Unlimited detail

Not really directly panda related, but still interesting for us 3D computer graphics people. I saw an article about it in NewScientist magazine.


They purport to be able to render better than movie-like polygon counts, in real time, in software, even on a limited platform like a phone. Oh, and unlimited high-poly terrain without LODing.

It sounds impossible right? The gist of the technique, as I understand it, is to convert the high-poly model into a kind of point cloud, compress said cloud very efficiently, load the whole thing into RAM, and then use their secret algorithm to very quickly search for only the one color point needed for each pixel on the screen. I’m not sure how they can manage lighting without normals.

Example videos are on youtube (search for “unlimited detail”). In the examples, there does seem to be some aliasing problems (especially the shadows) and all the examples are fairly bad programmer art, but still, they say these are temporary problems.

I wonder what the future for panda is, once this gets released?

Eh, I’ll believe it when I see it happen on my own phone.


i saw that engine earlier on. technically it is possible using an octree-like structured search for points which stops the search process on the first point close and small enough to the camera. while trying to get behind their working algorithms i estimated that this engine suffers a few very serious problems.
1st is: the ammount of data, while they claim infinite detail you may have noticed that there are many repeting elements. if you’d make everything unique you’d end up with an insanly ammount of data whichi you’d had to stuff in your RAM while searching through it.

other problems are dynamic lighting/shadows, each light will take as much time to compute as your original image.

next problem i fear will arise is animation. searching huge ammount of well structured data is no real problem these days. but as soon as you have animated point clouds searching the correct point turns into a computing chore.

so while it looks all nice an beautiful i belive it has some major drawbacks which makes it very unsuitable for beeing anything else than a point-cloud visualisation engine. if you dont belive it, see the video’s. they point out the “flaws” in todays engines… but they dont really mention how their’s perform in tasks you’d need every day. and if someone doesnt mention a functionality in an advertising video there usualy is a serious problem with it.

therefore… i dont belive this engine will affect panda in any way. i mean not even unreal3 did afaik. except we got inspirated to add a few new features :wink:

Like we’ve never heard of vector graphics! Though the by-pixel rendering instead of by-vertex has a certain appeal.