Hey, that’s a nice idea! Looks like a smart resampler.
4 Likes
This example looks like a really nice antialiasing effect 
4 Likes
Here are the shader files with some presets for the “new” Lanczos resampler.
Download link:
https://mega.nz/file/whAlFDYJ#H68MzUoJxR7jmppfQU3mTuVAQKKHjwJOIVuabKaAPVc
It’s a two pass composition, but can be also used separately if needed.
The benefits are that filter range can be adjusted, which can be really nice for downsampling…
Good use cases:
- downsampling to a low resolution
- downsampling super-sampled sources to viewport
- antialiasing scaled graphics (xbr, scalefx…)
- edge smoothing on high resolution content (PSX, PS2, DC,…)
- can be also used as an ordinary resampler
6 Likes
I was looking at screenshots from my CRT to better match what it does on filtering, and i think windowed are the closest indeed.
raw screenshot
After applied gaussian on GIMP to “remove” mask etc to view shapes better.
bright pixels definitely enlarge, mostly horizontally “shaping” a kind of curve, or “spikes” to better describe it
3 Likes
Tested here and they worked as expected! Very useful for downsampling! Congrats!
3 Likes
Tried with ps1 @1080p internal resolution,
- filter range 3.5
- sharpness 1.3
- downsample 5
“240p” content
raw
shader on
with “hi res” content I like double scan ON
raw
shader on
3 Likes
As single pass resamplers have some benefits, i have developed a new “g-sharp2” single pass resampler.
Compared with the first version, the benefits are:
- much faster with same filter range (approx 2x)
- better image processing
- can be much faster with downsampling (due to range boost with same # of lookups)
@HyperspaceMadness can you do some “DREZ” testings? Should do quite better then the first version. I belive it could be a good upgrade.
Edit: 2-pass version added.
Screenie (2-pass version):
Edit: 2-pass version tweak, better AR
Download link:
https://mega.nz/file/4kxVhawQ#N_GMHOnzqudbX4vL3jkyRmomYlDIrKGsgJDr1AaNVrU
10 Likes
That’s a nice flexible shader. Is there any possibility to derivate some fast bilateral from these ones? I think we lack more bilateral options for now.
1 Like
Hey @Hyllian (and everybody else interested), can you test it a bit? Should run nicely with base resolution. I added some options, so it isn’t so boring. 
#version 450
/*
Bilateral filter - dynamic range (upscaler, downsampler)
Copyright (C) 2024 guest(r)
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; either version 2
of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
layout(push_constant) uniform Push
{
vec4 SourceSize;
vec4 OriginalSize;
vec4 OutputSize;
uint FrameCount;
float FRANGE;
float FBSMOOTH;
float FBOOST;
} params;
#pragma parameter FRANGE "Filter Range" 1.0 0.5 5.0 0.5
#define FRANGE params.FRANGE
#pragma parameter FBSMOOTH "Filter Base Smoothing" 0.55 0.05 1.0 0.025
#define FBSMOOTH params.FBSMOOTH
#pragma parameter FBOOST "Boost Smoothing w. dist" 0.0 0.0 5.0 0.25
#define FBOOST params.FBOOST
layout(std140, set = 0, binding = 0) uniform UBO
{
mat4 MVP;
} global;
#pragma stage vertex
layout(location = 0) in vec4 Position;
layout(location = 1) in vec2 TexCoord;
layout(location = 0) out vec2 vTexCoord;
void main()
{
gl_Position = global.MVP * Position;
