Yeah... Good luck rendering an infinite number of infinitely-small polygons using rasterization techniques... Also, infinitely-small polygons are simply points, from any practical point of view, not really polygons.

@Felix: What was the point of your post? Also, if you're quoting me, at least don't omit key-words. Small polygons are simply small polygons.

To be awesome.

The only difference I could conceive is that points would be considered as single discrete objects, whereas a very small polygon would still have several vertices.

It makes sense for some things (you can make some really neat liquid effects with particle systems and real-time surface generation), but to use it to represent generic 3D volumes? Hell no. As someone else pointed out, if you want actual infinite detail, just use NURBS.

The video makes extraordinary claims, has little explanation of the technique, has lots of irrelevant detail and pseudo-tech talk, and is even deceptive. Did he just accidentally slip over five minutes in a long-prepared demo? Forget the tech side, the cues alone are enough to call this grade-A bullshit.

The way the point cloud concept works is like this.

Imagine a sphere which has a volume. You also have a cube that encloses the sphere with sides equal to the diameter of the sphere. Now you use the random function in C++ and find the modulus of the radius of that sphere from the origin like so:

randomX = rand()%radius;
randomY = rand()%radius;
randomZ = rand()%radius;

...if all three of these variables randomX, randomY, randomZ fall within the volume of the sphere then you plot the point in openGL (or whatever API you like). After a certain amount of time you finally get a point cloud and a probability distribution.

The point is that this it the point cloud they are talking about and this is the way you would find the volume integral of a sphere. The concept of using this could be extended to different colored points as well as using simple primitives such as cylinder, cubes, etc... to generate more complex shapes.

I think this is the easiest example I could explain here but Im sure most of you already know this. This is where Monte Carlo integration is also used for calculating integrals of any dimension.

Heres a program in Java that explains it (on the bottom).

I could sort of see a hierarchy being used to determine the output of data where the hierarchy could represent and object and all its sub objects but I really dont think that is something new and it really just reeks of bullshit.

{"name":"800px-Matryoshka_Russian_politicians.jpg","src":"http:\/\/static.allegro.cc\/image\/cache\/1\/c\/1ce8eaf61ef74ebdb181ce22b0832de8.jpg","w":800,"h":369,"tn":"http:\/\/static.allegro.cc\/image\/cache\/1\/c\/1ce8eaf61ef74ebdb181ce22b0832de8"}

A Matroshka doll. Invented in Russia a millenia ago.

But some questions remain:

a) how do they manage to rotate all the points while maintaining a good frame rate?
b) how do they manage to render partially clipped objects?

I don't think the 'point cloud search' algorithm has to do with a Matroshka doll type search. I think that's it's more about searching which voxels fall within a particular X/Y/Z.

It does impress me less than a MMORPG with giant maps (ala WOW)