Inconsistent coordinate space on output 3d bounding boxes

[free-soellingeraj]

I am finding that different models producing 3d bounding boxes output boxes in different coordinate spaces. The feature here would be conforming all of those models to the same coordinate space, or providing tooling and documentation to discover the coordinate space in which the model outputs results.

I experimented around with different coordinate space transforms, and it's not straight forward to discover in which coordinate space the outputs live. The test here is whether the boxes are aligned with the input image. In my case, the input is in Camera Oriented space, data taken from a 3 LiDAR innovusion system.