CfL_Prediction.glsl (12880B)
1 // MIT License 2 3 // Copyright (c) 2023 João Chrisóstomo 4 5 // Permission is hereby granted, free of charge, to any person obtaining a copy 6 // of this software and associated documentation files (the "Software"), to deal 7 // in the Software without restriction, including without limitation the rights 8 // to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 // copies of the Software, and to permit persons to whom the Software is 10 // furnished to do so, subject to the following conditions: 11 12 // The above copyright notice and this permission notice shall be included in all 13 // copies or substantial portions of the Software. 14 15 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 // IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 // FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 // AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 // LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 // OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 // SOFTWARE. 22 23 //!PARAM chroma_offset_x 24 //!TYPE float 25 0.0 26 27 //!PARAM chroma_offset_y 28 //!TYPE float 29 0.0 30 31 //!HOOK CHROMA 32 //!BIND LUMA 33 //!BIND CHROMA 34 //!SAVE LUMA_LR 35 //!WIDTH CHROMA.w 36 //!HEIGHT LUMA.h 37 //!WHEN CHROMA.w LUMA.w < 38 //!DESC Chroma From Luma Prediction (Hermite 1st step, Downscaling Luma) 39 40 float comp_wd(vec2 v) { 41 float x = min(length(v), 1.0); 42 return smoothstep(0.0, 1.0, 1.0 - x); 43 } 44 45 vec4 hook() { 46 vec2 luma_pos = LUMA_pos; 47 luma_pos.x += chroma_offset_x / LUMA_size.x; 48 float start = ceil((luma_pos.x - (1.0 / CHROMA_size.x)) * LUMA_size.x - 0.5); 49 float end = floor((luma_pos.x + (1.0 / CHROMA_size.x)) * LUMA_size.x - 0.5); 50 51 float wt = 0.0; 52 float luma_sum = 0.0; 53 vec2 pos = luma_pos; 54 55 for (float dx = start.x; dx <= end.x; dx++) { 56 pos.x = LUMA_pt.x * (dx + 0.5); 57 vec2 dist = (pos - luma_pos) * CHROMA_size; 58 float wd = comp_wd(dist); 59 float luma_pix = LUMA_tex(pos).x; 60 luma_sum += wd * luma_pix; 61 wt += wd; 62 } 63 64 vec4 output_pix = vec4(luma_sum /= wt, 0.0, 0.0, 1.0); 65 return clamp(output_pix, 0.0, 1.0); 66 } 67 68 //!HOOK CHROMA 69 //!BIND LUMA_LR 70 //!BIND CHROMA 71 //!BIND LUMA 72 //!SAVE LUMA_LR 73 //!WIDTH CHROMA.w 74 //!HEIGHT CHROMA.h 75 //!WHEN CHROMA.w LUMA.w < 76 //!DESC Chroma From Luma Prediction (Hermite 2nd step, Downscaling Luma) 77 78 float comp_wd(vec2 v) { 79 float x = min(length(v), 1.0); 80 return smoothstep(0.0, 1.0, 1.0 - x); 81 } 82 83 vec4 hook() { 84 vec2 luma_pos = LUMA_LR_pos; 85 luma_pos.y += chroma_offset_y / LUMA_LR_size.y; 86 float start = ceil((luma_pos.y - (1.0 / CHROMA_size.y)) * LUMA_LR_size.y - 0.5); 87 float end = floor((luma_pos.y + (1.0 / CHROMA_size.y)) * LUMA_LR_size.y - 0.5); 88 89 float wt = 0.0; 90 float luma_sum = 0.0; 91 vec2 pos = luma_pos; 92 93 for (float dy = start; dy <= end; dy++) { 94 pos.y = LUMA_LR_pt.y * (dy + 0.5); 95 vec2 dist = (pos - luma_pos) * CHROMA_size; 96 float wd = comp_wd(dist); 97 float luma_pix = LUMA_LR_tex(pos).x; 98 luma_sum += wd * luma_pix; 99 wt += wd; 100 } 101 102 vec4 output_pix = vec4(luma_sum /= wt, 0.0, 0.0, 1.0); 103 return clamp(output_pix, 0.0, 1.0); 104 } 105 106 //!HOOK CHROMA 107 //!BIND HOOKED 108 //!BIND LUMA 109 //!BIND LUMA_LR 110 //!WHEN CHROMA.w LUMA.w < 111 //!WIDTH LUMA.w 112 //!HEIGHT LUMA.h 113 //!OFFSET ALIGN 114 //!DESC Chroma From Luma Prediction (Upscaling Chroma) 115 116 #define USE_12_TAP_REGRESSION 1 117 #define USE_8_TAP_REGRESSIONS 1 118 #define DEBUG 0 119 120 float comp_wd(vec2 v) { 121 float d = min(length(v), 2.0); 122 float d2 = d * d; 123 float d3 = d2 * d; 124 125 if (d < 1.0) { 126 return 1.25 * d3 - 2.25 * d2 + 1.0; 127 } else { 128 return -0.75 * d3 + 3.75 * d2 - 6.0 * d + 3.0; 129 } 130 } 131 132 vec4 hook() { 133 float ar_strength = 0.8; 134 vec2 mix_coeff = vec2(0.8); 135 vec2 corr_exponent = vec2(4.0); 136 137 vec4 output_pix = vec4(0.0, 0.0, 0.0, 1.0); 138 float luma_zero = LUMA_texOff(0.0).x; 139 140 vec2 pp = HOOKED_pos * HOOKED_size - vec2(0.5); 141 vec2 fp = floor(pp); 142 pp -= fp; 143 144 #ifdef HOOKED_gather 145 vec2 quad_idx[4] = {{0.0, 0.0}, {2.0, 0.0}, {0.0, 2.0}, {2.0, 2.0}}; 146 147 vec4 luma_quads[4]; 148 vec4 chroma_quads[4][2]; 149 150 for (int i = 0; i < 4; i++) { 151 luma_quads[i] = LUMA_LR_gather(vec2((fp + quad_idx[i]) * HOOKED_pt), 0); 152 chroma_quads[i][0] = HOOKED_gather(vec2((fp + quad_idx[i]) * HOOKED_pt), 0); 153 chroma_quads[i][1] = HOOKED_gather(vec2((fp + quad_idx[i]) * HOOKED_pt), 1); 154 } 155 156 vec2 chroma_pixels[16]; 157 chroma_pixels[0] = vec2(chroma_quads[0][0].w, chroma_quads[0][1].w); 158 chroma_pixels[1] = vec2(chroma_quads[0][0].z, chroma_quads[0][1].z); 159 chroma_pixels[2] = vec2(chroma_quads[1][0].w, chroma_quads[1][1].w); 160 chroma_pixels[3] = vec2(chroma_quads[1][0].z, chroma_quads[1][1].z); 161 chroma_pixels[4] = vec2(chroma_quads[0][0].x, chroma_quads[0][1].x); 162 chroma_pixels[5] = vec2(chroma_quads[0][0].y, chroma_quads[0][1].y); 163 chroma_pixels[6] = vec2(chroma_quads[1][0].x, chroma_quads[1][1].x); 164 chroma_pixels[7] = vec2(chroma_quads[1][0].y, chroma_quads[1][1].y); 165 