I want to get the orientation (in 3d space if possible) of a marker object in a video stream using tensorflow lite on a coral tpu. My marker looks like:

The marker will be sticked on various objects to get their orientation without the need to train models on every separate object.

I can find many examples on how to train for object detection, but fail to find anything on orientation detection.
I suspect it should be something like pose estimation? But also there I can’t find anything besides using pretrained models.

As language I want to use either python or c++.