FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_decimate.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2012 Fredrik Mellbin
3  * Copyright (c) 2013 Clément Bœsch
4  *
5  * This file is part of FFmpeg.
6  *
7  * FFmpeg is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (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 GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with FFmpeg; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "libavutil/opt.h"
23 #include "libavutil/pixdesc.h"
24 #include "libavutil/timestamp.h"
25 #include "avfilter.h"
26 #include "internal.h"
27 
28 #define INPUT_MAIN 0
29 #define INPUT_CLEANSRC 1
30 
31 struct qitem {
33  int64_t maxbdiff;
34  int64_t totdiff;
35 };
36 
37 typedef struct {
38  const AVClass *class;
39  struct qitem *queue; ///< window of cycle frames and the associated data diff
40  int fid; ///< current frame id in the queue
41  int filled; ///< 1 if the queue is filled, 0 otherwise
42  AVFrame *last; ///< last frame from the previous queue
43  AVFrame **clean_src; ///< frame queue for the clean source
44  int got_frame[2]; ///< frame request flag for each input stream
45  double ts_unit; ///< timestamp units for the output frames
46  uint32_t eof; ///< bitmask for end of stream
47  int hsub, vsub; ///< chroma subsampling values
48  int depth;
49  int nxblocks, nyblocks;
50  int bdiffsize;
51  int64_t *bdiffs;
52 
53  /* options */
54  int cycle;
55  double dupthresh_flt;
56  double scthresh_flt;
57  int64_t dupthresh;
58  int64_t scthresh;
59  int blockx, blocky;
60  int ppsrc;
61  int chroma;
63 
64 #define OFFSET(x) offsetof(DecimateContext, x)
65 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
66 
67 static const AVOption decimate_options[] = {
68  { "cycle", "set the number of frame from which one will be dropped", OFFSET(cycle), AV_OPT_TYPE_INT, {.i64 = 5}, 2, 25, FLAGS },
69  { "dupthresh", "set duplicate threshold", OFFSET(dupthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 1.1}, 0, 100, FLAGS },
70  { "scthresh", "set scene change threshold", OFFSET(scthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 15.0}, 0, 100, FLAGS },
71  { "blockx", "set the size of the x-axis blocks used during metric calculations", OFFSET(blockx), AV_OPT_TYPE_INT, {.i64 = 32}, 4, 1<<9, FLAGS },
72  { "blocky", "set the size of the y-axis blocks used during metric calculations", OFFSET(blocky), AV_OPT_TYPE_INT, {.i64 = 32}, 4, 1<<9, FLAGS },
73  { "ppsrc", "mark main input as a pre-processed input and activate clean source input stream", OFFSET(ppsrc), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS },
74  { "chroma", "set whether or not chroma is considered in the metric calculations", OFFSET(chroma), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS },
75  { NULL }
76 };
77 
78 AVFILTER_DEFINE_CLASS(decimate);
79 
80 static void calc_diffs(const DecimateContext *dm, struct qitem *q,
81  const AVFrame *f1, const AVFrame *f2)
82 {
83  int64_t maxdiff = -1;
84  int64_t *bdiffs = dm->bdiffs;
85  int plane, i, j;
86 
87  memset(bdiffs, 0, dm->bdiffsize * sizeof(*bdiffs));
88 
89  for (plane = 0; plane < (dm->chroma ? 3 : 1); plane++) {
90  int x, y, xl;
91  const int linesize1 = f1->linesize[plane];
92  const int linesize2 = f2->linesize[plane];
93  const uint8_t *f1p = f1->data[plane];
94  const uint8_t *f2p = f2->data[plane];
95  int width = plane ? FF_CEIL_RSHIFT(f1->width, dm->hsub) : f1->width;
96  int height = plane ? FF_CEIL_RSHIFT(f1->height, dm->vsub) : f1->height;
97  int hblockx = dm->blockx / 2;
98  int hblocky = dm->blocky / 2;
99 
100  if (plane) {
101  hblockx >>= dm->hsub;
102  hblocky >>= dm->vsub;
103  }
104 
105  for (y = 0; y < height; y++) {
106  int ydest = y / hblocky;
107  int xdest = 0;
108 
109 #define CALC_DIFF(nbits) do { \
110  for (x = 0; x < width; x += hblockx) { \
111  int64_t acc = 0; \
112  int m = FFMIN(width, x + hblockx); \
113  for (xl = x; xl < m; xl++) \
114  acc += abs(((const uint##nbits##_t *)f1p)[xl] - \
115  ((const uint##nbits##_t *)f2p)[xl]); \
