The scratch labels look less certain than the images suggest (Fairino FR5)

RachelChan1131 · 14 May 2026, 17:52 UTC

Reply to discussion
RA
RachelChan1131
We're planning presenting reflective components for surface inspection, with Fairino FR5 presenting a brushed stainless trim plate to a fixed camera in a small finished-parts production cell. Bright streaks move with small presentation changes in our example images, but people keep calling them scratches. We haven't commissioned the cell. How do I sort the references and imaging before we tune thresholds around this mess?

15 replies

RE
ReeceArcher0384
Replying to RachelChan1131

@RachelChan1131 Were those scratches confirmed on the physical samples using your agreed appearance method? Start by linking each image to a sample and its inspection result.

16 points
RA
RachelChan1131
Replying to ReeceArcher0384

Some labels came from the pictures alone. We still have the samples, but those labels aren't tied to physical inspection results. That's irritatingly circular.

7 points
RE
ReeceArcher0384
Replying to RachelChan1131

Keep the image-only labels uncertain pending physical inspection. A controlled comparison of lighting and presentation can then investigate whether the moving streaks indicate reflection effects.

18 points
HA
HazelBrown0919
Replying to ReeceArcher0384

A scratch's visibility can change with presentation too. Don't let 'moving streak' quietly become the rule for relabelling something acceptable.

12 points
RE
ReeceArcher0384
Replying to HazelBrown0919

@HazelBrown0919 Correct. I meant a reason to investigate the imaging, not a new defect label. Physical acceptance stays with the agreed inspection method.

18 points
RE
ReeceAllen0297
Replying to ReeceArcher0384

@ReeceArcher0384 So uncertain isn't the same as good? Where do those images go?

2 points
RE
ReeceArcher0384
Replying to ReeceAllen0297

Keep them explicitly undecided, with the sample link. Don't use them as confirmed examples of either class until the physical review settles them.

24 points
RA
RachelChan1131
Replying to ReeceArcher0384

Our two comparison folders changed exposure and presentation together. I can't tell which change made the images look better. That comparison needs a caveat too.

-2 points
TH
TheoArcher0410
Replying to RachelChan1131

On my inspection setup, a folder called 'better lighting' also contained easier samples. The folder name was doing a lot of unearned work.

20 points
RA
RachelChan1131
Replying to TheoArcher0410

@TheoArcher0410 What did you keep beside each image? I can manage sample ID and settings, but I don't want a form nobody fills in.

8 points
TH
TheoArcher0410
Replying to RachelChan1131

@RachelChan1131 On my setup we kept sample ID, exposure, lighting arrangement and presentation condition. Enough to explain a comparison without reconstructing the whole session from memory.

20 points
RE
ReeceArcher0384
Replying to TheoArcher0410

Use those details to ask your vision specialist for controlled comparisons on representative surfaces. Keep the sample set fixed when assessing an acquisition change.

6 points
RE
ReeceAllen0297
Replying to ReeceArcher0384

@ReeceArcher0384 Could using a learned inspection model remove the need to resolve the lighting variation?

3 points
RE
ReeceArcher0384
Replying to ReeceAllen0297

It still needs trustworthy labels and representative evaluation. A different model doesn't establish whether your reference scratches are real or your acquisition comparison is fair.

9 points
HA
HazelBrown0919
Replying to ReeceArcher0384

@ReeceArcher0384 Keep physical sample identity in mind when forming evaluation sets. Different images of a sample used for tuning don't provide an independent sample-level evaluation.

22 points

Add to the discussion

Welcome to Application Robot

Everyone can read the forum. Sign in or create an account to start a discussion, reply, or upload photos.

Forgot your password?

By creating an account, you agree to our Terms and Conditions and community guidelines. Read our Privacy Policy for how your information is handled.