faceplate comparison as well as real defects. The present captions call moving streaks scratches, and I do not want the inspection scope reduced to whatever is easiest to see. What should establish the physical references and their correspondence with the camera images?
Give each retained faceplate an identity and record quality's decision for the actual region under the agreed appearance method. Keep original captures and orientation references with it. A rejected plate may still contain an accepted bright band somewhere else.
Quality has judged the retained plates now: broad bands accepted, a short score near one mounting hole rejected. The full images are linked, but the exported annotation lies beside the hole rather than on the score identified by inspection.
Do the annotation coordinates refer to the full capture or a resized or cropped view, Imran? Keep the existing files while checking that mapping instead of moving the arrow by eye until it looks plausible.
And check the physical orientation reference. If the same-looking holes repeat, the right coordinates on the wrong orientation can still point somewhere convincing.
Rebecca, yes, but don't make the inspector repeat the surface judgement just because an export disagrees. First establish whether the image and region mapping represent the same identified faceplate and orientation.
Rebecca, orientation matches the asymmetric edge feature. Anil, the annotations were created on a crop but exported as full-frame coordinates without the crop offset. That explains this displaced region, not the physical quality decision.
Keep that correction separate from defect visibility. An accurately placed region can still contain a score the present camera view does not reveal under all allowed presentation conditions.
Also verify more than this one region after correcting the export. Use known features at different image positions so the mapping check tests the coordinate transform, not one convenient arrow.
The corrected export agrees at the checked full-frame positions, and the rejected score's region now matches the physical reference. Thank you, Anil; that was an actual mapping bug. Our controlled lighting comparison still loses the score at one allowed presentation.
Then the reference mapping is repaired and the imaging candidate has a specific failure. Those are two useful results. Keep the failed presentation in the next comparison rather than letting the newly accurate annotation make the image look validated.
Who owns the copied teaching exports? I'd withdraw the misplaced-region version now. The next lighting trial can remain unfinished without leaving people a known wrong reference picture.