116  bdiffs[ydest * dm->nxblocks + xdest] += acc; \
117  xdest++; \
118  } \
119 } while (0)
120  if (dm->depth == 8) CALC_DIFF(8);
121  else CALC_DIFF(16);
122 
123  f1p += linesize1;
124  f2p += linesize2;
125  }
126  }
127 
128  for (i = 0; i < dm->nyblocks - 1; i++) {
129  for (j = 0; j < dm->nxblocks - 1; j++) {
130  int64_t tmp = bdiffs[ i * dm->nxblocks + j ]
131  + bdiffs[ i * dm->nxblocks + j + 1]
132  + bdiffs[(i + 1) * dm->nxblocks + j ]
133  + bdiffs[(i + 1) * dm->nxblocks + j + 1];
134  if (tmp > maxdiff)
135  maxdiff = tmp;
136  }
137  }
138 
139  q->totdiff = 0;
140  for (i = 0; i < dm->bdiffsize; i++)
141  q->totdiff += bdiffs[i];
142  q->maxbdiff = maxdiff;
143 }
144 
145 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
146 {
147  int scpos = -1, duppos = -1;
148  int drop = INT_MIN, i, lowest = 0, ret;
149  AVFilterContext *ctx = inlink->dst;
150  AVFilterLink *outlink = ctx->outputs[0];
151  DecimateContext *dm = ctx->priv;
152  AVFrame *prv;
153 
154  /* update frames queue(s) */
155  if (FF_INLINK_IDX(inlink) == INPUT_MAIN) {
156  dm->queue[dm->fid].frame = in;
157  dm->got_frame[INPUT_MAIN] = 1;
158  } else {
159  dm->clean_src[dm->fid] = in;
160  dm->got_frame[INPUT_CLEANSRC] = 1;
161  }
162  if (!dm->got_frame[INPUT_MAIN] || (dm->ppsrc && !dm->got_frame[INPUT_CLEANSRC]))
163  return 0;
165 
166  if (in) {
167  /* update frame metrics */
168  prv = dm->fid ? dm->queue[dm->fid - 1].frame : dm->last;
169  if (!prv)
170  prv = in;
171  calc_diffs(dm, &dm->queue[dm->fid], prv, in);
172  if (++dm->fid != dm->cycle)
173  return 0;
174  av_frame_free(&dm->last);
175  dm->last = av_frame_clone(in);
176  dm->fid = 0;
177 
178  /* we have a complete cycle, select the frame to drop */
179  lowest = 0;
180  for (i = 0; i < dm->cycle; i++) {
181  if (dm->queue[i].totdiff > dm->scthresh)
182  scpos = i;
183  if (dm->queue[i].maxbdiff < dm->queue[lowest].maxbdiff)
184  lowest = i;
185  }
186  if (dm->queue[lowest].maxbdiff < dm->dupthresh)
187  duppos = lowest;
188  drop = scpos >= 0 && duppos < 0 ? scpos : lowest;
189  }
190 
191  /* metrics debug */
192  if (av_log_get_level() >= AV_LOG_DEBUG) {
193  av_log(ctx, AV_LOG_DEBUG, "1/%d frame drop:\n", dm->cycle);
194  for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) {
195  av_log(ctx, AV_LOG_DEBUG," #%d: totdiff=%08"PRIx64" maxbdiff=%08"PRIx64"%s%s%s%s\n",
196  i + 1, dm->queue[i].totdiff, dm->queue[i].maxbdiff,
197  i == scpos ? " sc" : "",
198  i == duppos ? " dup" : "",
199  i == lowest ? " lowest" : "",
200  i == drop ? " [DROP]" : "");
201  }
202  }
203 
204  /* push all frames except the drop */
205  ret = 0;
206  for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) {
207  if (i == drop) {
208  if (dm->ppsrc)
209  av_frame_free(&dm->clean_src[i]);
210  av_frame_free(&dm->queue[i].frame);
211  } else {
212  AVFrame *frame = dm->queue[i].frame;
213  if (dm->ppsrc) {
214  av_frame_free(&frame);
215  frame = dm->clean_src[i];
216  }
217  frame->pts = outlink->frame_count * dm->ts_unit;
218  ret = ff_filter_frame(outlink, frame);
219  if (ret < 0)
220  break;
221  }
222  }
223 
224  return ret;
225 }
226 
227 static int config_input(AVFilterLink *inlink)
228 {
229  int max_value;
230  AVFilterContext *ctx = inlink->dst;
231  DecimateContext *dm = ctx->priv;
232  const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format);
233  const int w = inlink->w;
234  const int h = inlink->h;
235 
236  dm->hsub = pix_desc->log2_chroma_w;
237  dm->vsub = pix_desc->log2_chroma_h;
238  dm->depth = pix_desc->comp[0].depth_minus1 + 1;
239  max_value = (1 << dm->depth) - 1;
240  dm->scthresh = (int64_t)(((int64_t)max_value * w * h * dm->scthresh_flt) / 100);
241  dm->dupthresh = (int64_t)(((int64_t)max_value * dm->blockx * dm->blocky * dm->dupthresh_flt) / 100);
242  dm->nxblocks = (w + dm->blockx/2 - 1) / (dm->blockx/2);
243  dm->nyblocks = (h + dm->blocky/2 - 1) / (dm->blocky/2);
244  dm->bdiffsize = dm->nxblocks * dm->nyblocks;
245  dm->bdiffs = av_malloc(dm->bdiffsize * sizeof(*dm->bdiffs));
246  dm->queue = av_calloc(dm->cycle, sizeof(*dm->queue));
