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  int64_t start_pts; ///< base for output timestamps
47  uint32_t eof; ///< bitmask for end of stream
48  int hsub, vsub; ///< chroma subsampling values
49  int depth;
50  int nxblocks, nyblocks;
51  int bdiffsize;
52  int64_t *bdiffs;
53 
54  /* options */
55  int cycle;
56  double dupthresh_flt;
57  double scthresh_flt;
58  int64_t dupthresh;
59  int64_t scthresh;
60  int blockx, blocky;
61  int ppsrc;
62  int chroma;
64 
65 #define OFFSET(x) offsetof(DecimateContext, x)
66 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
67 
68 static const AVOption decimate_options[] = {
69  { "cycle", "set the number of frame from which one will be dropped", OFFSET(cycle), AV_OPT_TYPE_INT, {.i64 = 5}, 2, 25, FLAGS },
70  { "dupthresh", "set duplicate threshold", OFFSET(dupthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 1.1}, 0, 100, FLAGS },
71  { "scthresh", "set scene change threshold", OFFSET(scthresh_flt), AV_OPT_TYPE_DOUBLE, {.dbl = 15.0}, 0, 100, FLAGS },
72  { "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 },
73  { "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 },
74  { "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 },
75  { "chroma", "set whether or not chroma is considered in the metric calculations", OFFSET(chroma), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS },
76  { NULL }
77 };
78 
79 AVFILTER_DEFINE_CLASS(decimate);
80 
81 static void calc_diffs(const DecimateContext *dm, struct qitem *q,
82  const AVFrame *f1, const AVFrame *f2)
83 {
84  int64_t maxdiff = -1;
85  int64_t *bdiffs = dm->bdiffs;
86  int plane, i, j;
87 
88  memset(bdiffs, 0, dm->bdiffsize * sizeof(*bdiffs));
89 
90  for (plane = 0; plane < (dm->chroma && f1->data[2] ? 3 : 1); plane++) {
91  int x, y, xl;
92  const int linesize1 = f1->linesize[plane];
93  const int linesize2 = f2->linesize[plane];
94  const uint8_t *f1p = f1->data[plane];
95  const uint8_t *f2p = f2->data[plane];
96  int width = plane ? FF_CEIL_RSHIFT(f1->width, dm->hsub) : f1->width;
97  int height = plane ? FF_CEIL_RSHIFT(f1->height, dm->vsub) : f1->height;
98  int hblockx = dm->blockx / 2;
99  int hblocky = dm->blocky / 2;
100 
101  if (plane) {
102  hblockx >>= dm->hsub;
103  hblocky >>= dm->vsub;
104  }
105 
106  for (y = 0; y < height; y++) {
107  int ydest = y / hblocky;
108  int xdest = 0;
109 
110 #define CALC_DIFF(nbits) do { \
111  for (x = 0; x < width; x += hblockx) { \
112  int64_t acc = 0; \
113  int m = FFMIN(width, x + hblockx); \
114  for (xl = x; xl < m; xl++) \
115  acc += abs(((const uint##nbits##_t *)f1p)[xl] - \
116  ((const uint##nbits##_t *)f2p)[xl]); \
117  bdiffs[ydest * dm->nxblocks + xdest] += acc; \
118  xdest++; \
119  } \
120 } while (0)
121  if (dm->depth == 8) CALC_DIFF(8);
122  else CALC_DIFF(16);
123 
124  f1p += linesize1;
125  f2p += linesize2;
126  }
127  }
128 
129  for (i = 0; i < dm->nyblocks - 1; i++) {
130  for (j = 0; j < dm->nxblocks - 1; j++) {
131  int64_t tmp = bdiffs[ i * dm->nxblocks + j ]
132  + bdiffs[ i * dm->nxblocks + j + 1]
133  + bdiffs[(i + 1) * dm->nxblocks + j ]
134  + bdiffs[(i + 1) * dm->nxblocks + j + 1];
135  if (tmp > maxdiff)
136  maxdiff = tmp;
137  }
138  }
139 
140  q->totdiff = 0;
141  for (i = 0; i < dm->bdiffsize; i++)
142  q->totdiff += bdiffs[i];
143  q->maxbdiff = maxdiff;
144 }
145 
146 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
147 {
148  int scpos = -1, duppos = -1;
149  int drop = INT_MIN, i, lowest = 0, ret;
150  AVFilterContext *ctx = inlink->dst;
151  AVFilterLink *outlink = ctx->outputs[0];
152  DecimateContext *dm = ctx->priv;
153  AVFrame *prv;
154 
155  /* update frames queue(s) */
