DICe beta 8 malfunctionning

[Vincent730]

Hello DICe community,

I'm using DICe V3.0-beta.8 in stereo correlation.
The calibration I made was nice. I have a camera angle of ~30°. You can find the cal.xml here:

Then, I tried to analyse the results. I'm using .tif images of 8MPx each (here I provide you .png ones so that they can be displayed)
Case 1:
I successfully analysed some tensile test experiments. All went good with images like this:

Case 2:
I performed other tests later, but now, only a few part of the images are analysed. The images are:

Nothing has changed so far, I tried other parameter analysis, to play with the initialization methods, change the subset_size from 21 to 51, use some shape functions or not or all of them. Nothing changed.
I also tried to use the option "USE_RECTIFIED_CORRESPONDENCES" but it barely improved the results.

I really don't know what to do next, do you have some advises ? Some hints about what's going on ?

Thank you fir the reply,

[dicengine]

I would try lowering the SSSIG tolerance. It may be that your new pattern doesn't have enough variation in the contrast.

[Vincent730]

Thank you for your answer. I already tried with a SSSIG of 0 but nothing went out...
If you have another idea, this would be great !
Have a nice day

[dicengine]

Can you post a screenshot of DICe that shows the areas of the image that get analyzed? That might help narrow down what is going on.

[Vincent730]

Hi, here are two screen shots of what DICe is doing.
Case 1:

Case 2:

I hope it will help
Thank you

[dicengine]

Were there any errors reported in the console output at the bottom of the GUI window? Also, did you output the exodus file, DICe_solution.e? If so, you can look at the exodus file to see what the STATUS_FLAG field was for each of the nodes that failed. That would tell you if it's related to the initialization, the solver hitting the max iterations, etc.

[Vincent730]

Here you can find the STATUS FLAG representation in Paraview

Only few points were found, all the other failed the initialization...
What can I do to solve it ?

[dicengine]

Was the camera system moved between case 1 and case 2? It seems that the calibration is slightly off for case 2. I ran your images through DICe with debugging on and Case 2 doesn't successfully cross correlate between the left and right camera. The epipolar error is too large so it fails all the points you see above in red. Just curious if the camera could have been moved.

[Vincent730]

Thank you for your work.
The cameras may have moved indeed, but if it were the case, it would have only been a very small displacement because the focus was still very nice. Is there a mean to run the analysis even with this high epipolar error ?
Just for you to know, I ran all the tests with VIC-3D and no issue occured...

[dicengine]

I'll have to look more closely at why the initialization isn't working. One workaround would be to increase the epipolar error tolerance, but there isn't an option in the code for the user to set that. When I was looking at the epipolar errors, they were just marginally over the tolerance so it could be that moving the cameras barely exceeded that threshold. We should probably have a way for the user to just be warned about the errors and let the analysis proceed.

[Vincent730]

It would be nice yes. Or to be able to choose the tolerance so that the analysis can proceed but only for your threshold. In the "params.xml" for instance.
Thanks for the time you spend on this case. I can't wait to see this option in the next months !