| @@ -0,0 +1,349 @@ |
| 1 | +/* |
| 2 | + * Copyright (c) 2021 Boris Baracaldo |
| 3 | + * Copyright (c) 2022 Thilo Borgmann |
| 4 | + * |
| 5 | + * This file is part of FFmpeg. |
| 6 | + * |
| 7 | + * FFmpeg is free software; you can redistribute it and/or modify |
| 8 | + * it under the terms of the GNU General Public License as published by |
| 9 | + * the Free Software Foundation; either version 2 of the License, or |
| 10 | + * (at your option) any later version. |
| 11 | + * |
| 12 | + * FFmpeg is distributed in the hope that it will be useful, |
| 13 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 15 | + * GNU General Public License for more details. |
| 16 | + * |
| 17 | + * You should have received a copy of the GNU General Public License along |
| 18 | + * with FFmpeg; if not, write to the Free Software Foundation, Inc., |
| 19 | + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 20 | + */ |
| 21 | + |
| 22 | +/** |
| 23 | + * @file |
| 24 | + * Calculate Spatial Info (SI) and Temporal Info (TI) scores |
| 25 | + */ |
| 26 | + |
| 27 | +#include <math.h> |
| 28 | + |
| 29 | +#include "libavutil/imgutils.h" |
| 30 | +#include "libavutil/internal.h" |
| 31 | +#include "libavutil/opt.h" |
| 32 | + |
| 33 | +#include "avfilter.h" |
| 34 | +#include "formats.h" |
| 35 | +#include "internal.h" |
| 36 | +#include "video.h" |
| 37 | + |
| 38 | +static const int X_FILTER[9] = { |
| 39 | + 1, 0, -1, |
| 40 | + 2, 0, -2, |
| 41 | + 1, 0, -1 |
| 42 | +}; |
| 43 | + |
| 44 | +static const int Y_FILTER[9] = { |
| 45 | + 1, 2, 1, |
| 46 | + 0, 0, 0, |
| 47 | + -1, -2, -1 |
| 48 | +}; |
| 49 | + |
| 50 | +typedef struct SiTiContext { |
| 51 | + const AVClass *class; |
| 52 | + int pixel_depth; |
| 53 | + int width, height; |
| 54 | + uint64_t nb_frames; |
| 55 | + uint8_t *prev_frame; |
| 56 | + float max_si; |
| 57 | + float max_ti; |
| 58 | + float min_si; |
| 59 | + float min_ti; |
| 60 | + float sum_si; |
| 61 | + float sum_ti; |
| 62 | + float *gradient_matrix; |
| 63 | + float *motion_matrix; |
| 64 | + int full_range; |
| 65 | + int print_summary; |
| 66 | +} SiTiContext; |
| 67 | + |
| 68 | +static const enum AVPixelFormat pix_fmts[] = { |
| 69 | + AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, |
| 70 | + AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P, |
| 71 | + AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, |
| 72 | + AV_PIX_FMT_NONE |
| 73 | +}; |
| 74 | + |
| 75 | +static av_cold int init(AVFilterContext *ctx) |
| 76 | +{ |
| 77 | + // User options but no input data |
| 78 | + SiTiContext *s = ctx->priv; |
| 79 | + s->max_si = 0; |
| 80 | + s->max_ti = 0; |
| 81 | + return 0; |
| 82 | +} |
| 83 | + |
| 84 | +static av_cold void uninit(AVFilterContext *ctx) |
| 85 | +{ |
| 86 | + SiTiContext *s = ctx->priv; |
| 87 | + |
| 88 | + if (s->print_summary) { |
| 89 | + float avg_si = s->sum_si / s->nb_frames; |
| 90 | + float avg_ti = s->sum_ti / s->nb_frames; |
| 91 | + av_log(ctx, AV_LOG_INFO, |
| 92 | + "SITI Summary:\nTotal frames: %"PRId64"\n\n" |
| 93 | + "Spatial Information:\nAverage: %f\nMax: %f\nMin: %f\n\n" |
| 94 | + "Temporal Information:\nAverage: %f\nMax: %f\nMin: %f\n", |
| 95 | + s->nb_frames, avg_si, s->max_si, s->min_si, avg_ti, s->max_ti, s->min_ti |
| 96 | + ); |
| 97 | + } |
