This series illustrate my shaders on Shadertoy.com, which renders Instagram filters in real time. Brannan Filter Brannan filter emphasizes the grey and green colors, and paints a metallic tint upon the photos. Click the mouse for comparison. I write this shader to stylize your photos in batch on my own ShaderToy renderer 🙂 (To be…
PostProcessing: Brightness, Contrast, Hue, Saturation, and Vibrance
Real-time image manipulation is always fascinating. With the power of the modern GPU, it is possible to achieve the postprocessing effects all in a single shader. Remember that in GLSL, matrix are column major. Brightness is a float between [-1, 1], directly adding to the RGB Contrast is a float between [0, 1, ∞], directly…
Code Golf: Halftone Image
This is a code golf with the help of Dr. Neyret and coyote. Demo Code
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 |
// forked from my: https://www.shadertoy.com/view/lsSfWV // taken Dr. Neyret and coyote's advices void mainImage( out vec4 O, vec2 u ) { vec2 R = iResolution.xy, M = iMouse.xy / R, S = ( M.y < 1e-3 ? 6. : mix(3., 5.5, M.x/M.y) ) / R; O = vec4( length( M= mod( u/R, S) - S*.5 ) < dot( texture(iChannel0, u/R - M), vec4( .21, .72, .07, 0) ) * S.x * ( 1. + .3 * sin(iTime) ) ); } |
Hopf Fibration
According to Wikipedia: In the mathematical field of differential topology, the Hopf fibration (also known as the Hopf bundleor Hopf map) describes a 3-sphere (a hypersphere in four-dimensional space) in terms of circles and an ordinary sphere. Discovered by Heinz Hopf in 1931, it is an influential early example of a fiber bundle. Technically, Hopf found a many-to-one continuous function (or “map”) from the 3-sphere onto the 2-sphere such that each distinct point of the 2-sphere comes from…
Unified Gnomonic and Stereographic Projections
Gnomonic projection, or rectilinear projection, together with stereographic projection, are two most commonly used projection in rendering 360 degree videos, or other VR applications. Recently, I found the inverse converting function from screen coordinates to the two projections can be unified within a single function. It’s not really surprising since both projection uses spherical lens,…
Equirectangular Projection, Gnomonic Projection, and Cubemaps
Gnomonic Projection Background According to MathWorld, the gnomonic projection is a nonconformal map projection obtained by projecting points (or ) on the surface of sphere from a sphere’s center O to point P in a plane that is tangent to a point S (Coxeter 1969, p. 93). In a gnomonic projection, great circles are mapped…
0-4 Order of Spherical Harmonics
Spherical Harmonics is widely used in Computer Graphics. They are analogue to Fourier basis on a sphere, consists of a set of orthogonal functions to represent functions defined on the surface of a sphere. However, they are very tricky to implement due to lots of constants and integral functions. Here is a real-time visualization that…
Foveated Rendering via Quadtree
Today, I wrote a shader for foveated rendering uisng Prof. Neyret’s QuadTree: https://www.shadertoy.com/view/Ml3SDf The basic idea is: Calculate the depth of the QuadTree using the distance between the current coordinate to the foveat region Use the depth as the mipmap level to sample from the texture Code below:
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 |
// forked and remixed from Prof. Neyret's https://www.shadertoy.com/view/ltBSDV // Foveated Rendering via Quadtree: https://www.shadertoy.com/view/Ml3SDf# void mainImage( out vec4 o, vec2 U ) { float r = 0.1, t = iGlobalTime, H = iResolution.y; vec2 V = U.xy / iResolution.xy; U /= H; // foveated region : disc(P,r) vec2 P = .5 + .5 * vec2(cos(t), sin(t * 0.7)), fU; U *= .5; P *= .5; // unzoom for the whole domain falls within [0,1]^n float mipmapLevel = 4.0; for (int i = 0; i < 7; ++i) { // to the infinity, and beyond ! :-) //fU = min(U,1.-U); if (min(fU.x,fU.y) < 3.*r/H) { o--; break; } // cell border if (length(P - vec2(0.5)) - r > 0.7) break; // cell is out of the shape // --- iterate to child cell fU = step(.5, U); // select child U = 2.0 * U - fU; // go to new local frame P = 2.0 * P - fU; r *= 2.0; mipmapLevel -= 0.5; } o = texture2D(iChannel0, V, mipmapLevel); } |
