I'm preparing a UR5e inspection exercise around aluminium bezels and a fixed camera, and our example labels are already causing arguments. Move a bezel slightly and the bright streak moves. We haven't commissioned anything yet. How can I give the team references they can trust before they start adjusting thresholds?
I'd start with the physical bezels. Can somebody responsible for cosmetic acceptance examine them and record what each mark actually is? A clear photograph can still be a poor reference if its label came from a guess.
And keep the uncertain pieces. Sarah's right about the labels, but a set containing only obvious passes and obvious scratches can make the inspection look easier than it is. Put disputed examples aside for a decision.
We now have acceptance decisions against individual bezels, including a few still disputed. The moving streak was a reflection on two pieces, but there was a real mark beneath it on another. That rather ended the argument that movement alone proves there's no scratch
That mixed example is valuable. Use it when comparing lighting and presentation options. Record the conditions for every capture and check whether the actual mark remains visible through the presentation variation you expect.
We've started separate comparisons, Sarah. Thanks for the practical warning; I had put both adjustments on the same worksheet step. No threshold selected yet, and the disputed pieces are still awaiting a decision.
Separate comparisons are sensible here. Once an arrangement looks promising, check the combination as a whole too, including taking a bezel out and presenting it again. The commissioning evidence will need more than the best still image.