1 /*
2 * Copyright (c) 2011 Stefano Sabatini
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 /**
22 * @file
23 * filter for selecting which frame passes in the filterchain
24 */
25
26 #include "config_components.h"
27
28 #include "libavutil/avstring.h"
29 #include "libavutil/eval.h"
30 #include "libavutil/fifo.h"
31 #include "libavutil/imgutils.h"
32 #include "libavutil/internal.h"
33 #include "libavutil/opt.h"
34 #include "libavutil/pixdesc.h"
35 #include "avfilter.h"
36 #include "audio.h"
37 #include "formats.h"
38 #include "internal.h"
39 #include "video.h"
40 #include "scene_sad.h"
41
42 static const char *const var_names[] = {
43 "TB", ///< timebase
44
45 "pts", ///< original pts in the file of the frame
46 "start_pts", ///< first PTS in the stream, expressed in TB units
47 "prev_pts", ///< previous frame PTS
48 "prev_selected_pts", ///< previous selected frame PTS
49
50 "t", ///< timestamp expressed in seconds
51 "start_t", ///< first PTS in the stream, expressed in seconds
52 "prev_t", ///< previous frame time
53 "prev_selected_t", ///< previously selected time
54
55 "pict_type", ///< the type of picture in the movie
56 "I",
57 "P",
58 "B",
59 "S",
60 "SI",
61 "SP",
62 "BI",
63 "PICT_TYPE_I",
64 "PICT_TYPE_P",
65 "PICT_TYPE_B",
66 "PICT_TYPE_S",
67 "PICT_TYPE_SI",
68 "PICT_TYPE_SP",
69 "PICT_TYPE_BI",
70
71 "interlace_type", ///< the frame interlace type
72 "PROGRESSIVE",
73 "TOPFIRST",
74 "BOTTOMFIRST",
75
76 "consumed_samples_n",///< number of samples consumed by the filter (only audio)
77 "samples_n", ///< number of samples in the current frame (only audio)
78 "sample_rate", ///< sample rate (only audio)
79
80 "n", ///< frame number (starting from zero)
81 "selected_n", ///< selected frame number (starting from zero)
82 "prev_selected_n", ///< number of the last selected frame
83
84 "key", ///< tell if the frame is a key frame
85 "pos", ///< original position in the file of the frame
86
87 "scene",
88
89 "concatdec_select", ///< frame is within the interval set by the concat demuxer
90
91 NULL
92 };
93
94 enum var_name {
95 VAR_TB,
96
97 VAR_PTS,
98 VAR_START_PTS,
99 VAR_PREV_PTS,
100 VAR_PREV_SELECTED_PTS,
101
102 VAR_T,
103 VAR_START_T,
104 VAR_PREV_T,
105 VAR_PREV_SELECTED_T,
106
107 VAR_PICT_TYPE,
108 VAR_I,
109 VAR_P,
110 VAR_B,
111 VAR_S,
112 VAR_SI,
113 VAR_SP,
114 VAR_BI,
115 VAR_PICT_TYPE_I,
116 VAR_PICT_TYPE_P,
117 VAR_PICT_TYPE_B,
118 VAR_PICT_TYPE_S,
119 VAR_PICT_TYPE_SI,
120 VAR_PICT_TYPE_SP,
121 VAR_PICT_TYPE_BI,
122
123 VAR_INTERLACE_TYPE,
124 VAR_INTERLACE_TYPE_P,
125 VAR_INTERLACE_TYPE_T,
126 VAR_INTERLACE_TYPE_B,
127
128 VAR_CONSUMED_SAMPLES_N,
129 VAR_SAMPLES_N,
130 VAR_SAMPLE_RATE,
131
132 VAR_N,
133 VAR_SELECTED_N,
134 VAR_PREV_SELECTED_N,
135
136 VAR_KEY,
137 VAR_POS,
138
139 VAR_SCENE,
140
141 VAR_CONCATDEC_SELECT,
142
143 VAR_VARS_NB
144 };
145
146 typedef struct SelectContext {
147 const AVClass *class;
148 char *expr_str;
149 AVExpr *expr;
150 double var_values[VAR_VARS_NB];
151 int bitdepth;
152 int nb_planes;
153 ptrdiff_t width[4];
154 ptrdiff_t height[4];
155 int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise
156 ff_scene_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
157 double prev_mafd; ///< previous MAFD (scene detect only)
158 AVFrame *prev_picref; ///< previous frame (scene detect only)
159 double select;
160 int select_out; ///< mark the selected output pad index
161 int nb_outputs;
162 } SelectContext;
163
164 #define OFFSET(x) offsetof(SelectContext, x)
165 #define DEFINE_OPTIONS(filt_name, FLAGS) \
166 static const AVOption filt_name##_options[] = { \
167 { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
168 { "e", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
169 { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
170 { "n", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
171 { NULL } \
172 }
173
174 static int request_frame(AVFilterLink *outlink);
175
init(AVFilterContext *ctx)176 static av_cold int init(AVFilterContext *ctx)
177 {
178 SelectContext *select = ctx->priv;
