glsl convolution shader trouble

Beginners
1

Hi guys!

I’m trying to expand the blur shader in the wiki page into a convolution shader, and I’ve encountered some issues.
blur-example.jpg

convolve-glsl.png
convolve-glsl.png837×654 28.2 KB

  • The blur kernel is the only one that works, but using multiple passes would result in a darker output, although the weights seem correct.

  • The sharpen/edge kernels don’t work as expected.

  • Is there a way to write to a texture buffer that you could adress (exclusively) inside the glslTOP without using a TOP network? (without a render pass that you’d feed back to the glslTOP)?
    glsl_convolution.toe (5.68 KB)

2

On line 18 in your shader, I think you want a different expression.

int row = j+1; // make it zero indexed int col = i+1; // make it zero indexed int myIndex = col + row*3; // now myIndex goes from 0 to 8
Then you’d use

_kernel[myIndex]

Also refer to en.wikipedia.org/wiki/Kernel_(image_processing
The kerEdge should probably have an 8 in its center. Edge detection kernels sum to 0 but blur kernels sum to 1.

3

The indexing did the trick, thanks!
Now they work as intended.

4

Is there any way I could make all of them work in the same glslTOP?

For example applying the blur three times (ideally without using the ‘passes’ slider)

Right now the convolution is happening on a 2D input.

Is there any way to write to a framebuffer and then read from it without using external TOPs?

5

You could do this

layout (location=0) out vec4 fragColor1; layout (location=1) out vec4 fragColor2; layout (location=1) out vec4 fragColor3;
then

fragColor1 = color1;
fragColor2 = color2;
fragColor3 = color3;

You’d set the numcolorbufs parameter on the GLSL TOP to 3 and then use Render Select TOPs.

Can you blur three times without using the passes slider? No, I don’t think so. Perhaps you’d be interested in my gaussian blur tox viewtopic.php?f=4&t=9330&p=35278&hilit=gaussian#p35278

I’m not sure I understand your third question.

6

I’m pretty new to GLSL. Tried out a few GLSL viewers and they handle frame buffers differently, so I was wondering how it works in TD.

I was thinking if I want to apply a blur and then a sharpen in the same GLSL TOP, I’d have to use a Render Select TOP just to feed that back as a 2nd input to my GLSL TOP, so I was wondering if there was a better way to do it.

I was reading of texture buffers on the wiki.

texture-buffers.jpg
texture-buffers.jpg1437×183 88.6 KB

What is the difference between these and color buffers?

7

Sometimes it’s great to use multiple passes, but in this situation I think you should have a unique GLSL TOP for each step of the process.

However, you can do something like this:

	if (uTDPass == 0) {
// do your blur stuff and set color to something else
	} else {
// do your sharpen stuff and set color.
	}[/code]
I've read from a few sources that this conditional if-else branching is slow for the GPU. It's as if the GPU computes the blur and sharpen every pass.

As far as I know, you can do texelFetch on sTD2DInputs[], like
[code]
vec4 color = texelFetch(sTD2DInputs[0], ivec2(gl_FragCoord.xy),0);
[/code]
and texelFetch on a samplerBuffer that has been put on the Arrays page of the GLSL TOP.

[code]uniform samplerBuffer posBuffer;
void main() {
int id = 0; // an integer
fragColor = texelFetch(posBuffer, id);
}