Sign in to follow this  
SuperVGA

Aliasing in pixel shader-drawn grid.

Recommended Posts

SuperVGA    1132
Hey everyone,

i have a GLSL fragment shader:
[code]
#version 130
varying vec4 position_in_world_space;
out vec4 MyFragColor;

void main(void)\n
{[attachment=4792:aliasing.png]
// Grid
vec4 circ_dist_to_grid_lines = abs( position_in_world_space - floor(position_in_world_space) );
if(circ_dist_to_grid_lines.x > 0.01 && circ_dist_to_grid_lines.x < 0.99) { circ_dist_to_grid_lines.x = 1.0; }
else{ if(circ_dist_to_grid_lines.x >= 0.99) { circ_dist_to_grid_lines.x = 1.0 - circ_dist_to_grid_lines.x; } circ_dist_to_grid_lines.x *= 100.0; }
if(circ_dist_to_grid_lines.y > 0.01 && circ_dist_to_grid_lines.y < 0.99) { circ_dist_to_grid_lines.y = 1.0; }
else{ if(circ_dist_to_grid_lines.y >= 0.99) { circ_dist_to_grid_lines.y = 1.0 - circ_dist_to_grid_lines.y; } circ_dist_to_grid_lines.y *= 100.0; }
if(circ_dist_to_grid_lines.z > 0.01 && circ_dist_to_grid_lines.z < 0.99) { circ_dist_to_grid_lines.z = 1.0; }
else{ if(circ_dist_to_grid_lines.z >= 0.99) { circ_dist_to_grid_lines.z = 1.0 - circ_dist_to_grid_lines.z; } circ_dist_to_grid_lines.z *= 100.0; }

vec3 bri = vec3(1.0) - circ_dist_to_grid_lines.xyz;
MyFragColor.rgb = bri * 0.5;
}
[/code]
(Sorry about the 0.99 and 0.01 stuff. I've yet to finish the code and clean it up.)

-Which gets an interpolated world-space position for the fragment in position_in_world_space,
and through that draws gridlines for every axis in separate colors: X:red, Y:green, Z:blue.
Here we have its output for a straight X,Y plane:
[attachment=4792:aliasing.png]

Now I'm getting severe aliasing on the gridlines, and i could obviously better it by increasing the line thickness.
Do any of you guys have a suggestion as of how to avoid the aliasing? Should i sample nearby fragments?

Thanks a bunch! :)

Share this post


Link to post
Share on other sites
rouncer    294
hey sick shader man, i never would have thought of getting an infinite grid like that.

as for the aliasing... hmm... could you multisample it? as in render a chart 4 times the res then mip it back?

Share this post


Link to post
Share on other sites
SuperVGA    1132
[quote name='rouncer' timestamp='1312719237' post='4845749']
hey sick shader man, i never would have thought of getting an infinite grid like that.

as for the aliasing... hmm... could you multisample it? as in render a chart 4 times the res then mip it back?
[/quote]
Hi Rouncer, thanks, man, it was just something i thought natural as i want to be able to add planes with grids on the fly,
and i wouldn't want to restrain any line geometry to edges of the bounding box containing the subjects.
I tried with variable line widths with some success, and here's my shot at fading down the grid so it won't be visible further away:
[attachment=4794:faded_grid.png]
That's not alias-free either, it just makes the artifacts more subtle, and there are obvious downsides with not being able to see the grid
to a fairly long extend when measuring up parts of the edited subject.

The multisampling idea sounds like it would work. Just that I'd have to transform the nearby fragment coords from screen space to world coords following the plane,
before performing the posted steps again. And I'm not sure how I could go about doing that... Then there's the possibility of converting fragment depths to linear depth values,
and changing the line width based on that. That would depend on resolution, too, however i assume that'd be alright. Just that far away gridlines might be so fat they'd take
up entire grid cells.... Hmm...

... -Just re-read your post. Obviously i could render in a higher resolution and then downsample it like you said, i just misunderstood the multisampling as supersampling. I'm not sure which would work best...

Share this post


Link to post
Share on other sites
SuperVGA    1132
There, I tried the variable thickness approach with some success.
Now only areas very close to the player becomes non-blue (the blue color one huge line indicating a threshold on the z-axis)
That aside, i have diminished the aliasing artifacts somehow:
[attachment=4795:thickness.png]

[code]
float grid_thickness = (linear_depth(gl_FragCoord.z, 0.1, 1024.0)) * 0.5; // 0.5 is the thickness closest to the far clipping plane (here: 1024.0)
float inv_grid_thickness = 1.0 / grid_thickness;
[/code]
The grid will be invisible if shown at the near clipping planes, but I suppose that's ok.
I'm not sure it's consistent with changing near-far clipping planes, and higher resolutions will leave lines in the same non-relative thickness, as such, they will appear thinner on higher resolutions.

Share this post


Link to post
Share on other sites
CryZe    773
I'd project the pixels corners onto the plane and than compare the resulting trapezoid's area to the area where the lines are overlapping the trapezoid. This percentage is the perfect anti-aliasing value. Or you do it the other way 'round.

Share this post


Link to post
Share on other sites
SuperVGA    1132
[quote name='CryZe' timestamp='1312731507' post='4845802']
I'd project the pixels corners onto the plane and than compare the resulting trapezoid's area to the area where the lines are overlapping the trapezoid. This percentage is the perfect anti-aliasing value. Or you do it the other way 'round.
[/quote]
That does sound like an effective and very thorough idea. Although, my gridlines have been anti-aliased now, and the remaining artifact seems to be a [url="http://en.wikipedia.org/wiki/Moir%C3%A9_pattern"]moire effect[/url].
But thanks for the input. I realise that the gridlines will indeed be perfectly 1wd antialiased lines with your technique. :)

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this