179 int i, ret;
180
181 if ((ret = av_expr_parse(&select->expr, select->expr_str,
182 var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
183 av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
184 select->expr_str);
185 return ret;
186 }
187 select->do_scene_detect = !!strstr(select->expr_str, "scene");
188
189 for (i = 0; i < select->nb_outputs; i++) {
190 AVFilterPad pad = { 0 };
191
192 pad.name = av_asprintf("output%d", i);
193 if (!pad.name)
194 return AVERROR(ENOMEM);
195 pad.type = ctx->filter->inputs[0].type;
196 pad.request_frame = request_frame;
197 if ((ret = ff_append_outpad_free_name(ctx, &pad)) < 0)
198 return ret;
199 }
200
201 return 0;
202 }
203
204 #define INTERLACE_TYPE_P 0
205 #define INTERLACE_TYPE_T 1
206 #define INTERLACE_TYPE_B 2
207
config_input(AVFilterLink *inlink)208 static int config_input(AVFilterLink *inlink)
209 {
210 SelectContext *select = inlink->dst->priv;
211 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
212 int is_yuv = !(desc->flags & AV_PIX_FMT_FLAG_RGB) &&
213 (desc->flags & AV_PIX_FMT_FLAG_PLANAR) &&
214 desc->nb_components >= 3;
215
216 select->bitdepth = desc->comp[0].depth;
217 select->nb_planes = is_yuv ? 1 : av_pix_fmt_count_planes(inlink->format);
218
219 for (int plane = 0; plane < select->nb_planes; plane++) {
220 ptrdiff_t line_size = av_image_get_linesize(inlink->format, inlink->w, plane);
221 int vsub = desc->log2_chroma_h;
222
223 select->width[plane] = line_size >> (select->bitdepth > 8);
224 select->height[plane] = plane == 1 || plane == 2 ? AV_CEIL_RSHIFT(inlink->h, vsub) : inlink->h;
225 }
226
227 select->var_values[VAR_N] = 0.0;
228 select->var_values[VAR_SELECTED_N] = 0.0;
229
230 select->var_values[VAR_TB] = av_q2d(inlink->time_base);
231
232 select->var_values[VAR_PREV_PTS] = NAN;
233 select->var_values[VAR_PREV_SELECTED_PTS] = NAN;
234 select->var_values[VAR_PREV_SELECTED_T] = NAN;
235 select->var_values[VAR_PREV_T] = NAN;
236 select->var_values[VAR_START_PTS] = NAN;
237 select->var_values[VAR_START_T] = NAN;
238
239 select->var_values[VAR_I] = AV_PICTURE_TYPE_I;
240 select->var_values[VAR_P] = AV_PICTURE_TYPE_P;
241 select->var_values[VAR_B] = AV_PICTURE_TYPE_B;
242 select->var_values[VAR_SI] = AV_PICTURE_TYPE_SI;
243 select->var_values[VAR_SP] = AV_PICTURE_TYPE_SP;
244 select->var_values[VAR_BI] = AV_PICTURE_TYPE_BI;
245 select->var_values[VAR_PICT_TYPE_I] = AV_PICTURE_TYPE_I;
246 select->var_values[VAR_PICT_TYPE_P] = AV_PICTURE_TYPE_P;
247 select->var_values[VAR_PICT_TYPE_B] = AV_PICTURE_TYPE_B;
248 select->var_values[VAR_PICT_TYPE_SI] = AV_PICTURE_TYPE_SI;
249 select->var_values[VAR_PICT_TYPE_SP] = AV_PICTURE_TYPE_SP;
250 select->var_values[VAR_PICT_TYPE_BI] = AV_PICTURE_TYPE_BI;
251
252 select->var_values[VAR_INTERLACE_TYPE_P] = INTERLACE_TYPE_P;
253 select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T;
254 select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B;
255
256 select->var_values[VAR_PICT_TYPE] = NAN;
257 select->var_values[VAR_INTERLACE_TYPE] = NAN;
258 select->var_values[VAR_SCENE] = NAN;
259 select->var_values[VAR_CONSUMED_SAMPLES_N] = NAN;
260 select->var_values[VAR_SAMPLES_N] = NAN;
261
262 select->var_values[VAR_SAMPLE_RATE] =
263 inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
264
265 if (CONFIG_SELECT_FILTER && select->do_scene_detect) {
266 select->sad = ff_scene_sad_get_fn(select->bitdepth == 8 ? 8 : 16);
267 if (!select->sad)
268 return AVERROR(EINVAL);
269 }
270 return 0;
271 }
272
get_scene_score(AVFilterContext *ctx, AVFrame *frame)273 static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
274 {
275 double ret = 0;
276 SelectContext *select = ctx->priv;
277 AVFrame *prev_picref = select->prev_picref;
278
279 if (prev_picref &&
280 frame->height == prev_picref->height &&
281 frame->width == prev_picref->width) {
282 uint64_t sad = 0;
283 double mafd, diff;
284 uint64_t count = 0;
285
286 for (int plane = 0; plane < select->nb_planes; plane++) {
287 uint64_t plane_sad;
288 select->sad(prev_picref->data[plane], prev_picref->linesize[plane],
