Separate FR5 cover project from my UR5e faceplate thread. Same uncomfortable problem: small presentation changes move bright streaks, but example labels already call them scratches. I want the physical sample decisions and imaging conditions established before anyone tunes around them.
Who has the actual covers now, Alex? I'd make sure the numbered pieces stay available while quality and the vision team compare them. I mean the physical pieces, not just a shared folder that everyone assumes somebody else can explain.
Our stores once returned reference parts to ordinary stock because the box only said samples. Put an owner and purpose on these while they're waiting. Otherwise the next person tidying the bench can erase your best comparison without knowing it.
Only the box label, Lin. We found two again but couldn't reliably connect them to the photographs. That's why I care about the individual IDs, not just a better cupboard name.
Sam, I have four numbered covers in a retained-sample box. Jess Burke, they won't go back into stock; I'll keep that explicit. The scratch labels still lack quality's physical judgement.
Our image set once treated a reseated part as a new example. Group views by cover before anyone votes on labels; the contradiction may otherwise stay hidden.
Are those groups preserving the original images, Imran, or replacing them with cropped comparison pictures? I would want to trace a label back to what was actually seen.
Please keep that link visible in the training material too. A crop can be useful for a lesson, but somebody later needs to know where to look on the complete cover. We once trained a very reliable response to an arrow.
The imaging comparison needs the actual holder condition as well. I would inspect how the cover seats before spending money on another lamp, though an unreliable seat and unsuitable illumination can both be present.
We changed a lamp and a locating stop together, then nobody could tell which change helped, so I'd keep those comparisons separate even if the combined result produces nicer pictures.
Use the known defect locations as part of that comparison. Removing a reflection is helpful only if the chosen condition still exposes the surface damage you need to detect; I would not judge it by image neatness alone.
Quality has examined the four covers. Two are accepted, one is rejected for a dent and one remains disputed over a faint surface line. I've kept the disputed cover out of the reference answers, not out of the investigation.
Zara, would a dent reject be enough for Alex's scratch work? I would want a confirmed scratch too, or rather a reference for the actual defect class the proposal needs to detect.
No, I would ask quality for a confirmed scratch reference as well. The dent may remain useful for another check, but it does not establish scratch sensitivity or resolve the faint line.
Thanks for separating those, Sam and Zara. That is exactly the distinction a trainer could lose when a folder is simply called rejects and all the parts look shiny.
Lin, quality's appearance lead owns the faint-line decision. Sam and Zara, I've asked for confirmed scratch examples across the intended finish range. We haven't started the controlled holder and lighting comparison yet.
Keep the four existing covers available when that starts. They already show why the proposed method needs work, even though not all four qualify as settled reference answers.