Add panoptic driving perception demo using YOLOPv2

[DiegoFernandezC]

This PR introduces a new demo using the YOLOPv2 model.

[javiber]

Looks good, the video seems to perform worst than theirs but is actually much harder so it's expected in a sense.

skimming through the code I got the feeling that we included some unnecessary dependencies and utils, even some functionality that logs performance. I was expecting a minimal demo like the others but maybe we want to have a different approach in this one

[dekked]

Need to add this demo to the main README doc.

[javiber]

Can we add some comments on our code that points to the original code in the yolopv2 repo?

The motivation is to separate the code we used from their repo from the one e had to add to make the demo work. This will help us in the future to have the lineage of the code, and know what we did vs what is yolopv2 code.
Of course, this request is only valid if the distinction is clear, if you had to copy a function from them and then modify it then just add a general comment explaining that the code is based on their repo.

[javiber]

remove unnecessary return and commented code in this function and the above one
(I'm not against commented code that might be useful for debugging or future improvements but in that case, add a comment explaining why is there)

[dekked]

This comment is not helpful.

Why are we installing this version of PyTorch?

[DiegoFernandezC]

We had problems with the versions of Torch used. I didn't see that the actions failed. It was solved using the ones from the docker image.