| 98 | + |
| 99 | + av_freep(&s->prev_frame); |
| 100 | + av_freep(&s->gradient_matrix); |
| 101 | + av_freep(&s->motion_matrix); |
| 102 | +} |
| 103 | + |
| 104 | +static int config_input(AVFilterLink *inlink) |
| 105 | +{ |
| 106 | + // Video input data avilable |
| 107 | + AVFilterContext *ctx = inlink->dst; |
| 108 | + SiTiContext *s = ctx->priv; |
| 109 | + int max_pixsteps[4]; |
| 110 | + size_t pixel_sz; |
| 111 | + size_t data_sz; |
| 112 | + size_t gradient_sz; |
| 113 | + size_t motion_sz; |
| 114 | + |
| 115 | + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); |
| 116 | + av_image_fill_max_pixsteps(max_pixsteps, NULL, desc); |
| 117 | + |
| 118 | + // free previous buffers in case they are allocated already |
| 119 | + av_freep(&s->prev_frame); |
| 120 | + av_freep(&s->gradient_matrix); |
| 121 | + av_freep(&s->motion_matrix); |
| 122 | + |
| 123 | + s->pixel_depth = max_pixsteps[0]; |
| 124 | + s->width = inlink->w; |
| 125 | + s->height = inlink->h; |
| 126 | + pixel_sz = s->pixel_depth == 1 ? sizeof(uint8_t) : sizeof(uint16_t); |
| 127 | + data_sz = s->width * pixel_sz * s->height; |
| 128 | + |
| 129 | + s->prev_frame = av_malloc(data_sz); |
| 130 | + |
| 131 | + gradient_sz = (s->width - 2) * sizeof(float) * (s->height - 2); |
| 132 | + s->gradient_matrix = av_malloc(gradient_sz); |
| 133 | + |
| 134 | + motion_sz = s->width * sizeof(float) * s->height; |
| 135 | + s->motion_matrix = av_malloc(motion_sz); |
| 136 | + |
| 137 | + if (!s->prev_frame || ! s->gradient_matrix || !s->motion_matrix) { |
| 138 | + return AVERROR(ENOMEM); |
| 139 | + } |
| 140 | + |
| 141 | + return 0; |
| 142 | +} |
| 143 | + |
| 144 | +// Determine whether the video is in full or limited range. If not defined, assume limited. |
| 145 | +static int is_full_range(AVFrame* frame) |
| 146 | +{ |
| 147 | + // If color range not specified, fallback to pixel format |
| 148 | + if (frame->color_range == AVCOL_RANGE_UNSPECIFIED || frame->color_range == AVCOL_RANGE_NB) |
| 149 | + return frame->format == AV_PIX_FMT_YUVJ420P || frame->format == AV_PIX_FMT_YUVJ422P; |
| 150 | + return frame->color_range == AVCOL_RANGE_JPEG; |
| 151 | +} |
| 152 | + |
| 153 | +// Check frame's color range and convert to full range if needed |
| 154 | +static uint16_t convert_full_range(int factor, uint16_t y) |
| 155 | +{ |
| 156 | + int shift; |
| 157 | + int limit_upper; |
| 158 | + int full_upper; |
| 159 | + int limit_y; |
| 160 | + |
| 161 | + // For 8 bits, limited range goes from 16 to 235, for 10 bits the range is multiplied by 4 |
| 162 | + shift = 16 * factor; |
| 163 | + limit_upper = 235 * factor - shift; |
| 164 | + full_upper = 256 * factor - 1; |
| 165 | + limit_y = fminf(fmaxf(y - shift, 0), limit_upper); |
| 166 | + return (full_upper * limit_y / limit_upper); |
| 167 | +} |
| 168 | + |
| 169 | +// Applies sobel convolution |
| 170 | +static void convolve_sobel(SiTiContext *s, const uint8_t *src, float *dst, int linesize) |
| 171 | +{ |
| 172 | + double x_conv_sum; |
| 173 | + double y_conv_sum; |
| 174 | + float gradient; |
| 175 | + int ki; |
| 176 | + int kj; |
| 177 | + int index; |
| 178 | + uint16_t data; |
| 179 | + int filter_width = 3; |
| 180 | + int filter_size = filter_width * filter_width; |
| 181 | + int stride = linesize / s->pixel_depth; |
