Convert to Hybrid Log Gamma

This commit is contained in:
Sam Belliveau 2024-03-26 14:06:56 -04:00
parent 1efda863e4
commit 50512366c3
1 changed files with 49 additions and 29 deletions

View File

@ -17,45 +17,65 @@ DefaultValue = 2.5
/***** Transfer Function *****/
const float4 m_1 = float4(2610.0 / 16384.0);
const float4 m_2 = float4(128.0 * 2523.0 / 4096.0);
const float4 m_1_inv = float4(16384.0 / 2610.0);
const float4 m_2_inv = float4(4096.0 / (128.0 * 2523.0));
const float a = 0.17883277;
const float b = 1.0 - 4.0 * a;
const float c = 0.5 - a * log(4.0 * a);
const float4 c_1 = float4(3424.0 / 4096.0);
const float4 c_2 = float4(2413.0 / 4096.0 * 32.0);
const float4 c_3 = float4(2392.0 / 4096.0 * 32.0);
float HLG_f(float x)
{
if (x < 0.0) {
return 0.0;
}
float4 EOTF_inv(float4 lms) {
float4 y = pow(lms, m_1);
return pow((c_1 + c_2 * y) / (1.0 + c_3 * y), m_2);
else if (x < 1.0 / 12.0) {
return sqrt(3.0 * x);
}
return a * log(12.0 * x - b) + c;
}
float4 EOTF(float4 lms) {
float4 x = pow(lms, m_2_inv);
return pow(-(x - c_1) / (c_3 * x - c_2), m_1_inv);
float HLG_inv_f(float x)
{
if (x < 0.0) {
return 0.0;
}
else if (x < 1.0 / 2.0) {
return x * x / 3.0;
}
return (exp((x - c) / a) + b) / 12.0;
}
// This is required as scaling in EOTF space is not linear.
float EOTF_AMPLIFICATION = EOTF_inv(float4(AMPLIFICATION)).x;
float4 HLG(float4 lms)
{
return float4(HLG_f(lms.x), HLG_f(lms.y), HLG_f(lms.z), lms.w);
}
float4 HLG_inv(float4 lms)
{
return float4(HLG_inv_f(lms.x), HLG_inv_f(lms.y), HLG_inv_f(lms.z), lms.w);
}
/***** Linear <--> ICtCp *****/
const mat4 RGBtoLMS = mat4(
1688.0, 683.0, 99.0, 0.0,
2146.0, 2951.0, 309.0, 0.0,
262.0, 462.0, 3688.0, 0.0,
0.0, 0.0, 0.0, 4096.0) / 4096.0;
1688.0, 683.0, 99.0, 0.0,
2146.0, 2951.0, 309.0, 0.0,
262.0, 462.0, 3688.0, 0.0,
0.0, 0.0, 0.0, 4096.0)
/ 4096.0;
const mat4 LMStoICtCp = mat4(
+2048.0, +6610.0, +17933.0, 0.0,
+2048.0, -13613.0, -17390.0, 0.0,
+0.0, +7003.0, -543.0, 0.0,
+0.0, +0.0, +0.0, 4096.0) / 4096.0;
+2048.0, +3625.0, +9500.0, 0.0,
+2048.0, -7465.0, -9212.0, 0.0,
+0.0, +3840.0, -288.0, 0.0,
+0.0, +0.0, +0.0, 4096.0)
/ 4096.0;
float4 LinearRGBToICtCP(float4 c)
{
return LMStoICtCp * EOTF_inv(RGBtoLMS * c);
return LMStoICtCp * HLG(RGBtoLMS * c);
}
/***** ICtCp <--> Linear *****/
@ -65,7 +85,7 @@ mat4 LMStoRGB = inverse(RGBtoLMS);
float4 ICtCpToLinearRGB(float4 c)
{
return LMStoRGB * EOTF(ICtCptoLMS * c);
return LMStoRGB * HLG_inv(ICtCptoLMS * c);
}
void main()
@ -88,19 +108,19 @@ void main()
// Scale the color in perceptual space depending on the percieved luminance.
//
// At low luminances, ~0.0, pow(EOTF_AMPLIFICATION, ~0.0) ~= 1.0, so the
// At low luminances, ~0.0, pow(AMPLIFICATION, ~0.0) ~= 1.0, so the
// color will appear to be unchanged. This is important as we don't want to
// over expose dark colors which would not have otherwise been seen.
//
// At high luminances, ~1.0, pow(EOTF_AMPLIFICATION, ~1.0) ~= EOTF_AMPLIFICATION,
// which is equivilant to scaling the color by EOTF_AMPLIFICATION. This is
// At high luminances, ~1.0, pow(AMPLIFICATION, ~1.0) ~= AMPLIFICATION,
// which is equivilant to scaling the color by AMPLIFICATION. This is
// important as we want to get the most out of the display, and we want to
// get bright colors to hit their target brightness.
//
// For more information, see this desmos demonstrating this scaling process:
// https://www.desmos.com/calculator/syjyrjsj5c
const float luminance = ictcp_color.x;
ictcp_color *= pow(EOTF_AMPLIFICATION, luminance);
ictcp_color *= HLG_f(pow(AMPLIFICATION, luminance));
// Convert back to Linear RGB and output the color to the display.
// We use hdr_paper_white to renormalize the color to the comfortable