vTexCoord = TexCoord * 1.00001;
}
#pragma stage fragment
layout(location = 0) in vec2 vTexCoord;
layout(location = 0) out vec4 FragColor;
layout(set = 0, binding = 2) uniform sampler2D Source;
#define COMPAT_TEXTURE(c,d) texture(c,d)
#define SourceSize params.SourceSize
float wt(vec3 A, vec3 B)
{
return clamp(FBSMOOTH - ((6.0+0.25)/3.0) * dot(pow(abs(A-B),vec3(1.0/1.0)),1.0.xxx)/(dot(A+B,1.0.xxx)+0.25), 0.0001, 0.25);
}
float getw(float x, vec3 c, vec3 p)
{
float y = max(2.0-x, 0.0) + FBOOST;
float d = wt(c,p);
return y*d;
}
void main()
{
vec2 pos = vTexCoord * SourceSize.xy;
vec2 f = 0.5-fract(pos);
vec2 tex = floor(pos)*SourceSize.zw + 0.5*SourceSize.zw;
vec3 color = 0.0.xxx;
vec2 dx = vec2(SourceSize.z, 0.0);
vec2 dy = vec2(0.0, SourceSize.w);
float w, fp;
float wsum = 0.0;
vec3 pixel;
float FPR = FRANGE;
float FPR2 = 2.0*FPR;
float FPR3 = FPR2*FPR2;
float LOOPSIZE = FPR2;
float y = -LOOPSIZE;
float x = 0.0;
vec3 comp = COMPAT_TEXTURE(Source, tex).rgb;
do
{
x = -LOOPSIZE;
do
{
fp = dot(vec2(x+f.x,y+f.y),vec2(x+f.x,y+f.y));
if (fp >= FPR3) w = 0.0;
else
{
pixel = COMPAT_TEXTURE(Source, tex + x*dx + y*dy).rgb;
fp = sqrt(fp)/FPR;
w = getw(fp,comp,pixel);
color = color + w * pixel;
wsum = wsum + w;
}
x = x + 1.0;
} while (x <= LOOPSIZE);
y = y + 1.0;
} while (y <= LOOPSIZE);
color = color / wsum;
FragColor = vec4(color, 1.0);
}
3 Likes
Hey! Nice shader! I just tested here and it works great! For my tests and taste, the most sane options for me (tested with survival games) were these:
#pragma parameter FRANGE “Filter Range” 0.5 0.5 5.0 0.5 #pragma parameter FBSMOOTH “Filter Base Smoothing” 0.50 0.05 1.0 0.025 #pragma parameter FBOOST “Boost Smoothing w. dist” 1.0 0.0 5.0 0.25
I try to adjust the parameters to get rid of noise and let texture alone.
For instance, you can spot the transition between the sidewalk and the wall/door clearly at the original image here:
Using default params they are blended:
Using 0.5, 0.5 and 1.0 you can still see the transition between sidewalk and wall/door:
2 Likes
Thanks for testing!
The shader framework is OK i guess, but the effect largely depends on the metric. Please feel free to add something different (the float wt(vec3 A, vec3 B) function), maybe it could look better…
2 Likes
Small but nice update to g-sharp_resampler 2.0.
Small speed gains with somewhat nicer appearance.
Update:
- filtering quality improvements
- kernel optimizations, notable performance increase
- dedithering (source-1x) preset added, needs some parameter tweaks
Download link:
https://mega.nz/file/U1YHmIzS#KDhQceABx9g4oYvf3HkNHANTymyGt0X3fUDYIxDeXFg
5 Likes
I made a slightly change to the wt formula and I think it preserves more some details. No idea if it’s just placebo, though. Using luma to calc intensity differences instead adding color channels:
float wt(vec3 A, vec3 B)
{
// return clamp(FBSMOOTH - ((6.0+0.25)/3.0) * dot(pow(abs(A-B),vec3(1.0/1.0)),1.0.xxx)/(dot(A+B,1.0.xxx)+0.25), 0.0001, 0.25);
return clamp(FBSMOOTH - ((6.0+0.25)/1.0) * dot(pow(abs(A-B),vec3(1.0/1.0)),luma)/(dot(A+B,luma)+0.25), 0.0001, 0.25);
}EDIT: seeing some more examples now, I think it isn’t better or worse, just different. First formula prioritize color changes and the second, luma changes.
2 Likes