Lightburn Camera 8MP 120-W cannot complete calibration

I just got the camera and tried to calibrate since there is no preset for this one, but it seems to be impossible. Step 1 and 2 are very easy, but step 3 took roughly 120 tries to get a score below 1.
now in step 4 it is the same.
Pattern is glued to the surface, but it does not seem to help.


It does a lot of weird distortion with stupid scores like you see above even tho it always said “Pattern found”.
Once in a while there is a good one in there, but not enough to complete to the next step:

What’s with that super distorted capture image? Weird.

Try doing the calibration away from the laser. It doesn’t need to be in the final position for calibration. Only for alignment. Preferably somewhere with a nice clean background with good lighting like a light colored wall where you can control the environment. Try to get the pattern to fill a 1/9th of the visible frame for each capture.

Got them cam out of the mound and tried on the floor. Still no luck. Can’t even complete step 2 anymore. 99% of images are those distorted ones. I will give it another try tomorrow got way to late here ^^"

Something’s not right with those wacky images. When you get some rest try opening the camera app and see if it even behaves correctly there.

Yeah camera looks fine in any other app. Focus is fine, no delay and also the lightburn preview is fine.
My guess is, that something in the background confused the algorithm into detecting something else as the pattern, leading to very high distortion.
So after rearrangement:


It stays at “Pattern NOT found”.
At least the error changed…
Moving the camera back a little gives distorted images again even tho the score is quite good this time up to step 3 where it gets ridiculous again.

Maybe this is a bug in the fisheye calibration?
I’m calibrating using the fisheye option, since the frame in the preview is clearly curved and staff told to do so:

However if I use the standard option, calibration does work till the end without issues … except the result is mediocre, since there is no correction for the curves.

Okay. So just a matter of getting a clean capture. I love how in that last screenshot you got a 0.17 score.

I’d say try to find a completely blank background. Gonna be tough with that super wide lens. Looks like you should be able to bring the image closer to the pattern, similar to the first screenshot in your last post.

I think the surrounding elements are still confusing the capture.

This has to be a bug.
I cannot get the background more even.
And still score is not going below 1 and sometimes it still completely messes up.

Clueless now on what to do.

Can’t see the attached video…

does this work for you?

Nope. I get the same static video player with no ability to play.


Apparently your browser does not allow for webm, so there is a gif :slight_smile:

I can view webm fine so not sure what’s going on there.

That gif is a wild ride. Love the purple background. Perfect. The only thing at this point that I can think of is that the corner tape-downs are interfering but can’t imagine that would be the case. Some of the captures are insane. I wonder why so much variation.

Maybe someone else in the forum with the same camera can send you their calibration.

I’d suggest sending an email to support@lightburnsoftware.com asking for help on this as they may not see this in the forum. Include a link to this topic so that they can review the history.

The purple one was the only blanket at hand and probably better than a white one contrast-wise.
Thanks for your help I’ll go and ask the support, maybe they even have a calibration file for this camera, since it is an official one :man_shrugging:

Is there any potential for turning this into a manual process?

I covered the honeycomb, initially got good scores, but they soon turned awful and just could not get 1.0 for a hundred takes. When I finally got something, the result was inaccurate to several mm. That is, I fired a couple of crosses near the center of the bed, tweaked XY offset and size, and did another cross a foot away and got like 2mm of horizontal error. And that’s not just in the X instead of Y, but with the door interlock disabled so there was no door repositioning error.

It took so long to attempt the lens autocalibration that I think a manual process wouldn’t be a larger burden, if it’s accurate. I’d like to throw down some pieces of paper and burn a dozen or so reference points and manually adjust the lens calibration parameters- how many parameters are there and how interrelated are they?

I think an option would be to allow for an on-screen calibration or override. So instead of relying entirely on the auto-detection it could load an initial take, then provide controls to allow you to tweak the capture according to some rules or set goals (e.g. get the dots aligned). The thing is when the autodetection works it’s likely to be better and faster than manual tweaking although this is a source of frustration for many.

By the way, you can see the actual calibration figures if you export the calibration. It’s quite small.

Did you glue the pattern card to something extremely flat, like MDF or foam board? Or did you just tape the corners of the paper? If you did that, the pattern card would have curvature or ripples in it that are not standard lens distortion, and the code would have no way of correcting for that.

Ignore the distortion in the image for the first couple of captures - It’s showing you an “undistorted” image, but it’s only based on a single sample of data, so the results can be extremely misleading. This actually causes confusion for many people, so I’m likely going to make it not show anything at all for the first two captures - until you have three or more the data is basically useless.

You are also not moving the pattern at all when capturing. You’re just clicking the capture button repeatedly, which doesn’t do much. Try moving the card closer or further from the camera, and angling toward the camera. The code can be very sensitive, and there’s unfortunately not much I can do about that. I would calibrate one here and just upload the result for you but we’re out of stock on 120’s and I don’t have a spare.

Does this mean that data accumulates over multiple clicks? I had thought these were all isolated attempts. That could explain some of the things I’ve seen.

The data accumulates over the 9 images that it asks you to capture. If you’re on image #1 and keep re-capturing because the undistorted image looks funny, you’re sabotaging your own success.

Once you get an image with a good score, even if it looks bizarre, you should progress to the next capture.

For example, this is a great score:

… but the image on the right looks bizarre. That’s ok. This was the 2nd capture, also a great score, still a weird image:

But now, with the 3rd capture, there’s enough data, and the undistorted image looks correct:

I lied - I found a 120 we had as an exchange, and was able to calibrate it (see the above shots). The resulting calibration file is attached.

8mp-120.lbcm.txt (394 Bytes)
You will need to remove the .txt extension (had to add that to get it to upload here)