Extend Panoptic Quality usability, functionality and testing

[marcocaccin]

🚀 Feature

Expand the PanopticQuality code to

• allow batches in input (necessary feature)
• allow it to work on both images and point clouds (other inputs TBD)
• do not consider never seen classes in metric computation
• extend testing
• [let's see if something else comes up in the discussion here]

Motivation

The just landed PanopticQuality is a welcome addition to the toolbox, but there are a number of minor issues associated with it, from deal-breaking (does not take batches as inputs) to unnecessarily restrictive (PQ is a metric for panoptic segmentation both in point cloud and image domains), to suboptimal in terms of interpretability (if a class has TP + FP + FN = 0, why should it contribute to the final metric value?)

Pitch

Let's bring in these goodies, happy to contribute 😉

Alternatives

Additional context

[github-actions]

Hi! thanks for your contribution!, great first issue!

[bhack]

I will add:

• cover the Panoptic boundaries [Metrics] Panoptic Quality #50 (comment)