289 frame->data[plane], frame->linesize[plane],
290 select->width[plane], select->height[plane], &plane_sad);
291 sad += plane_sad;
292 count += select->width[plane] * select->height[plane];
293 }
294
295 emms_c();
296 mafd = (double)sad / count / (1ULL << (select->bitdepth - 8));
297 diff = fabs(mafd - select->prev_mafd);
298 ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
299 select->prev_mafd = mafd;
300 av_frame_free(&prev_picref);
301 }
302 select->prev_picref = av_frame_clone(frame);
303 return ret;
304 }
305
get_concatdec_select(AVFrame *frame, int64_t pts)306 static double get_concatdec_select(AVFrame *frame, int64_t pts)
307 {
308 AVDictionary *metadata = frame->metadata;
309 AVDictionaryEntry *start_time_entry = av_dict_get(metadata, "lavf.concatdec.start_time", NULL, 0);
310 AVDictionaryEntry *duration_entry = av_dict_get(metadata, "lavf.concatdec.duration", NULL, 0);
311 if (start_time_entry) {
312 int64_t start_time = strtoll(start_time_entry->value, NULL, 10);
313 if (pts >= start_time) {
314 if (duration_entry) {
315 int64_t duration = strtoll(duration_entry->value, NULL, 10);
316 if (pts < start_time + duration)
317 return -1;
318 else
319 return 0;
320 }
321 return -1;
322 }
323 return 0;
324 }
325 return NAN;
326 }
327
select_frame(AVFilterContext *ctx, AVFrame *frame)328 static void select_frame(AVFilterContext *ctx, AVFrame *frame)
329 {
330 SelectContext *select = ctx->priv;
331 AVFilterLink *inlink = ctx->inputs[0];
332 double res;
333
334 if (isnan(select->var_values[VAR_START_PTS]))
335 select->var_values[VAR_START_PTS] = TS2D(frame->pts);
336 if (isnan(select->var_values[VAR_START_T]))
337 select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
338
339 select->var_values[VAR_N ] = inlink->frame_count_out;
340 select->var_values[VAR_PTS] = TS2D(frame->pts);
341 select->var_values[VAR_T ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
342 select->var_values[VAR_POS] = frame->pkt_pos == -1 ? NAN : frame->pkt_pos;
343 select->var_values[VAR_KEY] = frame->key_frame;
344 select->var_values[VAR_CONCATDEC_SELECT] = get_concatdec_select(frame, av_rescale_q(frame->pts, inlink->time_base, AV_TIME_BASE_Q));
345
346 switch (inlink->type) {
347 case AVMEDIA_TYPE_AUDIO:
348 select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
349 break;
350
351 case AVMEDIA_TYPE_VIDEO:
352 select->var_values[VAR_INTERLACE_TYPE] =
353 !frame->interlaced_frame ? INTERLACE_TYPE_P :
354 frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
355 select->var_values[VAR_PICT_TYPE] = frame->pict_type;
356 if (select->do_scene_detect) {
357 char buf[32];
358 select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
359 // TODO: document metadata
360 snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
361 av_dict_set(&frame->metadata, "lavfi.scene_score", buf, 0);
362 }
363 break;
364 }
365
366 select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
367 av_log(inlink->dst, AV_LOG_DEBUG,
368 "n:%f pts:%f t:%f key:%d",
369 select->var_values[VAR_N],
370 select->var_values[VAR_PTS],
371 select->var_values[VAR_T],
372 frame->key_frame);
373
374 switch (inlink->type) {
375 case AVMEDIA_TYPE_VIDEO:
376 av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
377 (!frame->interlaced_frame) ? 'P' :
378 frame->top_field_first ? 'T' : 'B',
379 av_get_picture_type_char(frame->pict_type),
380 select->var_values[VAR_SCENE]);
381 break;
382 case AVMEDIA_TYPE_AUDIO:
383 av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%f",
384 frame->nb_samples,
385 select->var_values[VAR_CONSUMED_SAMPLES_N]);
386 break;
387 }
388
389 if (res == 0) {
390 select->select_out = -1; /* drop */
391 } else if (isnan(res) || res < 0) {
392 select->select_out = 0; /* first output */
393 } else {
394 select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
395 }
396
397 av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
398
399 if (res) {
400 select->var_values[VAR_PREV_SELECTED_N] = select->var_values[VAR_N];
401 select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