| 182 | + // For 8 bits, limited range goes from 16 to 235, for 10 bits the range is multiplied by 4 |
| 183 | + int factor = s->pixel_depth == 1 ? 1 : 4; |
| 184 | + |
| 185 | + // Dst matrix is smaller than src since we ignore edges that can't be convolved |
| 186 | + #define CONVOLVE(bps) \ |
| 187 | + { \ |
| 188 | + uint##bps##_t *vsrc = (uint##bps##_t*)src; \ |
| 189 | + for (int j = 1; j < s->height - 1; j++) { \ |
| 190 | + for (int i = 1; i < s->width - 1; i++) { \ |
| 191 | + x_conv_sum = 0.0; \ |
| 192 | + y_conv_sum = 0.0; \ |
| 193 | + for (int k = 0; k < filter_size; k++) { \ |
| 194 | + ki = k % filter_width - 1; \ |
| 195 | + kj = floor(k / filter_width) - 1; \ |
| 196 | + index = (j + kj) * stride + (i + ki); \ |
| 197 | + data = s->full_range ? vsrc[index] : convert_full_range(factor, vsrc[index]); \ |
| 198 | + x_conv_sum += data * X_FILTER[k]; \ |
| 199 | + y_conv_sum += data * Y_FILTER[k]; \ |
| 200 | + } \ |
| 201 | + gradient = sqrt(x_conv_sum * x_conv_sum + y_conv_sum * y_conv_sum); \ |
| 202 | + dst[(j - 1) * (s->width - 2) + (i - 1)] = gradient; \ |
| 203 | + } \ |
| 204 | + } \ |
| 205 | + } |
| 206 | + |
| 207 | + if (s->pixel_depth == 2) { |
| 208 | + CONVOLVE(16); |
| 209 | + } else { |
| 210 | + CONVOLVE(8); |
| 211 | + } |
| 212 | +} |
| 213 | + |
| 214 | +// Calculate pixel difference between current and previous frame, and update previous |
| 215 | +static void calculate_motion(SiTiContext *s, const uint8_t *curr, |
| 216 | + float *motion_matrix, int linesize) |
| 217 | +{ |
| 218 | + int stride = linesize / s->pixel_depth; |
| 219 | + float motion; |
| 220 | + int curr_index; |
| 221 | + int prev_index; |
| 222 | + uint16_t curr_data; |
| 223 | + // For 8 bits, limited range goes from 16 to 235, for 10 bits the range is multiplied by 4 |
| 224 | + int factor = s->pixel_depth == 1 ? 1 : 4; |
| 225 | + |
| 226 | + // Previous frame is already converted to full range |
| 227 | + #define CALCULATE(bps) \ |
| 228 | + { \ |
| 229 | + uint##bps##_t *vsrc = (uint##bps##_t*)curr; \ |
| 230 | + uint##bps##_t *vdst = (uint##bps##_t*)s->prev_frame; \ |
| 231 | + for (int j = 0; j < s->height; j++) { \ |
| 232 | + for (int i = 0; i < s->width; i++) { \ |
| 233 | + motion = 0; \ |
| 234 | + curr_index = j * stride + i; \ |
| 235 | + prev_index = j * s->width + i; \ |
| 236 | + curr_data = s->full_range ? vsrc[curr_index] : convert_full_range(factor, vsrc[curr_index]); \ |
| 237 | + if (s->nb_frames > 1) \ |
| 238 | + motion = curr_data - vdst[prev_index]; \ |
| 239 | + vdst[prev_index] = curr_data; \ |
| 240 | + motion_matrix[j * s->width + i] = motion; \ |
| 241 | + } \ |
| 242 | + } \ |
| 243 | + } |
| 244 | + |
| 245 | + if (s->pixel_depth == 2) { |
| 246 | + CALCULATE(16); |
| 247 | + } else { |
| 248 | + CALCULATE(8); |
| 249 | + } |
| 250 | +} |
| 251 | + |
| 252 | +static float std_deviation(float *img_metrics, int width, int height) |
| 253 | +{ |
| 254 | + int size = height * width; |
| 255 | + double mean = 0.0; |
| 256 | + double sqr_diff = 0; |
| 257 | + |
| 258 | + for (int j = 0; j < height; j++) |
| 259 | + for (int i = 0; i < width; i++) |
| 260 | + mean += img_metrics[j * width + i]; |
| 261 | + |
| 262 | + mean /= size; |
| 263 | + |
| 264 | + for (int j = 0; j < height; j++) { |
