Doing occlusion culling with machine learning methods

I have idea that is could be transitioned from ML world to game:
for example doing occlusion culling with classification, knn (k-nearest neighbours) or clustering, k-means

as you see
Model is parent
and child objects
for example model is the city and
children are buildings

and then the child objects could be classified or clustered
What do you think?
And also note the image above is not a represantation of screen space but this is a full scene

1 Like

Do these techniques take into account that in the case of a game, the objects are not points but have volume?

Good question I didn’t wonder if it works for 3D
but I think for classification, tags should be used as labels
also I found from

The biggest advantage to thinking of examples as vectors in a high
dimensional space is that it allows us to apply geometric concepts
to machine learning

and also imo clustering than classifying should be better when there is a lot of child objects,
so the clusters act like regions

Hi @grzesiekmq,

Occlusion culling is indeed very case specific on how to implement. In its most simplest form sector / vis areas can help on it.

You can search for forum posts by community member @Glenn_Ko, he did a lot of research on how to implement this in Playcanvas and has posted sample projects and code on this.

1 Like