chroma_pixels[8] = vec2(chroma_quads[2][0].w, chroma_quads[2][1].w); 166 chroma_pixels[9] = vec2(chroma_quads[2][0].z, chroma_quads[2][1].z); 167 chroma_pixels[10] = vec2(chroma_quads[3][0].w, chroma_quads[3][1].w); 168 chroma_pixels[11] = vec2(chroma_quads[3][0].z, chroma_quads[3][1].z); 169 chroma_pixels[12] = vec2(chroma_quads[2][0].x, chroma_quads[2][1].x); 170 chroma_pixels[13] = vec2(chroma_quads[2][0].y, chroma_quads[2][1].y); 171 chroma_pixels[14] = vec2(chroma_quads[3][0].x, chroma_quads[3][1].x); 172 chroma_pixels[15] = vec2(chroma_quads[3][0].y, chroma_quads[3][1].y); 173 174 float luma_pixels[16]; 175 luma_pixels[0] = luma_quads[0].w; 176 luma_pixels[1] = luma_quads[0].z; 177 luma_pixels[2] = luma_quads[1].w; 178 luma_pixels[3] = luma_quads[1].z; 179 luma_pixels[4] = luma_quads[0].x; 180 luma_pixels[5] = luma_quads[0].y; 181 luma_pixels[6] = luma_quads[1].x; 182 luma_pixels[7] = luma_quads[1].y; 183 luma_pixels[8] = luma_quads[2].w; 184 luma_pixels[9] = luma_quads[2].z; 185 luma_pixels[10] = luma_quads[3].w; 186 luma_pixels[11] = luma_quads[3].z; 187 luma_pixels[12] = luma_quads[2].x; 188 luma_pixels[13] = luma_quads[2].y; 189 luma_pixels[14] = luma_quads[3].x; 190 luma_pixels[15] = luma_quads[3].y; 191 #else 192 vec2 pix_idx[16] = {{-0.5,-0.5}, {0.5,-0.5}, {1.5,-0.5}, {2.5,-0.5}, 193 {-0.5, 0.5}, {0.5, 0.5}, {1.5, 0.5}, {2.5, 0.5}, 194 {-0.5, 1.5}, {0.5, 1.5}, {1.5, 1.5}, {2.5, 1.5}, 195 {-0.5, 2.5}, {0.5, 2.5}, {1.5, 2.5}, {2.5, 2.5}}; 196 197 float luma_pixels[16]; 198 vec2 chroma_pixels[16]; 199 200 for (int i = 0; i < 16; i++) { 201 luma_pixels[i] = LUMA_LR_tex(vec2((fp + pix_idx[i]) * HOOKED_pt)).x; 202 chroma_pixels[i] = HOOKED_tex(vec2((fp + pix_idx[i]) * HOOKED_pt)).xy; 203 } 204 #endif 205 206 #if (DEBUG == 1) 207 vec2 chroma_spatial = vec2(0.5); 208 mix_coeff = vec2(1.0); 209 #else 210 float wd[16]; 211 float wt = 0.0; 212 vec2 ct = vec2(0.0); 213 214 vec2 chroma_min = min(min(min(chroma_pixels[5], chroma_pixels[6]), chroma_pixels[9]), chroma_pixels[10]); 215 vec2 chroma_max = max(max(max(chroma_pixels[5], chroma_pixels[6]), chroma_pixels[9]), chroma_pixels[10]); 216 217 const int dx[16] = {-1, 0, 1, 2, -1, 0, 1, 2, -1, 0, 1, 2, -1, 0, 1, 2}; 218 const int dy[16] = {-1, -1, -1, -1, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2}; 219 220 for (int i = 0; i < 16; i++) { 221 wd[i] = comp_wd(vec2(dx[i], dy[i]) - pp); 222 wt += wd[i]; 223 ct += wd[i] * chroma_pixels[i]; 224 } 225 226 vec2 chroma_spatial = ct / wt; 227 chroma_spatial = clamp(mix(chroma_spatial, clamp(chroma_spatial, chroma_min, chroma_max), ar_strength), 0.0, 1.0); 228 #endif 229 