I'm looking for something similar to ROAM (Realtime Optimally-Adapting Meshes) in houdini.

Is it possible to perform subdivision on a mesh-grid based on distance to the camera using a Sop, or do I need to write my own code for that?

Thanks.

Houdini has it built-in.
Not a very pretty example, but it gets the point across. Just move the camera forward and backward.

Thanks for the quick reply and also for the example.

Do you know if it's possibly to do it the other way around? I would like a Sop like reduce which splits polygons based on distance to the camera. Otherwise I have to create a huge mesh and the reduce it which is not very efficient.

I'm slightly confused by your terminology, but aren't you basically talking about render time tessellation of a standard mesh based subdivision surface? Mantra does this without any special effort.

http://www.sidefx.com/docs/houdini9.5/rendering/subdivision [sidefx.com]

See this link for a how to. If I've not understood you, pardon my interjection into your thread.