git.sesse.net Git - ffmpeg/blob - libavfilter/vf_estdif.c

   1 /*
   2  * Copyright (c) 2021 Paul B Mahol
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/common.h"
  22 #include "libavutil/imgutils.h"
  23 #include "libavutil/opt.h"
  24 #include "libavutil/pixdesc.h"
  25 #include "avfilter.h"
  26 #include "formats.h"
  27 #include "internal.h"
  28 #include "video.h"
  29
  30 typedef struct ESTDIFContext {
  31     const AVClass *class;
  32
  33     int mode;             ///< 0 is frame, 1 is field
  34     int parity;           ///< frame field parity
  35     int deint;            ///< which frames to deinterlace
  36     int rslope;           ///< best edge slope search radius
  37     int redge;            ///< best edge match search radius
  38     int linesize[4];      ///< bytes of pixel data per line for each plane
  39     int planewidth[4];    ///< width of each plane
  40     int planeheight[4];   ///< height of each plane
  41     int field;            ///< which field are we on, 0 or 1
  42     int eof;
  43     int depth;
  44     int half;
  45     int nb_planes;
  46     int nb_threads;
  47     int64_t pts;
  48     AVFrame *prev;
  49
  50     void (*interpolate)(uint8_t *dst,
  51                         const uint8_t *prev_line, const uint8_t *next_line,
  52                         const uint8_t *prev2_line, const uint8_t *next2_line,
  53                         int x, int width, int rslope, int redge, unsigned half,
  54                         int depth, int *K);
  55 } ESTDIFContext;
  56
  57 #define MAX_R 15
  58 #define S     (MAX_R * 2 + 1)
  59
  60 #define OFFSET(x) offsetof(ESTDIFContext, x)
  61 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
  62 #define CONST(name, help, val, unit) { name, help, 0, AV_OPT_TYPE_CONST, {.i64=val}, 0, 0, FLAGS, unit }
  63
  64 static const AVOption estdif_options[] = {
  65     { "mode", "specify the mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "mode" },
  66     CONST("frame", "send one frame for each frame", 0, "mode"),
  67     CONST("field", "send one frame for each field", 1, "mode"),
  68     { "parity", "specify the assumed picture field parity", OFFSET(parity), AV_OPT_TYPE_INT, {.i64=-1}, -1, 1, FLAGS, "parity" },
  69     CONST("tff",  "assume top field first",    0, "parity"),
  70     CONST("bff",  "assume bottom field first", 1, "parity"),
  71     CONST("auto", "auto detect parity",       -1, "parity"),
  72     { "deint",  "specify which frames to deinterlace", OFFSET(deint), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS, "deint" },
  73     CONST("all",        "deinterlace all frames",                       0, "deint"),
  74     CONST("interlaced", "only deinterlace frames marked as interlaced", 1, "deint"),
  75     { "rslope", "specify the search radius for edge slope tracing", OFFSET(rslope), AV_OPT_TYPE_INT, {.i64=1}, 1, MAX_R, FLAGS, },
  76     { "redge",  "specify the search radius for best edge matching", OFFSET(redge),  AV_OPT_TYPE_INT, {.i64=2}, 0, MAX_R, FLAGS, },
  77     { NULL }
  78 };
  79
  80 AVFILTER_DEFINE_CLASS(estdif);
  81
  82 static int query_formats(AVFilterContext *ctx)
  83 {
  84     static const enum AVPixelFormat pix_fmts[] = {
  85         AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
  86         AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
  87         AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
  88         AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
  89         AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
  90         AV_PIX_FMT_YUVJ411P,
  91         AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
  92         AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
  93         AV_PIX_FMT_GRAY8,
  94         AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
  95         AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
  96         AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
  97         AV_PIX_FMT_YUV440P10,
  98         AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
  99         AV_PIX_FMT_YUV440P12,
 100         AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
 101         AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
 102         AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
 103         AV_PIX_FMT_YUVA444P9, AV_PIX_FMT_YUVA444P10, AV_PIX_FMT_YUVA444P12, AV_PIX_FMT_YUVA444P16,
 104         AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA422P16,