156  if (FF_INLINK_IDX(inlink) == INPUT_MAIN) {
157  dm->queue[dm->fid].frame = in;
158  dm->got_frame[INPUT_MAIN] = 1;
159  } else {
160  dm->clean_src[dm->fid] = in;
161  dm->got_frame[INPUT_CLEANSRC] = 1;
162  }
163  if (!dm->got_frame[INPUT_MAIN] || (dm->ppsrc && !dm->got_frame[INPUT_CLEANSRC]))
164  return 0;
166 
167  if (in) {
168  /* update frame metrics */
169  prv = dm->fid ? dm->queue[dm->fid - 1].frame : dm->last;
170  if (!prv)
171  prv = in;
172  calc_diffs(dm, &dm->queue[dm->fid], prv, in);
173  if (++dm->fid != dm->cycle)
174  return 0;
175  av_frame_free(&dm->last);
176  dm->last = av_frame_clone(in);
177  dm->fid = 0;
178 
179  /* we have a complete cycle, select the frame to drop */
180  lowest = 0;
181  for (i = 0; i < dm->cycle; i++) {
182  if (dm->queue[i].totdiff > dm->scthresh)
183  scpos = i;
184  if (dm->queue[i].maxbdiff < dm->queue[lowest].maxbdiff)
185  lowest = i;
186  }
187  if (dm->queue[lowest].maxbdiff < dm->dupthresh)
188  duppos = lowest;
189  drop = scpos >= 0 && duppos < 0 ? scpos : lowest;
190  }
191 
192  /* metrics debug */
193  if (av_log_get_level() >= AV_LOG_DEBUG) {
194  av_log(ctx, AV_LOG_DEBUG, "1/%d frame drop:\n", dm->cycle);
195  for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) {
196  av_log(ctx, AV_LOG_DEBUG," #%d: totdiff=%08"PRIx64" maxbdiff=%08"PRIx64"%s%s%s%s\n",
197  i + 1, dm->queue[i].totdiff, dm->queue[i].maxbdiff,
198  i == scpos ? " sc" : "",
199  i == duppos ? " dup" : "",
200  i == lowest ? " lowest" : "",
201  i == drop ? " [DROP]" : "");
202  }
203  }
204 
205  /* push all frames except the drop */
206  ret = 0;
207  for (i = 0; i < dm->cycle && dm->queue[i].frame; i++) {
208  if (i == drop) {
209  if (dm->ppsrc)
210  av_frame_free(&dm->clean_src[i]);
211  av_frame_free(&dm->queue[i].frame);
212  } else {
213  AVFrame *frame = dm->queue[i].frame;
214  if (frame->pts != AV_NOPTS_VALUE && dm->start_pts == AV_NOPTS_VALUE)
215  dm->start_pts = frame->pts;
216  if (dm->ppsrc) {
217  av_frame_free(&frame);
218  frame = dm->clean_src[i];
219  }
220  frame->pts = outlink->frame_count * dm->ts_unit +
221  (dm->start_pts == AV_NOPTS_VALUE ? 0 : dm->start_pts);
222  ret = ff_filter_frame(outlink, frame);
223  if (ret < 0)
224  break;
225  }
226  }
227 
228  return ret;
229 }
230 
231 static int config_input(AVFilterLink *inlink)
232 {
233  int max_value;
234  AVFilterContext *ctx = inlink->dst;
235  DecimateContext *dm = ctx->priv;
236  const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format);
237  const int w = inlink->w;
238  const int h = inlink->h;
239 
240  dm->hsub = pix_desc->log2_chroma_w;
241  dm->vsub = pix_desc->log2_chroma_h;
242  dm->depth = pix_desc->comp[0].depth_minus1 + 1;
243  max_value = (1 << dm->depth) - 1;
244  dm->scthresh = (int64_t)(((int64_t)max_value * w * h * dm->scthresh_flt) / 100);
245  dm->dupthresh = (int64_t)(((int64_t)max_value * dm->blockx * dm->blocky * dm->dupthresh_flt) / 100);
246  dm->nxblocks = (w + dm->blockx/2 - 1) / (dm->blockx/2);
247  dm->nyblocks = (h + dm->blocky/2 - 1) / (dm->blocky/2);
248  dm->bdiffsize = dm->nxblocks * dm->nyblocks;
249  dm->bdiffs = av_malloc_array(dm->bdiffsize, sizeof(*dm->bdiffs));
250  dm->queue = av_calloc(dm->cycle, sizeof(*dm->queue));
251 
252  if (!dm->bdiffs || !dm->queue)
253  return AVERROR(ENOMEM);
254 
255  if (dm->ppsrc) {
256  dm->clean_src = av_calloc(dm->cycle, sizeof(*dm->clean_src));
257  if (!dm->clean_src)
258  return AVERROR(ENOMEM);
259  }
260 
261  return 0;
262 }
263 
265 {
266  DecimateContext *dm = ctx->priv;
267  AVFilterPad pad = {
268  .name = av_strdup("main"),
269  .type = AVMEDIA_TYPE_VIDEO,
270  .filter_frame = filter_frame,
271  .config_props = config_input,
