New use of Cube Mapping with Rev?

[EndR]

Don´t know if this is old... but sounded interesting..

(taken from a swedish forum)

"As ripx points out, the technique has been used for a while to provide reflection effects. This is mentioned in the Nintendo patent. Nintendo's patent describes using Cube Mapping for an entirely different purpose - displaying a pre-rendered game environment from any camera angle in real time. The game environment data can be created with the most advanced rendering techniques that the developer can afford, or even from high-res photography of real world locations or detailed models.

This has two advantages. First, it provides a more attractive and realistic game environment than is possible with current real-time rendering techniques. Second, it does not use as much of the game system's hardware resources as the aforementioned real-time rendering techniques. This allows the developer to allocate more resources to the tasks of game logic, physics, A.I., and the rendering of highly detailed characters and objects to populate the game environment.

For most intents and purposes, this pushes the ceiling of how good the game environment can look all the way to infinity. The developer's limits are only the resolution of the display, the capacity of the game media, and the quality of his source material.

If a developer wants to spend a week to ray-trace an environment that will be played in-game for ten minutes, he can. It's like having a render farm of arbitrary power on the game disc.

Square-Enix has expressed more interest in Revolution than perhaps any other third party developer. They also happen to have more investment in high-end rendering technology than any other game publisher I can think of. Perhaps they believe that this technology will give them a competitive advantage.

The patent is very specific in that it applies to use of this technology in a home video-game console, attached to a normal display (color television is the example in the patent), with a wired or wireless controller, to play video games. It is not as weird or nebulous as many Nintendo patents I have looked at - it is downright straightforward. I recommend reading it."

http://www.gamecore.se/forum_read.php?id=261292


LINK
-----------

Anyone that maybe understands this better? My way of seeing this is that they can "manipulate" pre rendered enviroments and stuff, in realtime. So instead of having fixed camera angles in a pre rendered scene, you could have a "realtime" camera solution in that pre rendered scene.. or am I way off?

Once again, sorry if this is old, this sounded interesting and I was looking for maybe more explanation about it..

 

[Vaan]

Well, here I go with my thoughts...


A tradicional pre-rendered environment is nothing more than a plain hi-resolution render. A simple image. To this image you add a layer with the paths, and another layer with the Z information, I mean, a map with the point when a character must "hide" behind the render. (Or at least this is the way I always imagined)

What nintendo could bring to gaming is something similar, but much more powerful. Pre-rendered objects cubemapped.

Imagine an statue. Put it inside an imaginary cube. Now project the image of the statue on all the 6 sides of the cube. You have a cubemapped statue (Add a pair of buffers for depth info...)

We now want to represent this statue, so we calculate the spacial position of every point, based on the depth buffers and the color map and the interesection of the six sides of the cube.

So when rotating, when going backwards thus getting far and far, you only are apliying the transformations to the cube, and then the statue is rendered. And you see a photorealistic statue in the room.





Chances of being wrong => 98%

 

[Shifty Geezer]

Creating geometry from 6 projections won't work with concave shapes. Really why not just model the objects, bake textures on, and then you can view them from wherever?

Unless they mean taking 6 hires renders and projecting them onto a scene to illuminate objects? That again can't work were objects overlap the POVs of the cube mappings. Let's say you have two boxes standing close to each other. 5 out of 6 sides of each box would be captured ont eh cube map, but the sides facing each other won't have any detail captured. Plus the results are all static scenery. You couldn't move anything.

It seems to be mostly for backgrounds, which at what's considered virtually infinite distance anyway I don't see why you'd need depth information.

I've seen talk of this patent before. No-one could make sense of it then either :?

 

[Npl]

It seems to be mostly for backgrounds, which at what's considered virtually infinite distance anyway I don't see why you'd need depth information.

If you have a fixed POV and only rotate the Camera it could work like the classic Lucas-Arts adventures - having Layers of background, Objects and Characters will be in front or behind specific Layers.
Only instead of a few Layers there are effectively as much as the Z-Buffer allows, and the camera is rotateable.
But movement parallel to one of the Maps ( speak: any movement in 3D-Space ) is not reflected by the Cubemap - its rendered from a fixed Point.

Might work good enough for a very restricted POV around the Center of the Cubemap.

 

[Shifty Geezer]

If the scenery is defined by the depth map, surely edges of objects won't be clearly defined? Let's say you've got a view in the centre of a street. The edges of the houses won't be clearly defined in a lmited resolution depth map. How would that look? Also who wants to be stuck to one location?

With the option of baked on textures the only possible advantage is memory conservation, if the technology works, which I have trouble believing.

 

[Npl]

If the scenery is defined by the depth map, surely edges of objects won't be clearly defined? Let's say you've got a view in the centre of a street. The edges of the houses won't be clearly defined in a lmited resolution depth map. How would that look?

You could calculate an additional Alpha-Map and then shade occluding Pixels - Effectively some sort of poor FSAA
As long as the Resolution is high enough so you dont have to upscale.. it should look good

Also who wants to be stuck to one location?

Its Nintendo!!!

With the option of baked on textures the only possible advantage is memory conservation, if the technology works, which I have trouble believing.

Why shouldnt it?

Most of the Patents look like someone had a really bad cold and wrote down everything that passed his mind anyways. Doesnt means it has to be used.

 

[Ooh-videogames]

Maybe it involves a W buffer.