 105         AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA420P16,
 106         AV_PIX_FMT_GBRAP10,   AV_PIX_FMT_GBRAP12,    AV_PIX_FMT_GBRAP16,
 107         AV_PIX_FMT_NONE
 108     };
 109
 110     AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
 111     if (!fmts_list)
 112         return AVERROR(ENOMEM);
 113     return ff_set_common_formats(ctx, fmts_list);
 114 }
 115
 116 static int config_output(AVFilterLink *outlink)
 117 {
 118     AVFilterContext *ctx = outlink->src;
 119     AVFilterLink *inlink = ctx->inputs[0];
 120
 121     outlink->time_base.num = inlink->time_base.num;
 122     outlink->time_base.den = inlink->time_base.den * 2;
 123     outlink->frame_rate.num = inlink->frame_rate.num * 2;
 124     outlink->frame_rate.den = inlink->frame_rate.den;
 125
 126     return 0;
 127 }
 128
 129 typedef struct ThreadData {
 130     AVFrame *out, *in;
 131 } ThreadData;
 132
 133 #define MIDL(type, ss)                                         \
 134 static unsigned midl_##ss(const type *const prev,              \
 135                           const type *const next,              \
 136                           int end, int x, int k)               \
 137 {                                                              \
 138     return (prev[av_clip(x + k, 0, end)] +                     \
 139             next[av_clip(x - k, 0, end)] + 1) >> 1;            \
 140 }
 141
 142 MIDL(uint8_t, 8)
 143 MIDL(uint16_t, 16)
 144
 145 #define MID4(type, ss)                                         \
 146 static unsigned mid4_##ss(const type *const prev,              \
 147                           const type *const next,              \
 148                           const type *const prev2,             \
 149                           const type *const next2,             \
 150                           int end, int x, int k, int depth)    \
 151 {                                                              \
 152     return av_clip_uintp2_c((                                  \
 153             9 * (prev[av_clip(x + k, 0, end)] +                \
 154                  next[av_clip(x - k, 0, end)]) -               \
 155             1 * (prev2[av_clip(x + k*3, 0, end)] +             \
 156                  next2[av_clip(x - k*3, 0, end)]) + 8) >> 4,   \
 157                  depth);                                       \
 158 }
 159
 160 MID4(uint8_t, 8)
 161 MID4(uint16_t, 16)
 162
 163 #define DIFF(type, ss)                                         \
 164 static unsigned diff_##ss(const type *const prev,              \
 165                           const type *const next,              \
 166                           int end, int x, int k, int j)        \
 167 {                                                              \
 168     return FFABS(prev[av_clip(x + k + j, 0, end)] -            \
 169                  next[av_clip(x - k + j, 0, end)]);            \
 170 }
 171
 172 DIFF(uint8_t, 8)
 173 DIFF(uint16_t, 16)
 174
 175 #define COST(type, ss)                                         \
 176 static unsigned cost_##ss(const type *const prev,              \
 177                           const type *const next,              \
 178                           int end, int x, int k)               \
 179 {                                                              \
 180     const int m = midl_##ss(prev, next, end, x, k);            \
 181     const int p = prev[x];                                     \
 182     const int n = next[x];                                     \
 183                                                                \
 184     return FFABS(p - m) + FFABS(n - m);                        \
 185 }
 186
 187 COST(uint8_t, 8)
 188 COST(uint16_t, 16)
 189
 190 #define INTERPOLATE(type, atype, max, ss)                                      \
 191 static void interpolate_##ss(uint8_t *ddst,                                    \
 192                              const uint8_t *const pprev_line,                  \
 193                              const uint8_t *const nnext_line,                  \
 194                              const uint8_t *const pprev2_line,                 \
 195                              const uint8_t *const nnext2_line,                 \
 196                              int x, int width, int rslope,                     \
 197                              int redge, unsigned h, int depth,                 \
 198                              int *K)                                           \
