Calibration failed too many times and needs some advice

Hello new user here :smile:
Got my pop2 around 2 weeks ago but can do a testing journey around one week. It nice little product with a lot of potentials. As most users say in this forum the weakness is software and tracking and can improve in the future.

My current problem is calibration always failed in the end and accuracy tests are unstable. In this forum, many people got around 0.05 which is impossible for me :joy:

After many tries, the best result I got so far is 0.1 from too many fails. :smiling_face_with_tear: ( 5-6 passed out of 50-60)


My record video (I cut out the calculation phase a little to save time)

Things I discovered during this test were

  1. level the table and have enough light help a lot. Low light makes tracking unable to track big dots or stuck.

  2. Step 14 is likely to have “find corner fail” in the log file so need to twist your hand a little( dark corner cut off).

  3. Even if some image fails in log file software can pass the calculation phase.

I am not sure anymore what causes this problem with software or hardware because it also passes. the same in another machine and virtual machine

  • The software version is Revo_Calibration_for_win_v1.0.1.7,

  • Camera permission, firewall set. using a good usb3 port.

This is really strange because it shouldn’t let me pass 23 steps or show accuracy beyond 0.5 if there are many errors instead of failed. I hope this is a software issue.

My question

  1. Is there someone in the same boat with me or can give me some advice? :blush:

  2. I was wondering about the difference between good calibration as I know it decreases noise but how much? Maybe seeing other scans can help to identify. Is there a model of Agrippa scan with pop2 share somewhere or does anyone with good calibrate result wants to share? I want to compare it with mine
    this is my Agrippa from Pop2 default out of the box meshing in cloudcompare.
    Model

I checked your Agrippa and I don’t see you have any calibration issues , meshing at higher level will always create level of noises , objects too close with lots of red gain will also create little more noises , noises are part of this technology , and proper meshing can remove or add extra noises , most important here is the cloud point , and how it looks after scanning and not the mesh as mesh is processing product , focus how point cloud looks like , it is smooth , it have noises , if you see any imperfections it will be hard to fix it automatic . Your mesh can be as good as your point cloud is.

So next time you mesh , think about , do you need denoiser, do the mesh needs all the high pitch point settings when fusing point cloud , do you need higher values for meshing .

It is not the calibration issue here that is clear , it is about how you process it .
If you had calibrations issues it would looks like covered with carpet . Not the noises you have on your scan .
As with any scanner in the world , you need to process the files after scanning the proper way .

Regarding calibration software , it is harder now that it was before and we all go through the same as you did . Sometimes I pass , sometimes I don’t , and there is nothing beside the proper table level that you can do better .
I even used leveler on top of my POP2 , and guess what ? it didn’t matters , POP2 uses only the Depth Camera sensors and IR to calibrate , so you can do that in complete darkness and will still works .

Thank you @PopUpTheVolume :smile: that makes sense. At my accuracy level test, 0.1 should be fine. It is more about we see a cloud on the screen. Also how we process the mesh is as much important as the cloud we see.

So my issue is more about their software user experience, instead of “calibration failed” it would be better to show a higher value than the reference point or as it can calculate and write parameter greyed out. which can make users feel better about passing 23 steps got something :roll_eyes:

Did you made at least 3 tests calibrations before calibration ? To see the accurate number ?

If your point cloud looks nice and smooth and the model after meshing don’t, then you meshed it at too high value .

You can also export the point cloud to Revo Studio and smooth it out before meshing , it will result in much smoother mesh with nice details .

yep usually 2 times before each calibrate session. My factory accuracy is around 0.13
Passed calibration results I got around 0.1-0.13 from a total of 6. Accuracy tests swing quite a lot sometimes going up to 0.2 if just plugged I guess it needs time to warm up but normally in the range of calibration results.
I also did an experimental on some failed sessions to write data from the folder and was surprised the result was 0.08 :rofl:
I feel like something is wrong with my calculation phase or a different batch of pop2 but yea 0.1 is not too bad. It can be USB but the same for both USB version in a different machines.
I hope will have an improvement in the next calibration software update or can test with wifi.
for now, I should rely on the scanning process and learn a different method of scan :laughing:
Btw your scanning thread is really helped and a lot to learn from :blush:

1 Like

@lotte I made 4 tests my calibration changed from 0.0500 to 0.0900 , then I calibrate it and with many errors finally got 0.0680 , then I tested it after the calibration and the test shower 0.0900 , then I tested 5 times in a row and it finally shows 0.0680 , so I will take it with a grain of salt and not get too obsessed with it . The real problems accrues when it actually reach 0.5000 value .
I never saw anybody reaching bellow 0.0400 anyway , your results at 0.1000 are still fine , as long your see your cloud point nice and smooth without "frosting " you are all good . If your cloud point is frosty it can be also due to overlapped scans , too many rotations , but Cloud Compare can clean it out nicely .

So next time check the cloud point after capturing 360 degree and 300 frames for any issues on the surface , keeping an eye on the gain setting and have as low red gain indicator as possible ,

Secondo voi e’accettabile come risultato?

Sì lo è, non è affatto male

grazie molto seguo con interesse i tuoi lavori e sto imparando tanto

1 Like

Grazie Giorgio, felice di sentirlo! :+1:

Can anyone reply here with the proper dimensions for the stick used to prop up the calibration board? Mine was not included with the premium package (POP2).

I asked Revopoint for the dimensions and they sent me a diagram that is obviously wrong (the width is way too wide to fit in the slot on the board)

Also, does it matter what side of the calibration board the stick would mount to? The socket on the back of my board is on the opposite end of the one in the video.

My calibration score is 1.15 so I would be thrilled to see results like you are getting, but since I saw nothing actually telling me what the numbers mean, I didn’t know 1.15 was bad until I read this post. So far my scans are junk and the POP2 has been a huge, frustrating, expensive (for me) disappointment.

Just measured mine with caliper:
30mm x 120.4mm x 4.85mm

Thank you!!! The diagram they sent me was 35mm wide so I did not trust the length.

1 Like

If you calibrate without the proper angle then you get 1,15 or was that 0.115 ? that are huge differences
The calibration video has a wrong preview , it should be reversed.
The original white stick was
30mm x 120.3mm x 4.8mm
my opening for the stick is
30mm x 4.8mm x 8mm deep
it was hard to get the stick in the first time
now it is more of a 28.85mm wide on top

you have the board so you can measure the exactly opening of yours.
Most important is the length as it will give you the proper angle .

Haha yes that was really 1.15 because I did not have the proper angle. I will print a new stick when I get home and start again.

Thanks for letting me know about the board reversal, I was worried I had a defective board :slight_smile:

1 Like

to hard calibration !
22/23 to hard for accurate the dot !

I know , I go through it many times , before it was the 17/23, now the 22/23

1 Like

I have followed the same problem as the person above but the accuracy is always at 1.1/1.0 never below that, pinpointing the point is very difficult, especially calibration.
I’ve tried a lot but nothing works, where did the error come from ?

why is my calibration point by point on the board very difficult to meet in stages 3 and 4, along with 22/23 is very difficult and always fails, I have tried a lot and it is tiring, is it possible the software has a bug?

I managed to get 0.0500 but after couple of times trying . The best result was when I level the table , nothing else there you can do about , but you know ? 0.9000 or 0.0500 there is not a difference in scanning , I think you will need at least 0.5000 score to see bad scan as that is when you actually should calibrate .