[XNA] Aligning texels to pixels doesn't work for small textures

Started by
11 comments, last by Spa8nky 12 years, 2 months ago
I'm attempting to draw a texture that is 3 pixels by 3 pixels in size.

If I draw it using SpriteBatch then the texels are correctly aligned to the pixels

If I draw it using Texture Atlas look up then the texels are not correctly aligned and the x coordinate doesn't line up with the SpriteBatch result.

If I draw it straight from the original texture without the Atlas look up then the texels are not correctly aligned either.

Here is a picture demonstrating the results (All 3 quads have an x position of 100 yet they do not line up):

SmallTexturePixelAlignProblem.png

I'm attempting to line up the pixels to texels as follows:

(C#):



// [DirectX 9 - Align pixels to texels]
// • Find size of half a pixel for render targets based on their size
// In this case render target size is backBufferWidth * backBufferHeight
// • The reason that it is 1 / width instead of 0.5 / width is because, in clip space,
// the coordinates range from -1 to 1 (width 2), and not from 0 to 1 (width 1)
// like they do in textures, so you need to double the movement to account for that
Vector4 halfPixelTexel = new Vector4();

// Half Pixel
//halfPixelTexel.X = 1f / textureAtlas.Texture.Width;
//halfPixelTexel.Y = -1f / textureAtlas.Texture.Height;
halfPixelTexel.X = 1f / GraphicsDevice.Viewport.Width;
halfPixelTexel.Y = -1f / GraphicsDevice.Viewport.Height;
//halfPixelTexel.X = 1f / GraphicsDevice.PresentationParameters.BackBufferWidth;
//halfPixelTexel.Y = -1f / GraphicsDevice.PresentationParameters.BackBufferHeight;

// Half Texel
halfPixelTexel.Z = 0.5f / textureAtlas.Texture.Width;
halfPixelTexel.W = 0.5f / textureAtlas.Texture.Height;
//halfPixelTexel.Z = 0.5f / backBufferDim.X;
//halfPixelTexel.W = 0.5f / backBufferDim.Y;

effects[EffectName.Quadrangle].Parameters["HalfPixelTexel"].SetValue(halfPixelTexel);
effects[EffectName.QuadrangleBatch].Parameters["HalfPixelTexel"].SetValue(halfPixelTexel);


HLSL:



VertexShaderOutput VertexShaderFunction(VertexShaderInput input)
{
VertexShaderOutput Output = (VertexShaderOutput)0;

// Set world transform here (don't use a static identity matrix as it is slower)
float4x4 World =
{
// [Rotation, Scale]
input.Column1.x * input.TexCoordDimension.z, // * Scale X
input.Column1.y * input.TexCoordDimension.z,
input.Column1.z * input.TexCoordDimension.z,
input.Column1.w,
input.Column2.x * input.TexCoordDimension.w, // * Scale Y
input.Column2.y * input.TexCoordDimension.w,
input.Column2.z * input.TexCoordDimension.w,
input.Column2.w,
input.Column3.x,
input.Column3.y,
input.Column3.z,
input.Column3.w,

// [Translation]
input.Column4.x,
input.Column4.y,
input.Column4.z,
input.Column4.w
};

// 'input.Position.w' should always equal 1
input.Position.w = 1;

// [Transformation]
// • Multiplying input.Position by World, then the result by ViewProjection is fast
// • Concatenating World and ViewProjection matrices then multiplying input.Position by the result is slower
input.Position = mul(input.Position, World);
Output.Position = mul(input.Position, ViewProjection);

// [Texel To Pixel Align]
// • Half pixel offset for correct texel centering)
// • Should be done AFTER transformation?
Output.Position.xy -= HalfPixelTexel.xy;

// [UV Coordinates]
Output.TextureCoordinates.xy = input.TexCoordDimension.xy;

// [Alpha]
Output.TextureCoordinates.w = input.Normal.w;

return Output;
}


When using textures larger than 8 pixels this problem does not occur.

Can anyone please explain why the texels are not lining up correctly with the pixels with smaller textures?
Advertisement
The half-pixel thing is relative to your screen, not to the texture size. That means that you would commonly just subtract 0.5 from x and y in your vertex shader, before transforming the input position.

Direct3D has need of instancing, but we do not. We have plenty of glVertexAttrib calls.

I still get the same results with:



VertexShaderOutput VertexShaderFunction(VertexShaderInput input)
{
VertexShaderOutput Output = (VertexShaderOutput)0;

// Set world transform here (don't use a static identity matrix as it is slower)
float4x4 World =
{
// [Rotation, Scale]
input.Column1.x * input.TexCoordDimension.z, // * Scale X
input.Column1.y * input.TexCoordDimension.z,
input.Column1.z * input.TexCoordDimension.z,
input.Column1.w,
input.Column2.x * input.TexCoordDimension.w, // * Scale Y
input.Column2.y * input.TexCoordDimension.w,
input.Column2.z * input.TexCoordDimension.w,
input.Column2.w,
input.Column3.x,
input.Column3.y,
input.Column3.z,
input.Column3.w,

// [Translation]
input.Column4.x,
input.Column4.y,
input.Column4.z,
input.Column4.w
};