 199 {                                                                              \
 200     type *dst = (type *)ddst;                                                  \
 201     const type *const prev_line = (const type *const)pprev_line;               \
 202     const type *const prev2_line = (const type *const)pprev2_line;             \
 203     const type *const next_line = (const type *const)nnext_line;               \
 204     const type *const next2_line = (const type *const)nnext2_line;             \
 205     const int end = width - 1;                                                 \
 206     const atype f = redge + 2;                                                 \
 207     atype sd[S], sD[S], di = 0;                                                \
 208     atype dmin = max;                                                          \
 209     int k = *K;                                                                \
 210                                                                                \
 211     for (int i = -rslope; i <= rslope && abs(k) > rslope; i++) {               \
 212         atype sum = 0;                                                         \
 213                                                                                \
 214         for (int j = -redge; j <= redge; j++) {                                \
 215             sum += diff_##ss(prev_line,  next_line,  end, x, i, j);            \
 216             sum += diff_##ss(prev2_line, prev_line,  end, x, i, j);            \
 217             sum += diff_##ss(next_line,  next2_line, end, x, i, j);            \
 218         }                                                                      \
 219                                                                                \
 220         sD[i + rslope]  =     sum;                                             \
 221         sD[i + rslope] += f * cost_##ss(prev_line,  next_line,  end, x, i);    \
 222         sD[i + rslope] += h * abs(i);                                          \
 223                                                                                \
 224         dmin = FFMIN(sD[i + rslope], dmin);                                    \
 225     }                                                                          \
 226                                                                                \
 227     for (int i = -rslope; i <= rslope; i++) {                                  \
 228         atype sum = 0;                                                         \
 229                                                                                \
 230         for (int j = -redge; j <= redge; j++) {                                \
 231             sum += diff_##ss(prev_line,  next_line,  end, x, k + i, j);        \
 232             sum += diff_##ss(prev2_line, prev_line,  end, x, k + i, j);        \
 233             sum += diff_##ss(next_line,  next2_line, end, x, k + i, j);        \
 234         }                                                                      \
 235                                                                                \
 236         sd[i + rslope]  =     sum;                                             \
 237         sd[i + rslope] += f * cost_##ss(prev_line, next_line, end, x, k + i);  \
 238         sd[i + rslope] += h * abs(k + i);                                      \
 239                                                                                \
 240         dmin = FFMIN(sd[i + rslope], dmin);                                    \
 241     }                                                                          \
 242                                                                                \
 243     for (int i = -rslope; i <= rslope && abs(k) > rslope; i++) {               \
 244         if (dmin == sD[i + rslope]) {                                          \
 245             di = 1;                                                            \
 246             k = i;                                                             \
 247             break;                                                             \
 248         }                                                                      \
 249     }                                                                          \
 250                                                                                \