402 select->var_values[VAR_PREV_SELECTED_T] = select->var_values[VAR_T];
403 select->var_values[VAR_SELECTED_N] += 1.0;
404 if (inlink->type == AVMEDIA_TYPE_AUDIO)
405 select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
406 }
407
408 select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
409 select->var_values[VAR_PREV_T] = select->var_values[VAR_T];
410 }
411
filter_frame(AVFilterLink *inlink, AVFrame *frame)412 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
413 {
414 AVFilterContext *ctx = inlink->dst;
415 SelectContext *select = ctx->priv;
416
417 select_frame(ctx, frame);
418 if (select->select)
419 return ff_filter_frame(ctx->outputs[select->select_out], frame);
420
421 av_frame_free(&frame);
422 return 0;
423 }
424
request_frame(AVFilterLink *outlink)425 static int request_frame(AVFilterLink *outlink)
426 {
427 AVFilterLink *inlink = outlink->src->inputs[0];
428 int ret = ff_request_frame(inlink);
429 return ret;
430 }
431
uninit(AVFilterContext *ctx)432 static av_cold void uninit(AVFilterContext *ctx)
433 {
434 SelectContext *select = ctx->priv;
435
436 av_expr_free(select->expr);
437 select->expr = NULL;
438
439 if (select->do_scene_detect) {
440 av_frame_free(&select->prev_picref);
441 }
442 }
443
444 #if CONFIG_ASELECT_FILTER
445
446 DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
447 AVFILTER_DEFINE_CLASS(aselect);
448
aselect_init(AVFilterContext *ctx)449 static av_cold int aselect_init(AVFilterContext *ctx)
450 {
451 SelectContext *select = ctx->priv;
452 int ret;
453
454 if ((ret = init(ctx)) < 0)
455 return ret;
456
457 if (select->do_scene_detect) {
458 av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
459 return AVERROR(EINVAL);
460 }
461
462 return 0;
463 }
464
465 static const AVFilterPad avfilter_af_aselect_inputs[] = {
466 {
467 .name = "default",
468 .type = AVMEDIA_TYPE_AUDIO,
469 .config_props = config_input,
470 .filter_frame = filter_frame,
471 },
472 };
473
474 const AVFilter ff_af_aselect = {
475 .name = "aselect",
476 .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
477 .init = aselect_init,
478 .uninit = uninit,
479 .priv_size = sizeof(SelectContext),
480 FILTER_INPUTS(avfilter_af_aselect_inputs),
481 .priv_class = &aselect_class,
482 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
483 };
484 #endif /* CONFIG_ASELECT_FILTER */
485
486 #if CONFIG_SELECT_FILTER
487
query_formats(AVFilterContext *ctx)488 static int query_formats(AVFilterContext *ctx)
489 {
490 SelectContext *select = ctx->priv;
491
492 if (!select->do_scene_detect) {
493 return ff_default_query_formats(ctx);
494 } else {
495 static const enum AVPixelFormat pix_fmts[] = {
496 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24, AV_PIX_FMT_RGBA,
497 AV_PIX_FMT_ABGR, AV_PIX_FMT_BGRA, AV_PIX_FMT_GRAY8,
498 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
499 AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVJ422P,
500 AV_PIX_FMT_YUV420P10,
501 AV_PIX_FMT_NONE
502 };
503 return ff_set_common_formats_from_list(ctx, pix_fmts);
504 }
505 }
506
507 DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
508 AVFILTER_DEFINE_CLASS(select);
509
select_init(AVFilterContext *ctx)510 static av_cold int select_init(AVFilterContext *ctx)
511 {
512 int ret;
513
514 if ((ret = init(ctx)) < 0)
515 return ret;
516
517 return 0;
518 }
519
520 static const AVFilterPad avfilter_vf_select_inputs[] = {
521 {
522 .name = "default",
523 .type = AVMEDIA_TYPE_VIDEO,
524 .config_props = config_input,
525 .filter_frame = filter_frame,
526 },
527 };
528
529 const AVFilter ff_vf_select = {
530 .name = "select",
531 .description = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
532 .init = select_init,
533 .uninit = uninit,
534 .priv_size = sizeof(SelectContext),
535 .priv_class = &select_class,
536 FILTER_INPUTS(avfilter_vf_select_inputs),
537 FILTER_QUERY_FUNC(query_formats),
538 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS | AVFILTER_FLAG_METADATA_ONLY,
539 };
540 #endif /* CONFIG_SELECT_FILTER */
541