Unlimited Detail Tech Demo

Total bullshit or the Messiah of future graphics?

http://www.youtube.com/watch?v=Q-ATtrImCx4

If it's true what they say it's kind of a breakthrough...

..... holy shit. o_o

how many dots a model have to have?
e.g. if you render a tree from distance its about 500 dots (rendered pixles) but if you zoom to a leaf fullscreen (1600x1400) only this leaf must have ~2.2Mpix!
(and then zoom to more detail )

The challenge of dot-cloud-models is not the select the visible dots but to handle the dots in memory ("unlimited amount of data..." [vpos 7:02])

(for me it's nonsens [the way it is showed])

ps: cool mozart music!
pps: nothing is impossible - so let's stay tuned

There also is another video on their homepage:
http://unlimiteddetailtechnology.com/home.html

Unfortunatey it doesn't tell THAT much more...

quite interesting and sounds actually possible.

interesting.... a good way to waste ram, but in a few years why not?

There's a lot of duplication -- even in their forest scenes they use the same tree over and over. I suspect this is because of the memory cost of each object.

I also suspect it doesn't handle transparency so well. That's really just a hunch based on some things that have been said.

Jibb

It must have downsides, but nevertheless a combination with polygons may be a perfect match. Certain things are not easy to make with polygons but are with this tech, and so is the other way around.

The ideal game editor conveniently gives a list of render methods to apply on your objects . If you want unlimited dots for grave detail, then you can't have transparency. But you can add transparent polygons to the parts that should be transparent, and leave a gap in the unlimited dots...

yeah, i dont think it was expressly said, but i'd assume that its similar to the same issues deffered rendering has.

Just... awesome.

How many times did he said "unlimited"?

unlimited amount of times

I was wondering, how would collision work with this? I think he'll have to go back to polygons for that...?

I think a bigger problem than collission will be animation. I guess thats why they havent shown a bit of animated stuff.

This has been done a long, long time ago. Check out this page. Don't forget to look at a demo and some screenshots at this page, too.

He talked like he was speaking to an idiot, it made the showcase alot dumber. Same goes to his site.

agreed, also collision I'd think would be a lot easier, but as puppeteer said, animation would take much too much ram right now.

I remember an interview with John Carmack about new id-Tech. And he told alot about terrains that dont need polygons anymore, some kind of voxel data maybe similar to this. But he also said that animated characters still will be made from polygons. And he told something about getting into problems with shadowing this environment.

From i see the guy seem to be hiding the scene from any single even tiny movement at all apart from moving along the camera projection,that's where i think where is the problem is,because when the movement(explosion,deformation,etc) happened it means there is no shortcut for his concept to run the movement of his instanced trilion of point to be simulated into new position?i don't have any phd in math besides my names but let see shall we ,from my opinion this could be good for brand new rendering engine for 3d modeler?.

it is insane

You're right, I haven't even thought about animation at all. Would probably be pretty hard as well.
Still, I wouldn't know how to do collision. You can't possibly calculate the collision between every one of this points and his neighbour...?

It really bugs me how he doesn't talk about any limitations at all in the video...