Please try the "New Proof of Concept," and let me know what you think. The source code is available for the "Old Proof of Concept."
The demo intends to showcase the "visibility precomputation" concept. I realize that the geometry in the scene looks dated--like N64, according to my girlfriend--and I might get around to something with a higher polygon count. I'm starting simple!
I'm not getting what's so special about this exactly, is it just some kind of new, more efficient/accurate way to cull unseen/obstructed geometry? All of the references in the video to Ready Player One and Stadia etc. don't really seem directly connected to geometry culling any more so than any game would be.
You have made the standard comment that tells me I need a better video explanation. The primary goal is to decide what geometry to send right now, which is a different question than, "What optimization techniques can I use to cull geometry to speed up drawing a frame?"
Plenty of games already delivery content over the internet--think Minecraft--but they're loading entire areas in bulk. I'm trying to sent only what is visible right now, just-in-time.
I'm going to develop an Android client for the same scene next. I hope that will make the use case more obvious.