We are experiencing severe issues with both tracking and jitter removal in ipi mocap studio pro.

In latest version (3.4.12.208) the algorithm gives unpredictable results (i.e. arms and legs start jumping around after 2 or 3 tracked frames) on both single kinect 2 and dual kinect 1 recordings.

Jitter removal also is affected. On long shots time to time the limbs (more often on the arms) jumps randomly in some frame to a non recorded and unpredictable position. Moreover removing the JR (setting the value to 0) doesn't revert to jittered motion (the wrong poses created by the JR algorithm are kept).


T this moment the only version we fond to work for us is the 3.2.3.190 where both tracking and jitter removal are working as expected whilst processing takes lot more time.

Our current setup is:
Windows 10 Pro x64
Intel Corer i5-4278U CPU @ 2.60GHz
8,00 GB RAM
Intel Iris Graphics5100

This new version has faster tracking, which utilizes video card to bigger extent.
We go similar report for owner of Intel integrated graphics card (older than yours).
So this may be related to video card drivers.
I'll send you beta version with several fixes in a separate message, that may help solving your issue.

Please be aware, that instead of rolling back to the previous software release, you can use old tracking algorithm in the latest version as well.
To do this, you need to check "Use old tracking algorithm" flag in "Advanced Tracking Options" expander at the bottom of Tracking tab.

_________________
iPi Soft

unfortunately even in the version you linked the jitter removal seems to be broken. It continues to give unpredictable results and corrupt existing correctly tracked data.
This is the reason i can't use the 'old tracking algorithm' option (that seems to have no effect on the new JR filter) and i have to downgrade the software.

the drivers are updated and are actually the same drivers i use with the older version (which is correctly working) so i guess it's probably something that has been changed in the JR algorithm.

...

Are you trying to re-track a previously tracked project that was tracked using the older algorithm?

OR, is this a brand new recording being tracked for the first time?

I don't believe you will get good results just trying to re-track any project that was already tracked with the old algorithm first.

Have you tried opening the original video again as a new project and tracking from scratch with the new algorithm?

If you saved the "corrupted tracking" as the same file name as the original without adding some tag to differentiate each, unfortunately you will have to track the video again from scratch with whatever version will work best for you, or possibly just use the latest version and select the old algorithm from the start.

...

After your first answer we tried a new tracking starting from an untouched video recording with the version you linked using the "old algorithm" option for the tracking. After the tracking was done we tried the JR and the results are always the same as reported in the previous post:

A) in some frames the bones jump to a wrong and untracked position (moreover you will always see this happen while the JR is ongoing)

B) the wrong poses created by JR seems to overwrite the good ones from the tracking pass. Appling JR=0 will not revert anymore to the previous tracking. Setting higher values (JR=2, JR=3) will add some more wrong poses and enanche the mismatch on the already mistracked ones from the previous JR pass.

C) opening a previously saved and tracked file done with the old version with the latest beta will give exactly the same results.

We really hope you dig up some more and hopefully find a solution before our license expires.

...

Possibly you can post a link to a Cloud upload of the video and calibration file, if needed, so iPi can test it on their ends to see exactly what may be happening, they usually request that in situations like you describe.

The JR process should not be changing any of the correctly tracked positioning of any body parts, so seems odd what you are stating, but obviously something is going on for you.

...

there's no calibration file, the shot is recorded with a single kinect 2 sensor.
I'll try to post the source video and the mocapfile pre-JR filter so that you can test it. Anyway keep in mind that this is only happening in latest versions of the software.

...

Ok, that would be more helpful, and yes, I only use the latest version of the software, although it is a bit different than earlier versions, I don't really have many issues with it, nothing I can't easily work through, I do only use 6 PS Eyes, but the tracking process is very similar, as I have used dual Kinect v1 and v2, as well and single sensor set ups, although I only tested a single Kinect v1 with the latest version, just to find out the speed increase on my machine, not really to use the animation.

...

Hello,
I checked you project. I used developer version to remove all data created with Jitter Removal, and leave only data from tracking.
I see that your project really contains issues introduced with Jitter Removal. But is also contains tracking errors (like right arm starting from frame 1775).
Jitter removal cannot remove any tracking errors. In your case tracking errors were caused with occlusion (part of arm occluded from sensor by body).
We recommend to face the sensor directly when doing motions, to minimize occlusion.

As to jitter removal issue, we can't reproduce it on Nvidia / AMD cards we have at hand. So I suspect this is specific to Intel integrated graphics (or maybe even particular card model). Intel's graphics drivers not always support all DirectX features correctly. And yes, jitter removal itself is not affected by "Use old tracking algorithm" flag.

I'll send you more details by e-mail, in order we could tracking down the issue.

_________________
iPi Soft