é sério mesmo pessoal
a Nvidia fez 3 "concursos" pra ver quem fazia a melhor demo usando o CG......
um Brasileiro ficou em primeiro lugar no terceiro concurso e em segundo lugar no segundo concurso........faturou uma GeForceFX zero bala!
.......
.......
Legal mesmo um Brasileiro ter feito isso.....
é bom ver que tem gente que faz MESMO aqui no Brasil.......não como certas pessoas que eu conheço que só ficam nas teorias....
O nome dele é Fabio Policarpo
abaixo uma amostra do trabalho dele:
==============================================
This Cg shader demo implements a nice image based rendering technique. The idea is to render a very complex object from only 6 texture maps.
For a given object, its bound box is calculated. Then 6 images are generated, one for each bound box face. We store the object normal into the RGB component of the images. In the alpha component of the images we store the depth information, depth 0.0 means on the beginning of the bound box and depth 1.0 at the end of it.
At render time, we draw the bound box texture mapped with the RGBA images calculated before (one image per bound box face). The fragment program transforms the pixels in the images based on the current view position and pixel depth giving the illusion of a real mesh. You can rotate and view the object from any position and it will always look as the real large resolution mesh.
Some restrictions are applied to the type of objects we can use. They should be convex for a 100% perfect image based solution but in practice we only need that at each of the 6 views, no part of the object occludes some other part of it. Anyway, you can always break down a concave object into a set of convex ones.
Several other options are possible and implemented in this demo like per-pixel lightning, texture mapping and environment mapping. A LOD system allows faster drawing of the objects by specifying the number of levels to consider. Also a correction term has been added that can smooth the solution for low LOD draws.
There is NO geometry in this demo at all (only the object bound box is drawn (6 quads, 3 visible at each frame at most). Using the NV30 emulator will make things VERY slow! You must select the options you want (object, program, etc.), move the camera to the desired position and then activate the fragment program (key 'K'). Then wait until you see a result... for large images can take minutes on the emulator. Try for small images (reduce window or move camera away) for a faster update.
The first object (relief) is made of a single quad but as you move sideways you will see the statues moving in front of each other given the illusion of the actual geometry. The secont object (satyr) is made of 6 images and allows viewing from any direction. The best quality can be achieved using the highest LOD (256 levels) and the correction option on.
============================================================================================
Show de Bola Fabio....
Parabéns mesmo
a Nvidia fez 3 "concursos" pra ver quem fazia a melhor demo usando o CG......
um Brasileiro ficou em primeiro lugar no terceiro concurso e em segundo lugar no segundo concurso........faturou uma GeForceFX zero bala!
.......
.......
Legal mesmo um Brasileiro ter feito isso.....
é bom ver que tem gente que faz MESMO aqui no Brasil.......não como certas pessoas que eu conheço que só ficam nas teorias....
O nome dele é Fabio Policarpo
abaixo uma amostra do trabalho dele:
==============================================
This Cg shader demo implements a nice image based rendering technique. The idea is to render a very complex object from only 6 texture maps.
For a given object, its bound box is calculated. Then 6 images are generated, one for each bound box face. We store the object normal into the RGB component of the images. In the alpha component of the images we store the depth information, depth 0.0 means on the beginning of the bound box and depth 1.0 at the end of it.
At render time, we draw the bound box texture mapped with the RGBA images calculated before (one image per bound box face). The fragment program transforms the pixels in the images based on the current view position and pixel depth giving the illusion of a real mesh. You can rotate and view the object from any position and it will always look as the real large resolution mesh.
Some restrictions are applied to the type of objects we can use. They should be convex for a 100% perfect image based solution but in practice we only need that at each of the 6 views, no part of the object occludes some other part of it. Anyway, you can always break down a concave object into a set of convex ones.
Several other options are possible and implemented in this demo like per-pixel lightning, texture mapping and environment mapping. A LOD system allows faster drawing of the objects by specifying the number of levels to consider. Also a correction term has been added that can smooth the solution for low LOD draws.
There is NO geometry in this demo at all (only the object bound box is drawn (6 quads, 3 visible at each frame at most). Using the NV30 emulator will make things VERY slow! You must select the options you want (object, program, etc.), move the camera to the desired position and then activate the fragment program (key 'K'). Then wait until you see a result... for large images can take minutes on the emulator. Try for small images (reduce window or move camera away) for a faster update.
The first object (relief) is made of a single quad but as you move sideways you will see the statues moving in front of each other given the illusion of the actual geometry. The secont object (satyr) is made of 6 images and allows viewing from any direction. The best quality can be achieved using the highest LOD (256 levels) and the correction option on.
============================================================================================
Show de Bola Fabio....
Parabéns mesmo