When storing skeleton data that comes from running a script using body format 38 I notice that the keypoint position of some joints, the keypoint confidences and the global root orientation for some skeletons on occasion is null/NaN even though body fitting is enabled. This is not the case when using body format 34. How can I get body fitting to work with body format 38?
Hi @haakonflaar,
The confidence can indeed be null/nan on the keypoints not detected, that is the expected behavior. The confidence is not re-estimated by the fitting.
The global_root_orientation, on the other hand, should not be null at any point when the fitting is active. However, the fitting is active only when the tracking_state is OK (this will be added to the next release notes, it’s an oversight on our part 
).
What may be happening is that for a few frames, the tracking_state is not OK (but searching for example) and some data are being grabbed by your script, with some NaN. And since body34 depends entirely on the fitting, you have no detection for it, so there is no NaN.
That being said, it could also be a bug.
- Are you using the last version of the SDK (4.0.3)?
- Can you send an SVO where it happens to support@stereolabs.com, and ideally a script reproducing the issue? It will accelerate the investigation on our side.
Best regards,
Jean-Loup