230 #if (USE_12_TAP_REGRESSION == 1 || USE_8_TAP_REGRESSIONS == 1) 231 const int i12[12] = {1, 2, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14}; 232 const int i4y[4] = {1, 2, 13, 14}; 233 const int i4x[4] = {4, 7, 8, 11}; 234 const int i4[4] = {5, 6, 9, 10}; 235 236 float luma_sum_4 = 0.0; 237 float luma_sum_4y = 0.0; 238 float luma_sum_4x = 0.0; 239 vec2 chroma_sum_4 = vec2(0.0); 240 vec2 chroma_sum_4y = vec2(0.0); 241 vec2 chroma_sum_4x = vec2(0.0); 242 243 for (int i = 0; i < 4; i++) { 244 luma_sum_4 += luma_pixels[i4[i]]; 245 luma_sum_4y += luma_pixels[i4y[i]]; 246 luma_sum_4x += luma_pixels[i4x[i]]; 247 chroma_sum_4 += chroma_pixels[i4[i]]; 248 chroma_sum_4y += chroma_pixels[i4y[i]]; 249 chroma_sum_4x += chroma_pixels[i4x[i]]; 250 } 251 252 float luma_avg_12 = (luma_sum_4 + luma_sum_4y + luma_sum_4x) / 12.0; 253 float luma_min_12 = luma_pixels[i12[0]]; 254 float luma_max_12 = luma_pixels[i12[0]]; 255 float luma_var_12 = 0.0; 256 vec2 chroma_avg_12 = (chroma_sum_4 + chroma_sum_4y + chroma_sum_4x) / 12.0; 257 vec2 chroma_var_12 = vec2(0.0); 258 vec2 luma_chroma_cov_12 = vec2(0.0); 259 260 for (int i = 0; i < 12; i++) { 261 luma_min_12 = min(luma_min_12, luma_pixels[i12[i]]); 262 luma_max_12 = max(luma_max_12, luma_pixels[i12[i]]); 263 luma_var_12 += pow(luma_pixels[i12[i]] - luma_avg_12, 2.0); 264 chroma_var_12 += pow(chroma_pixels[i12[i]] - chroma_avg_12, vec2(2.0)); 265 luma_chroma_cov_12 += (luma_pixels[i12[i]] - luma_avg_12) * (chroma_pixels[i12[i]] - chroma_avg_12); 266 } 267 268 vec2 corr = clamp(abs(luma_chroma_cov_12 / max(sqrt(luma_var_12 * chroma_var_12), 1e-6)), 0.0, 1.0); 269 mix_coeff = pow(corr, corr_exponent) * mix_coeff; 270 271 float luma_range_12 = max(luma_max_12 - luma_min_12, 1e-6); 272 float extrapolation = max(max(luma_zero - luma_max_12, luma_min_12 - luma_zero), 0.0) / luma_range_12; 273 mix_coeff *= clamp((8.0 - extrapolation) / 4.0, 0.0, 1.0); 274 #endif 275 276 #if (USE_12_TAP_REGRESSION == 1) 277 vec2 alpha_12 = luma_chroma_cov_12 / max(luma_var_12, 1e-6); 278 vec2 beta_12 = chroma_avg_12 - alpha_12 * luma_avg_12; 279 vec2 chroma_pred_12 = clamp(alpha_12 * luma_zero + beta_12, 0.0, 1.0); 280 #endif 281 282 #if (USE_8_TAP_REGRESSIONS == 1) 283 const int i8y[8] = {1, 2, 5, 6, 9, 10, 13, 14}; 284 const int i8x[8] = {4, 5, 6, 7, 8, 9, 10, 11}; 285 286 float luma_avg_8y = (luma_sum_4 + luma_sum_4y) / 8.0; 287 float luma_avg_8x = (luma_sum_4 + luma_sum_4x) / 8.0; 288 float luma_var_8y = 0.0; 289 float luma_var_8x = 0.0; 290 vec2 chroma_avg_8y = (chroma_sum_4 + chroma_sum_4y) / 8.0; 291 vec2 chroma_avg_8x = (chroma_sum_4 + chroma_sum_4x) / 8.0; 292 vec2 luma_chroma_cov_8y = vec2(0.0); 293 vec2 luma_chroma_cov_8x = vec2(0.0); 294 295 for (int i = 0; i < 8; i++) { 296 luma_var_8y += pow(luma_pixels[i8y[i]] - luma_avg_8y, 2.0); 297 luma_var_8x += pow(luma_pixels[i8x[i]] - luma_avg_8x, 2.0); 298 luma_chroma_cov_8y += (luma_pixels[i8y[i]] - luma_avg_8y) * (chroma_pixels[i8y[i]] - chroma_avg_8y); 299 luma_chroma_cov_8x += (luma_pixels[i8x[i]] - luma_avg_8x) * (chroma_pixels[i8x[i]] - chroma_avg_8x); 300 } 301 302 vec2 alpha_8y = luma_chroma_cov_8y / max(luma_var_8y, 1e-6); 303 vec2 alpha_8x = luma_chroma_cov_8x / max(luma_var_8x, 1e-6); 304 vec2 beta_8y = chroma_avg_8y - alpha_8y * luma_avg_8y; 305 vec2 beta_8x = chroma_avg_8x - alpha_8x * luma_avg_8x; 306 vec2 chroma_pred_8y = clamp(alpha_8y * luma_zero + beta_8y, 0.0, 1.0); 307 vec2 chroma_pred_8x = clamp(alpha_8x * luma_zero + beta_8x, 0.0, 1.0); 308 vec2 chroma_pred_8 = mix(chroma_pred_8y, chroma_pred_8x, 0.5); 309 #endif 310 311 #if (USE_12_TAP_REGRESSION == 1 && USE_8_TAP_REGRESSIONS == 1) 312 output_pix.xy = mix(chroma_spatial, mix(chroma_pred_12, chroma_pred_8, 0.5), mix_coeff); 313 #elif (USE_12_TAP_REGRESSION == 1 && USE_8_TAP_REGRESSIONS == 0) 314 output_pix.xy = mix(chroma_spatial, chroma_pred_12, mix_coeff); 315 #elif (USE_12_TAP_REGRESSION == 0 && USE_8_TAP_REGRESSIONS == 1) 316 output_pix.xy = mix(chroma_spatial, chroma_pred_8, mix_coeff); 317 #else 318 output_pix.xy = chroma_spatial; 319 #endif 320 321 output_pix.xy = clamp(output_pix.xy, 0.0, 1.0); 322 return output_pix; 323 } 324 325 //!PARAM distance_coeff 326 //!TYPE float 327 //!MINIMUM 0.0 328 2.0 329 330 //!PARAM intensity_coeff 331 //!TYPE float 332 //!MINIMUM 0.0 333 128.0 334 335 //!HOOK CHROMA 336 //!BIND CHROMA 337 //!BIND LUMA 338 //!BIND LUMA_LR 339 //!DESC Chroma From Luma Prediction (Smoothing Chroma) 340 341 float comp_w(vec2 spatial_distance, float intensity_distance) { 342 return max(100.0 * exp(-distance_coeff * pow(length(spatial_distance), 2.0) - intensity_coeff * pow(intensity_distance, 2.0)), 1e-32); 343 } 344 345 vec4 hook() { 346 vec4 output_pix = vec4(0.0, 0.0, 0.0, 1.0); 347 float luma_zero = LUMA_texOff(0).x; 348 float wt = 0.0; 349 vec2 ct = vec2(0.0); 350 351 for (int i = -1; i < 2; i++) { 352 for (int j = -1; j < 2; j++) { 353 vec2 chroma_pixels = CHROMA_texOff(vec2(i, j)).xy; 354 float luma_pixels = LUMA_texOff(vec2(i, j)).x; 355 float w = comp_w(vec2(i, j), luma_zero - luma_pixels); 356 wt += w; 357 ct += w * chroma_pixels; 358 } 359 } 360 361 output_pix.xy = clamp(ct / wt, 0.0, 1.0); 362 return output_pix; 363 }