272  };
273 
274  if (!pad.name)
275  return AVERROR(ENOMEM);
276  ff_insert_inpad(ctx, INPUT_MAIN, &pad);
277 
278  if (dm->ppsrc) {
279  pad.name = av_strdup("clean_src");
280  pad.config_props = NULL;
281  if (!pad.name)
282  return AVERROR(ENOMEM);
283  ff_insert_inpad(ctx, INPUT_CLEANSRC, &pad);
284  }
285 
286  if ((dm->blockx & (dm->blockx - 1)) ||
287  (dm->blocky & (dm->blocky - 1))) {
288  av_log(ctx, AV_LOG_ERROR, "blockx and blocky settings must be power of two\n");
289  return AVERROR(EINVAL);
290  }
291 
293 
294  return 0;
295 }
296 
298 {
299  int i;
300  DecimateContext *dm = ctx->priv;
301 
302  av_frame_free(&dm->last);
303  av_freep(&dm->bdiffs);
304  av_freep(&dm->queue);
305  av_freep(&dm->clean_src);
306  for (i = 0; i < ctx->nb_inputs; i++)
307  av_freep(&ctx->input_pads[i].name);
308 }
309 
310 static int request_inlink(AVFilterContext *ctx, int lid)
311 {
312  int ret = 0;
313  DecimateContext *dm = ctx->priv;
314 
315  if (!dm->got_frame[lid]) {
316  AVFilterLink *inlink = ctx->inputs[lid];
317  ret = ff_request_frame(inlink);
318  if (ret == AVERROR_EOF) { // flushing
319  dm->eof |= 1 << lid;
320  ret = filter_frame(inlink, NULL);
321  }
322  }
323  return ret;
324 }
325 
326 static int request_frame(AVFilterLink *outlink)
327 {
328  int ret;
329  AVFilterContext *ctx = outlink->src;
330  DecimateContext *dm = ctx->priv;
331  const uint32_t eof_mask = 1<<INPUT_MAIN | dm->ppsrc<<INPUT_CLEANSRC;
332 
333  if ((dm->eof & eof_mask) == eof_mask) // flush done?
334  return AVERROR_EOF;
335  if ((ret = request_inlink(ctx, INPUT_MAIN)) < 0)
336  return ret;
337  if (dm->ppsrc && (ret = request_inlink(ctx, INPUT_CLEANSRC)) < 0)
338  return ret;
339  return 0;
340 }
341 
343 {
344  static const enum AVPixelFormat pix_fmts[] = {
345 #define PF_NOALPHA(suf) AV_PIX_FMT_YUV420##suf, AV_PIX_FMT_YUV422##suf, AV_PIX_FMT_YUV444##suf
346 #define PF_ALPHA(suf) AV_PIX_FMT_YUVA420##suf, AV_PIX_FMT_YUVA422##suf, AV_PIX_FMT_YUVA444##suf
347 #define PF(suf) PF_NOALPHA(suf), PF_ALPHA(suf)
348  PF(P), PF(P9), PF(P10), PF_NOALPHA(P12), PF_NOALPHA(P14), PF(P16),
352  };
354  return 0;
355 }
356 
357 static int config_output(AVFilterLink *outlink)
358 {
359  AVFilterContext *ctx = outlink->src;
360  DecimateContext *dm = ctx->priv;
361  const AVFilterLink *inlink =
362  ctx->inputs[dm->ppsrc ? INPUT_CLEANSRC : INPUT_MAIN];
363  AVRational fps = inlink->frame_rate;
364 
365  if (!fps.num || !fps.den) {
366  av_log(ctx, AV_LOG_ERROR, "The input needs a constant frame rate; "
367  "current rate of %d/%d is invalid\n", fps.num, fps.den);
368  return AVERROR(EINVAL);
369  }
370  fps = av_mul_q(fps, (AVRational){dm->cycle - 1, dm->cycle});
371  av_log(ctx, AV_LOG_VERBOSE, "FPS: %d/%d -> %d/%d\n",
372  inlink->frame_rate.num, inlink->frame_rate.den, fps.num, fps.den);
373  outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
374  outlink->time_base = inlink->time_base;
375  outlink->frame_rate = fps;
376  outlink->sample_aspect_ratio = inlink->sample_aspect_ratio;
377  outlink->w = inlink->w;
378  outlink->h = inlink->h;
379  dm->ts_unit = av_q2d(av_inv_q(av_mul_q(fps, outlink->time_base)));
380  return 0;
381 }
382 
383 static const AVFilterPad decimate_outputs[] = {
384  {
385  .name = "default",
386  .type = AVMEDIA_TYPE_VIDEO,
387  .request_frame = request_frame,
388  .config_props = config_output,
389  },
390  { NULL }
391 };
392 
394  .name = "decimate",
395  .description = NULL_IF_CONFIG_SMALL("Decimate frames (post field matching filter)."),
396  .init = decimate_init,
397  .uninit = decimate_uninit,
398  .priv_size = sizeof(DecimateContext),
400  .outputs = decimate_outputs,
401  .priv_class = &decimate_class,
403 };