UV rotation through adjacent pixels?

[Victor]

First of all is excuse me for writting in English but I can't speak German. I really think that this is maybe the best forum to ask for help about MR issues.

I can acces to the UV position of one texture in this way:

state->tex_list[0].x;

But I need the orientation too, it could be easy if I could get the .x coordinate of the adjacent pixels, it is trivial using pixel shaders:

Code: Select all

// (red channel of the texture sampler represents the u coordinate)
currentU=Tex2D(mySampler, float2(myCoords.x, myCoords.y));
nextU=Tex2D(mySampler, float2(myCoords.x+1/widthOfImage, myCoords.y));
previousU=Tex2D(mySampler, float2(myCoords.x-1/widthOfImage, myCoords.y));
upperU=Tex2D(mySampler, float2(myCoords.x, myCoords.y+1/heightOfImage));
lowerU=Tex2D(mySampler, float2(myCoords.x, myCoords.y-1/heightOfImage));

Is there any way to access to other pixels? Is there anyway to detect the orientation?

Thanks in advance, Víctor.

[Seeda]

Hi Victor.

First of all you are Welcome here on GMR.
Sorry I cant help you, but i am sure others here can do, I.e Hotchip.
I hope he will help you soon.
Regards Seeda.

[haggi]

The problem here is that you dont deal with pixels but with samples. In adaptive sampling you dont know how many samples you will have per pixel, it depends on contrast and min/max settings.

The only way to get a fixed sampling is to use the rasterizer or fixed adaptive sampling. But this will result in a bad antialiased image.

Lets imgine you have a fixed one pixel sampling.
One way to get the neighbour values would be to transform the current point in raster space. Then go one pixel further and send your own ray into the scene. But this could result in an error because you dont know which objects will be hitten by the ray. Maybe you get the coordinates of a complete different object.

[Victor]

Thanks for the answer haggi.

Code: Select all

But this could result in an error because you dont know which objects will be hitten by the ray. Maybe you get the coordinates of a complete different object.


That's not a problem i really don't need a extreme precision but an aproximate one (some pixel could be wrong and is not a real problem)