 251     for (int i = -rslope; i <= rslope && !di; i++) {                           \
 252         if (dmin == sd[i + rslope]) {                                          \
 253             k += i;                                                            \
 254             break;                                                             \
 255         }                                                                      \
 256     }                                                                          \
 257                                                                                \
 258     dst[x] = mid4_##ss(prev_line, next_line, prev2_line, next2_line,           \
 259                        end, x, k, depth);                                      \
 260                                                                                \
 261     *K = k;                                                                    \
 262 }
 263
 264 INTERPOLATE(uint8_t,  unsigned, UINT_MAX, 8)
 265 INTERPOLATE(uint16_t, uint64_t, UINT64_MAX, 16)
 266
 267 static int deinterlace_slice(AVFilterContext *ctx, void *arg,
 268                              int jobnr, int nb_jobs)
 269 {
 270     ESTDIFContext *s = ctx->priv;
 271     ThreadData *td = arg;
 272     AVFrame *out = td->out;
 273     AVFrame *in = td->in;
 274     const int rslope = s->rslope;
 275     const int redge = s->redge;
 276     const int half = s->half;
 277     const int depth = s->depth;
 278     const int interlaced = in->interlaced_frame;
 279     const int tff = (s->field == (s->parity == -1 ? interlaced ? in->top_field_first : 1 :
 280                                   s->parity ^ 1));
 281
 282     for (int plane = 0; plane < s->nb_planes; plane++) {
 283         const uint8_t *src_data = in->data[plane];
 284         uint8_t *dst_data = out->data[plane];
 285         const int linesize = s->linesize[plane];
 286         const int width = s->planewidth[plane];
 287         const int height = s->planeheight[plane];
 288         const int src_linesize = in->linesize[plane];
 289         const int dst_linesize = out->linesize[plane];
 290         const int start = (height * jobnr) / nb_jobs;
 291         const int end = (height * (jobnr+1)) / nb_jobs;
 292         const uint8_t *prev_line, *prev2_line, *next_line, *next2_line, *in_line;
 293         uint8_t *out_line;
 294         int y_out;
 295
 296         y_out = start + (tff ^ (start & 1));
 297
 298         in_line  = src_data + (y_out * src_linesize);
 299         out_line = dst_data + (y_out * dst_linesize);
 300
 301         while (y_out < end) {
 302             memcpy(out_line, in_line, linesize);
 303             y_out += 2;
 304             in_line  += src_linesize * 2;
 305             out_line += dst_linesize * 2;
 306         }
 307
 308         y_out = start + ((!tff) ^ (start & 1));
 309         out_line = dst_data + (y_out * dst_linesize);
 310
 311         for (int y = y_out; y < end; y += 2) {
 312             int y_prev2_in = y - 3;
 313             int y_next2_in = y + 3;
 314             int y_prev_in = y - 1;
 315             int y_next_in = y + 1;
 316             int k;
 317
 318             while (y_prev2_in < 0)
 319                 y_prev2_in += 2;
 320
 321             while (y_next2_in >= height)
 322                 y_next2_in -= 2;
 323
 324             while (y_prev_in < 0)
 325                 y_prev_in += 2;
 326
 327             while (y_next_in >= height)
 328                 y_next_in -= 2;
 329
 330             prev2_line = src_data + (y_prev2_in * src_linesize);
 331             next2_line = src_data + (y_next2_in * src_linesize);
 332
 333             prev_line = src_data + (y_prev_in * src_linesize);
 334             next_line = src_data + (y_next_in * src_linesize);
 335
 336             k = 0;
 337
 338             for (int x = 0; x < width; x++) {
 339                 s->interpolate(out_line,
 340                                prev_line, next_line,
 341                                prev2_line, next2_line,
 342                                x, width, rslope, redge, half, depth, &k);
 343             }
 344
 345             out_line += 2 * dst_linesize;
 346         }
 347     }
 348
 349     return 0;
 350 }
 351
 352 static int filter(AVFilterContext *ctx, int is_second, AVFrame *in)
 353 {
 354     ESTDIFContext *s = ctx->priv;
 355     AVFilterLink *outlink = ctx->outputs[0];
 356     AVFrame *out;
 357     ThreadData td;
 358
 359     out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
