ST02A High Amount of NaNs


Using the latest SDK v0.11.1, we compared the Mark 2 sensor’s (ST02A) depth frames with ST01 depth frames. Model ST02A has a much greater amount of NaNs than model ST01, see table.

Can you please explain why ST02A has so many more NaNs than ST01?
With ST02A, how can we obtain the same results we have with ST01?

Please note:
-we have been reading the Occipital forum twice daily for any updates on how to fix this issue in ST02A.
-custom settings mentioned below are those mentioned in various pages of the forum.

Context of our test:
Scans of a non-reflective wall were taken using the Scanner app shipped with SDK v0.11.1, at a distance of 55 cm. Illustrated below is one image representation of the depth frames for ST01 and ST02A. We can clearly see that ST02A outputs a much higher number of NaNs than ST01.

BLUE -> Depths
RED -> NaNs

Occipital’s assistance in this matter would be greatly appreciated.

Hi Paul, I think I may have been experiencing the same issue. Out of curiosity, before working with v0.11.1, did you observe the same lack of depth info with earlier versions?

Also I think your sensor might have IR offset issues that are described by this comment. If this is the case, you should get in touch with the developer team to get access to the expert calibrator. I was able to improve the depth maps I receive from the Mark II that look more consistent with the image you posted from ST01. The maps still do contain some holes but the scanning experience is much better now after using the calibrator.

I hope this helps.

Hi @ecajuste, thanks for the tip. I have tried what you suggest with no improvement on depth coverage. NaNs % remain as high as ever in ST02A, while ST01 gives us much better depth coverage for the above test, around 90%. For other use cases of ours, ST01 is able to give us between 99 and 100% depth coverage. It’s a wonder why Occipital decided to discontinue a working device, and replace it with an unstable and unreliable device.

We are experiencing the same issue: High amount of NaN values from the sensor.

We are using the sensor to measure wear on some industrial parts and ST01 work quite well for our use. We where quite surprised when we no longer could order the ST01 but when we bought and tried the ST02A, we were even more surprised by the poor depth values we got out of it. Sure, we had stability issues as well but the poor framerate and high number of NaNs makes the device unusable for us.

We have had some dialog with the support but they have so far dodged this (for us) critical issue. The silence from Occipital about this issue makes us believe that this may very well be an unfixable hardware defect. But I really hope we are wrong.

Hey Paul, no problem, but sorry I couldn’t help. I thought recalibrating would make some sort of improvement for you. I could still use some tips myself to get even smoother gradients for the Mark 2 depth maps since I still encounter enough holes that it affects the user experience. If you find any new info or tips about this that would helpful, let me know.

I also wonder about the abrupt switch to the new sensor sometimes but as more SDK releases come out, the more confidence I have in the product.