Our UR5e camera proposal has a scratch dataset where the bright line moves across the same stainless plate between images. The physical samples are still in a drawer, thankfully. I'd like to sort the labels and presentation before we buy more lighting kit.
Retrieve the samples and have the acceptance owner classify them using the agreed physical inspection method. Preserve uncertain cases. Your image labels currently mix a visual observation with an unverified defect judgement.
Keep the unmatched images out of definite training references. You can still use them to investigate imaging problems, provided their unknown physical status stays visible.
Photograph the sample identifier beside the part for the record, then use the same identifier on the inspection images. We lost ours when somebody renamed folders for convenience.
Keep identification clear of the inspected surface. And record which side and area each image represents; a part number alone may not identify the feature being judged.
Can you use the existing light to compare identified good and bad samples first, before ordering the replacement everyone has already decided will fix it?
Yes, a controlled comparison can establish what the present arrangement fails to reveal. It will not guarantee that a different light solves it, but it makes the supplier discussion much less speculative.
We have an agreed sample set now. Existing-light comparison is next; purchase request paused. Unmatched images are retained as investigation material, not quietly relabelled as good parts.