Marker augmentation returns random point cloud

[adam-kozlowski-280]

I'm using your solution for almost 2 years now and recently I've noticed that marker augmentation step returns strange point clouds for specific sets of markers. Example is show on this image:

For reference: those points are in tens of meters apart from each other, input markers are in the middle of the cloud as they remain unchanged. This is a simple t-pose recording.

As you can see points explode into some unrealistic point cloud that doesn't resemble in any meaningful way original markers.
Input data order and format are correct as I've done hundreds of those calculations so far and only recently I've encountered them.

I have 0.53 pixel error during extrinsic calibration (I have separate pipeline for this) and I have recomputed it multiple times: everything always fails during augmentation step.

The most annoying part is that I have examples of people that either have the same issue (most of them from this year) and even larger group were it works

Have you ever encountered such behaviour previously? What might be the issue?

[mattpetrucci]

Hello, this is a question better suited for the forum, so please make a post there.

What devices are you using? You also say that the pose marker stay in the middle of the cloud, do any of them have discontinuous jumps across the capture volume even for a few frames? This could throw off the augmenter.