Why is CLS_ORIENTATION False for barriers in dataset?

[harsanyidani]

I see that in the code: https://github.com/tjiiv-cprg/EPro-PnP-v2/blob/85215de8002dbd8523ee8eaaf1bae85b47179ebe/EPro-PnP-Det_v2/epropnp_det/datasets/nuscenes3d_dataset.py#L41 CLS_ORIENTATION is set to false for barriers. Why is it this way?

nuScenes states about the TP metrics in the detection task:

We omit measurements for classes where they are not well defined: AVE for cones and barriers since they are stationary; AOE of cones since they do not have a well defined orientation; and AAE for cones and barriers since there are no attributes defined on these classes.

So orientation should be computed for barriers too.

Or did I misunderstood the meaning of CLS_ORIENTATION?

[harsanyidani]

Now I see that AOE is calculated for barriers too during evaluation. Am I seeing it right that CLS_ORIENTATION is only used for visualization purposes here:

https://github.com/tjiiv-cprg/EPro-PnP-v2/blob/main/EPro-PnP-Det_v2/epropnp_det/models/detectors/epropnp_det.py#L113-L119

And nowhere else? Can you please clarify what it's purpose is?

[Lakonik]

Now I see that AOE is calculated for barriers too during evaluation. Am I seeing it right that CLS_ORIENTATION is only used for visualization purposes here:

https://github.com/tjiiv-cprg/EPro-PnP-v2/blob/main/EPro-PnP-Det_v2/epropnp_det/models/detectors/epropnp_det.py#L113-L119

And nowhere else? Can you please clarify what it's purpose is?

That's right, CLS_ORIENTATION is only used for visualization. Orientations of all objects are handled in a probabilistic manner during training and inference so it doesn't matter whether an object is symmetric or not.

[harsanyidani]

Thanks!