Ever seen those cool 3D cloud points in movies? Yeah, those are point clouds. Compressing them is a big deal, and MPEG has a special way to do it called V-PCC. This method can squeeze point clouds into tiny sizes, but sometimes, the original ones go missing. That's where no-reference or reduced-reference quality metrics come in. These metrics are like judges who score how good the compressed point clouds look without needing the originals.

The big challenge is figuring out which features to use to judge these clouds. We're talking about things like how much they're compressed, their shape, how pointy they are, and how bright they are. We've come up with a set of these features. But just having them isn't enough. We need to pick the best ones, and that's where a method called LASSO comes in. It's like a smart judge who picks the most important features.

Once we have the best features, we need to turn them into a score. This score is like a beauty pageant score for the point clouds. We do this in a nonlinear space, which means it's not a straight line. Our method, PCQAML, uses 19 features right now, but it's flexible enough to use any number.

We tested PCQAML on two big datasets, WPC2. 0 and M-PCCD. Guess what? It beat all the other methods in terms of how well it agreed with human opinions and how close it was to the real scores. That's pretty cool, right?

A Smart Way to Rate Compressed Point Clouds

Actions