| 265 | + for (int i = 0; i < width; i++) { |
| 266 | + float mean_diff = img_metrics[j * width + i] - mean; |
| 267 | + sqr_diff += (mean_diff * mean_diff); |
| 268 | + } |
| 269 | + } |
| 270 | + sqr_diff = sqr_diff / size; |
| 271 | + return sqrt(sqr_diff); |
| 272 | +} |
| 273 | + |
| 274 | +static void set_meta(AVDictionary **metadata, const char *key, float d) |
| 275 | +{ |
| 276 | + char value[128]; |
| 277 | + snprintf(value, sizeof(value), "%0.2f", d); |
| 278 | + av_dict_set(metadata, key, value, 0); |
| 279 | +} |
| 280 | + |
| 281 | +static int filter_frame(AVFilterLink *inlink, AVFrame *frame) |
| 282 | +{ |
| 283 | + AVFilterContext *ctx = inlink->dst; |
| 284 | + SiTiContext *s = ctx->priv; |
| 285 | + float si; |
| 286 | + float ti; |
| 287 | + |
| 288 | + s->full_range = is_full_range(frame); |
| 289 | + s->nb_frames++; |
| 290 | + |
| 291 | + // Calculate si and ti |
| 292 | + convolve_sobel(s, frame->data[0], s->gradient_matrix, frame->linesize[0]); |
| 293 | + calculate_motion(s, frame->data[0], s->motion_matrix, frame->linesize[0]); |
| 294 | + si = std_deviation(s->gradient_matrix, s->width - 2, s->height - 2); |
| 295 | + ti = std_deviation(s->motion_matrix, s->width, s->height); |
| 296 | + |
| 297 | + // Calculate statistics |
| 298 | + s->max_si = fmaxf(si, s->max_si); |
| 299 | + s->max_ti = fmaxf(ti, s->max_ti); |
| 300 | + s->sum_si += si; |
| 301 | + s->sum_ti += ti; |
| 302 | + s->min_si = s->nb_frames == 1 ? si : fminf(si, s->min_si); |
| 303 | + s->min_ti = s->nb_frames == 1 ? ti : fminf(ti, s->min_ti); |
| 304 | + |
| 305 | + // Set si ti information in frame metadata |
| 306 | + set_meta(&frame->metadata, "lavfi.siti.si", si); |
| 307 | + set_meta(&frame->metadata, "lavfi.siti.ti", ti); |
| 308 | + |
| 309 | + return ff_filter_frame(inlink->dst->outputs[0], frame); |
| 310 | +} |
| 311 | + |
| 312 | +#define OFFSET(x) offsetof(SiTiContext, x) |
| 313 | +#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM |
| 314 | + |
| 315 | +static const AVOption siti_options[] = { |
| 316 | + { "print_summary", "Print summary showing average values", OFFSET(print_summary), AV_OPT_TYPE_BOOL, { .i64=0 }, 0, 1, FLAGS }, |
| 317 | + { NULL } |
| 318 | +}; |
| 319 | + |
| 320 | +AVFILTER_DEFINE_CLASS(siti); |
| 321 | + |
| 322 | +static const AVFilterPad avfilter_vf_siti_inputs[] = { |
| 323 | + { |
| 324 | + .name = "default", |
| 325 | + .type = AVMEDIA_TYPE_VIDEO, |
| 326 | + .config_props = config_input, |
| 327 | + .filter_frame = filter_frame, |
| 328 | + }, |
| 329 | +}; |
| 330 | + |
| 331 | +static const AVFilterPad avfilter_vf_siti_outputs[] = { |
| 332 | + { |
| 333 | + .name = "default", |
| 334 | + .type = AVMEDIA_TYPE_VIDEO |
| 335 | + }, |
| 336 | +}; |
| 337 | + |
| 338 | +AVFilter ff_vf_siti = { |
| 339 | + .name = "siti", |
| 340 | + .description = NULL_IF_CONFIG_SMALL("Calculate spatial information (SI) and temporal information (TI)."), |
| 341 | + .priv_size = sizeof(SiTiContext), |
| 342 | + .priv_class = &siti_class, |
| 343 | + .init = init, |
| 344 | + .uninit = uninit, |
| 345 | + .flags = AVFILTER_FLAG_METADATA_ONLY, |
| 346 | + FILTER_PIXFMTS_ARRAY(pix_fmts), |
| 347 | + FILTER_INPUTS(avfilter_vf_siti_inputs), |
| 348 | + FILTER_OUTPUTS(avfilter_vf_siti_outputs), |
| 349 | +}; |