// 'input.Position.w' should always equal 1
input.Position.w = 1;

input.Position.xy -= 0.5;

// [Transformation]
// • Multiplying input.Position by World, then the result by ViewProjection is fast
// • Concatenating World and ViewProjection matrices then multiplying input.Position by the result is slower
input.Position = mul(input.Position, World);
Output.Position = mul(input.Position, ViewProjection);

// [UV Coordinates]
Output.TextureCoordinates.xy = input.TexCoordDimension.xy;

// [Alpha]
Output.TextureCoordinates.w = input.Normal.w;

// [Source Rectangles]
// • Pass the source rectangles to the pixel shader for processing
Output.SourceRectangle0 = input.SourceRectangle0;
Output.SourceRectangle1 = input.SourceRectangle1;

return Output;
}


The SpriteBatch shader subtracts the 0.5 after transformation:




// Apply the matrix transform.
outputPosition = mul(float4(position, depth, 1), transpose(MatrixTransform));

// Half pixel offset for correct texel centering.
outputPosition.xy -= 0.5;


This method also produces the same incorrect result with my shader. :(
Subtract before transformation, not after.

You could even encode the subtraction into your world matrix, but I'd advise to get it working this way first.

Direct3D has need of instancing, but we do not. We have plenty of glVertexAttrib calls.


Subtract before transformation, not after.

You could even encode the subtraction into your world matrix, but I'd advise to get it working this way first.


Sorry if I didn't make it clear in my last post but I have already tried that.


[color=#000000]input[color=#666600].[color=#660066]Position[color=#666600].[color=#000000]xy [color=#666600]-=[color=#000000] [color=#006666]0.5[color=#666600];

[color=#880000]// [Transformation]
[color=#880000]// • Multiplying input.Position by World, then the result by ViewProjection is fast
[color=#880000]// • Concatenating World and ViewProjection matrices then multiplying input.Position by the result is slower
[color=#000000]input[color=#666600].[color=#660066]Position[color=#000000] [color=#666600]=[color=#000000] mul[color=#666600]([color=#000000]input[color=#666600].[color=#660066]Position[color=#666600],[color=#000000] [color=#660066]World[color=#666600]);[color=#000000]
[color=#660066]Output[color=#666600].[color=#660066]Position[color=#000000] [color=#666600]=[color=#000000] mul[color=#666600]([color=#000000]input[color=#666600].[color=#660066]Position[color=#666600],[color=#000000] [color=#660066]ViewProjection[color=#666600]);

The lines above in the last post are what I am referring to.

It doesn't work either.
Just glancing at this, 3x3 texture is an odd size? Some hardware doesnt like non-power of 2 textures (although is this day and age its less of an issue?) I never tend to use anything lower than 8x8. Also are you clamping your textures?

why not use a larger texture and just leave space?

what is your final aim?

Just glancing at this, 3x3 texture is an odd size? Some hardware doesnt like non-power of 2 textures (although is this day and age its less of an issue?) I never tend to use anything lower than 8x8.


Going the reflection of SpriteBatch, and its accompanying shader, I can't see anything that resizes the 3x3 sprite. SpriteBatch draws it correctly too.


why not use a larger texture and just leave space?
[/quote]

I could do but I wanted to get to the bottom of what was causing this problem.


[color=#282828][font=helvetica, arial, verdana, tahoma, sans-serif]

Also are you clamping your textures?

[/font]
[/quote]

Yes:



sampler TextureSampler = sampler_state
{
Texture = <Texture>;

MinFilter = Point;
MagFilter = Point;
MipFilter = Point; // None

AddressU = Clamp;
AddressV = Clamp;
};

The uvs you supply you need to pinch them 1/2 texel, not just move them all by the same value.

htw = half texel width, hth = half texel height.

As in, u(0) += htw, v(0) += hth - u(1) -= htw, v(1) -= hth

Other corner can be worked out from these.

The uvs you supply you need to pinch them 1/2 texel, not just move them all by the same value.

As in, u(0) += htw, v(0) += hth - u(1) -= htw, v(1) -= hth


Would you agree that the following clamping method is equivalent to your pinching method?


// Half Pixel Clamp (Pinch)

//input.Position.x = max(input.Position.x, HalfPixelTexel.x);
//input.Position.y = max(input.Position.y, HalfPixelTexel.y);
//input.Position.x = min(input.Position.x, 1.0 - HalfPixelTexel.x);
//input.Position.y = min(input.Position.y, 1.0 - HalfPixelTexel.y);

// Half Texel Clamp (Pinch)
input.Position.x = max(input.Position.x, HalfPixelTexel.z);
input.Position.y = max(input.Position.y, HalfPixelTexel.w);
input.Position.x = min(input.Position.x, 1.0 - HalfPixelTexel.z);
input.Position.y = min(input.Position.y, 1.0 - HalfPixelTexel.w);


Neither the half texel clamp or the half pixel clamp produce the correct result when applied either before or after transformation.

Would you agree that the following clamping method is equivalent to your pinching method?


Only if you're using Position as the UV coords, if position is what it implies then it won't help, forget about modifying the vertices positions and change the UV input.

EDIT: and pinch regardless rather than use min, max.

This topic is closed to new replies.

Advertisement