247 
248  if (!dm->bdiffs || !dm->queue)
249  return AVERROR(ENOMEM);
250 
251  if (dm->ppsrc) {
252  dm->clean_src = av_calloc(dm->cycle, sizeof(*dm->clean_src));
253  if (!dm->clean_src)
254  return AVERROR(ENOMEM);
255  }
256 
257  return 0;
258 }
259 
261 {
262  const DecimateContext *dm = ctx->priv;
263  AVFilterPad pad = {
264  .name = av_strdup("main"),
265  .type = AVMEDIA_TYPE_VIDEO,
266  .filter_frame = filter_frame,
267  .config_props = config_input,
268  };
269 
270  if (!pad.name)
271  return AVERROR(ENOMEM);
272  ff_insert_inpad(ctx, INPUT_MAIN, &pad);
273 
274  if (dm->ppsrc) {
275  pad.name = av_strdup("clean_src");
276  pad.config_props = NULL;
277  if (!pad.name)
278  return AVERROR(ENOMEM);
279  ff_insert_inpad(ctx, INPUT_CLEANSRC, &pad);
280  }
281 
282  if ((dm->blockx & (dm->blockx - 1)) ||
283  (dm->blocky & (dm->blocky - 1))) {
284  av_log(ctx, AV_LOG_ERROR, "blockx and blocky settings must be power of two\n");
285  return AVERROR(EINVAL);
286  }
287 
288  return 0;
289 }
290 
292 {
293  int i;
294  DecimateContext *dm = ctx->priv;
295 
296  av_frame_free(&dm->last);
297  av_freep(&dm->bdiffs);
298  av_freep(&dm->queue);
299  av_freep(&dm->clean_src);
300  for (i = 0; i < ctx->nb_inputs; i++)
301  av_freep(&ctx->input_pads[i].name);
302 }
303 
304 static int request_inlink(AVFilterContext *ctx, int lid)
305 {
306  int ret = 0;
307  DecimateContext *dm = ctx->priv;
308 
309  if (!dm->got_frame[lid]) {
310  AVFilterLink *inlink = ctx->inputs[lid];
311  ret = ff_request_frame(inlink);
312  if (ret == AVERROR_EOF) { // flushing
313  dm->eof |= 1 << lid;
314  ret = filter_frame(inlink, NULL);
315  }
316  }
317  return ret;
318 }
319 
320 static int request_frame(AVFilterLink *outlink)
321 {
322  int ret;
323  AVFilterContext *ctx = outlink->src;
324  DecimateContext *dm = ctx->priv;
325  const uint32_t eof_mask = 1<<INPUT_MAIN | dm->ppsrc<<INPUT_CLEANSRC;
326 
327  if ((dm->eof & eof_mask) == eof_mask) // flush done?
328  return AVERROR_EOF;
329  if ((ret = request_inlink(ctx, INPUT_MAIN)) < 0)
330  return ret;
331  if (dm->ppsrc && (ret = request_inlink(ctx, INPUT_CLEANSRC)) < 0)
332  return ret;
333  return 0;
334 }
335 
337 {
338  static const enum AVPixelFormat pix_fmts[] = {
339 #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
340 #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
341 #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
342  PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
346  };
348  return 0;
349 }
350 
351 static int config_output(AVFilterLink *outlink)
352 {
353  AVFilterContext *ctx = outlink->src;
354  DecimateContext *dm = ctx->priv;
355  const AVFilterLink *inlink =
356  ctx->inputs[dm->ppsrc ? INPUT_CLEANSRC : INPUT_MAIN];
357  AVRational fps = inlink->frame_rate;
358 
359  if (!fps.num || !fps.den) {
360  av_log(ctx, AV_LOG_ERROR, "The input needs a constant frame rate; "
361  "current rate of %d/%d is invalid\n", fps.num, fps.den);
362  return AVERROR(EINVAL);
363  }
364  fps = av_mul_q(fps, (AVRational){dm->cycle - 1, dm->cycle});
365  av_log(ctx, AV_LOG_VERBOSE, "FPS: %d/%d -> %d/%d\n",
366  inlink->frame_rate.num, inlink->frame_rate.den, fps.num, fps.den);
367  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
368  outlink->time_base = inlink->time_base;
369  outlink->frame_rate = fps;
370  outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
371  outlink->w = inlink->w;
372  outlink->h = inlink->h;
373  dm->ts_unit = av_q2d(av_inv_q(av_mul_q(fps, outlink->time_base)));
374  return 0;
375 }
376 
377 static const AVFilterPad decimate_outputs[] = {
378  {
379  .name = "default",
380  .type = AVMEDIA_TYPE_VIDEO,
381  .request_frame = request_frame,
382  .config_props = config_output,
383  },
384  { NULL }
385 };
386 
388  .name = "decimate",
389  .description = NULL_IF_CONFIG_SMALL("Decimate frames (post field matching filter)."),
390  .init = decimate_init,
391  .uninit = decimate_uninit,
392  .priv_size = sizeof(DecimateContext),
394  .outputs = decimate_outputs,
395  .priv_class = &decimate_class,
397 };