Bilateral Filter to Look Younger on GPU
Bilateral filter can be used to smooth the texture while preserving significant edges / contrast. Below shows a live demo in ShaderToy. Press mouse for comparison. Thanks to mrharicot’s awesome bilateral filter: https://www.shadertoy.com/view/4dfGDH With performance improvement proposed by athlete. With gamma correction by iq: https://www.shadertoy.com/view/XtsSzH Skin detection forked from carlolsson’s Skin Detection https://www.shadertoy.com/view/MlfSzn#
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 |
// Bilateral Filter for Younger. starea. // URL: https://www.shadertoy.com/view/XtVGWG // Press mouse for comparison. // Filter forked from mrharicot: https://www.shadertoy.com/view/4dfGDH // Skin detection forked from carlolsson's Skin Detection https://www.shadertoy.com/view/MlfSzn# // With performance improvement by athlete #define SIGMA 10.0 #define BSIGMA 0.1 #define MSIZE 15 #define USE_CONSTANT_KERNEL #define SKIN_DETECTION const bool GAMMA_CORRECTION = true; float kernel[MSIZE]; float normpdf(in float x, in float sigma) { return 0.39894 * exp(-0.5 * x * x/ (sigma * sigma)) / sigma; } float normpdf3(in vec3 v, in float sigma) { return 0.39894 * exp(-0.5 * dot(v,v) / (sigma * sigma)) / sigma; } float normalizeColorChannel(in float value, in float min, in float max) { return (value - min)/(max-min); } void mainImage( out vec4 fragColor, in vec2 fragCoord ) { vec3 c = texture2D(iChannel0, (fragCoord.xy / iResolution.xy)).rgb; const int kSize = (MSIZE - 1) / 2; vec3 final_colour = vec3(0.0); float Z = 0.0; #ifdef USE_CONSTANT_KERNEL // unfortunately, WebGL 1.0 does not support constant arrays... kernel[0] = kernel[14] = 0.031225216; kernel[1] = kernel[13] = 0.033322271; kernel[2] = kernel[12] = 0.035206333; kernel[3] = kernel[11] = 0.036826804; kernel[4] = kernel[10] = 0.038138565; kernel[5] = kernel[9] = 0.039104044; kernel[6] = kernel[8] = 0.039695028; kernel[7] = 0.039894000; float bZ = 0.2506642602897679; #else //create the 1-D kernel for (int j = 0; j <= kSize; ++j) { kernel[kSize+j] = kernel[kSize-j] = normpdf(float(j), SIGMA); } float bZ = 1.0 / normpdf(0.0, BSIGMA); #endif vec3 cc; float factor; //read out the texels for (int i=-kSize; i <= kSize; ++i) { for (int j=-kSize; j <= kSize; ++j) { cc = texture2D(iChannel0, (fragCoord.xy+vec2(float(i),float(j))) / iResolution.xy).rgb; factor = normpdf3(cc-c, BSIGMA) * bZ * kernel[kSize+j] * kernel[kSize+i]; Z += factor; if (GAMMA_CORRECTION) { final_colour += factor * pow(cc, vec3(2.2)); } else { final_colour += factor * cc; } } } if (GAMMA_CORRECTION) { fragColor = vec4(pow(final_colour / Z, vec3(1.0/2.2)), 1.0); } else { fragColor = vec4(final_colour / Z, 1.0); } bool isSkin = true; #ifdef SKIN_DETECTION isSkin = false; vec4 rgb = fragColor * 255.0; vec4 ycbcr = rgb; ycbcr.x = 16.0 + rgb.x*0.257 + rgb.y*0.504 + rgb.z*0.098; ycbcr.y = 128.0 - rgb.x*0.148 - rgb.y*0.291 + rgb.z*0.439; ycbcr.z = 128.0 + rgb.x*0.439 - rgb.y*0.368 - rgb.z*0.071; if (ycbcr.y > 100.0 && ycbcr.y < 118.0 && ycbcr.z > 121.0 && ycbcr.z < 161.0) { isSkin = true; } #endif if (iMouse.z > 0.0 || !isSkin) { fragColor = vec4(texture2D(iChannel0, fragCoord.xy / iResolution.xy).xyz, 1.0); } } |
…
Interactive Poisson Blending on GPU
Recently, I have implemented two fragment shaders for interactive Poisson Blending (seamless blending). Here is my real-time demo and code on ShaderToy: Press 1 for normal mode, press 2 for mixed gradients, press space to clear the canvas. Technical Brief I followed a simplified routine of the Poisson Image Editing paper [P. Pérez, M. Gangnet,…
Slideshow
