Our UR5e planning folder has bright-line pictures labelled scratch, but the sample sleeves beside the bench carry different identifiers from the image names. I want the physical faceplates matched to their decisions before we compare presentation and tune anything. Otherwise we could spend a decent lighting budget teaching the camera somebody's filing mistake
Match the original images to the retained faceplates with quality, using visible features and the original capture record where available, and leave any unassignable image unassigned rather than letting its filename supply the missing evidence.
Quality matched most of them. Two short labels were reused for different faceplates, and one image cannot be assigned. We've separated the physical decisions from the old captions; the unassigned picture isn't a threshold reference.
Keep the original images as well as the corrected set. Then compare a repeatable presentation with the accepted reflective examples and identified real rejects. A tidier folder shouldn't quietly remove the difficult surfaces.
One with a located scratch beside the corner fixing hole. The long central bright band on another faceplate has no corresponding physical defect. Both are retained, with their original full views and those separate decisions.
Check the required corner area, not just whether this scratch happens to be visible. Its presentation may need a different view from the broad face; include the handling needed for that view in the proposal.
The scratch is in frame but the holder lip hides part of the required corner. Tooling is assessing a revised support with the vision owner, using the same physical pair. We aren't cropping the awkward corner out of the job.
It changes the seated angle. We've treated the resulting pictures as a new presentation comparison, not improvements caused only by removing the lip. The corner is exposed now, but repeat loading remains to check.
The ordinary loader and a cover operator both tried the revised support with the identified pair. Seating was consistent and the required corner stayed visible. Quality checked the physical references again. We're keeping this presentation for the wider image study; the inspection rule itself isn't accepted yet.
Useful result. Add ordinary accepted finish variation to that wider study, alongside the awkward accepted band and real reject, so the selected presentation has to work beyond the first two well-known examples.