 360     if (!out)
 361         return AVERROR(ENOMEM);
 362     av_frame_copy_props(out, in);
 363     out->interlaced_frame = 0;
 364     out->pts = s->pts;
 365
 366     td.out = out; td.in = in;
 367     ctx->internal->execute(ctx, deinterlace_slice, &td, NULL,
 368                            FFMIN(s->planeheight[1] / 2, s->nb_threads));
 369
 370     if (s->mode)
 371         s->field = !s->field;
 372
 373     return ff_filter_frame(outlink, out);
 374 }
 375
 376 static int config_input(AVFilterLink *inlink)
 377 {
 378     AVFilterContext *ctx = inlink->dst;
 379     ESTDIFContext *s = ctx->priv;
 380     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
 381     int ret;
 382
 383     if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
 384         return ret;
 385
 386     s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
 387     s->planeheight[0] = s->planeheight[3] = inlink->h;
 388     s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
 389     s->planewidth[0] = s->planewidth[3] = inlink->w;
 390
 391     if (inlink->h < 3) {
 392         av_log(ctx, AV_LOG_ERROR, "Video of less than 3 lines is not supported\n");
 393         return AVERROR(EINVAL);
 394     }
 395
 396     s->nb_planes = av_pix_fmt_count_planes(inlink->format);
 397     s->nb_threads = ff_filter_get_nb_threads(ctx);
 398     s->depth = desc->comp[0].depth;
 399     s->interpolate = s->depth <= 8 ? interpolate_8 : interpolate_16;
 400     s->half = 1 << (s->depth - 1);
 401
 402     return 0;
 403 }
 404  static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 405 {
 406     AVFilterContext *ctx = inlink->dst;
 407     ESTDIFContext *s = ctx->priv;
 408     int ret;
 409
 410     if (!s->prev) {
 411         s->prev = in;
 412         return 0;
 413     }
 414
 415     if ((s->deint && !in->interlaced_frame) || ctx->is_disabled) {
 416         s->prev->pts *= 2;
 417         ret = ff_filter_frame(ctx->outputs[0], s->prev);
 418         s->prev = in;
 419         return ret;
 420     }
 421
 422     s->pts = s->prev->pts * 2;
 423     ret = filter(ctx, 0, s->prev);
 424     if (ret < 0 || s->mode == 0) {
 425         av_frame_free(&s->prev);
 426         s->prev = in;
 427         return ret;
 428     }
 429
 430     s->pts = s->prev->pts + in->pts;
 431     ret = filter(ctx, 1, s->prev);
 432     av_frame_free(&s->prev);
 433     s->prev = in;
 434     return ret;
 435 }
 436
 437 static int request_frame(AVFilterLink *link)
 438 {
 439     AVFilterContext *ctx = link->src;
 440     ESTDIFContext *s = ctx->priv;
 441     int ret;
 442
 443     if (s->eof)
 444         return AVERROR_EOF;
 445
 446     ret = ff_request_frame(ctx->inputs[0]);
 447
 448     if (ret == AVERROR_EOF && s->prev) {
 449         AVFrame *next = av_frame_clone(s->prev);
 450
 451         if (!next)
 452             return AVERROR(ENOMEM);
 453
 454         next->pts = s->prev->pts + av_rescale_q(1, av_inv_q(ctx->outputs[0]->frame_rate),
 455                                                 ctx->outputs[0]->time_base);
 456         s->eof = 1;
 457         ret = filter_frame(ctx->inputs[0], next);
 458     } else if (ret < 0) {
 459         return ret;
 460     }
 461
 462     return ret;
 463 }
 464
 465 static av_cold void uninit(AVFilterContext *ctx)
 466 {
 467     ESTDIFContext *s = ctx->priv;
 468
 469     av_frame_free(&s->prev);
 470 }
 471
 472 static const AVFilterPad estdif_inputs[] = {
 473     {
 474         .name          = "default",
 475         .type          = AVMEDIA_TYPE_VIDEO,
 476         .filter_frame  = filter_frame,
 477         .config_props  = config_input,
 478     },
 479     { NULL }
 480 };
 481
 482 static const AVFilterPad estdif_outputs[] = {
 483     {
 484         .name          = "default",
 485         .type          = AVMEDIA_TYPE_VIDEO,
 486         .config_props  = config_output,
 487         .request_frame = request_frame,
 488     },
 489     { NULL }
 490 };
 491
 492 AVFilter ff_vf_estdif = {
 493     .name          = "estdif",
 494     .description   = NULL_IF_CONFIG_SMALL("Apply Edge Slope Tracing deinterlace."),
 495     .priv_size     = sizeof(ESTDIFContext),
 496     .priv_class    = &estdif_class,
 497     .uninit        = uninit,
 498     .query_formats = query_formats,
 499     .inputs        = estdif_inputs,
 500     .outputs       = estdif_outputs,
 501     .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
 502     .process_command = ff_filter_process_command,
 503 };