The same identified trim plate has three labelled images, and the alleged scratch occupies a different physical area in each. The camera settings are recorded, but the seating changed. Nobody can show me a corresponding physical scratch at all three marked locations.
I have preserved the originals and stopped those labels entering the training set. Before commissioning, I want quality's actual surface judgements linked to the images, then an imaging comparison using the intended presentation range. Which reference checks should be completed before anyone resumes threshold work?
Start with the actual plate and all three marked regions together; quality can then say which, if any, correspond to an unacceptable physical feature rather than inherit the image caption.
Quality found one unacceptable short scratch near the notch. Two bright central bands are accepted finish and reflection. The original captions marked those bands but missed the scratch; I have withdrawn all three as labelled training examples.
Keep the original images available for the lighting comparison, with the corrected physical findings beside them. The missed notch scratch may be the more demanding case even though the central bands attract attention.
Also retain the required inspection area in the imaging trial. Enlarging the notch detail may help this example while losing coverage of another region the job still requires.
The tutor identified the real scratch with quality and has replaced the lesson's arrows. We now have an accepted plate and this reject with matched physical findings. The first revised view reveals the notch scratch but loses the far edge, so the camera comparison continues with full coverage required.