2 * Copyright (c) 2011 Stefano Sabatini
4 * This file is part of FFmpeg.
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.
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.
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
23 * filter for selecting which frame passes in the filterchain
26 #include "libavutil/avstring.h"
27 #include "libavutil/eval.h"
28 #include "libavutil/fifo.h"
29 #include "libavutil/internal.h"
30 #include "libavutil/opt.h"
36 #include "scene_sad.h"
38 static const char *const var_names[] = {
41 "pts", ///< original pts in the file of the frame
42 "start_pts", ///< first PTS in the stream, expressed in TB units
43 "prev_pts", ///< previous frame PTS
44 "prev_selected_pts", ///< previous selected frame PTS
46 "t", ///< timestamp expressed in seconds
47 "start_t", ///< first PTS in the stream, expressed in seconds
48 "prev_t", ///< previous frame time
49 "prev_selected_t", ///< previously selected time
51 "pict_type", ///< the type of picture in the movie
67 "interlace_type", ///< the frame interlace type
72 "consumed_samples_n",///< number of samples consumed by the filter (only audio)
73 "samples_n", ///< number of samples in the current frame (only audio)
74 "sample_rate", ///< sample rate (only audio)
76 "n", ///< frame number (starting from zero)
77 "selected_n", ///< selected frame number (starting from zero)
78 "prev_selected_n", ///< number of the last selected frame
80 "key", ///< tell if the frame is a key frame
81 "pos", ///< original position in the file of the frame
85 "concatdec_select", ///< frame is within the interval set by the concat demuxer
96 VAR_PREV_SELECTED_PTS,
120 VAR_INTERLACE_TYPE_P,
121 VAR_INTERLACE_TYPE_T,
122 VAR_INTERLACE_TYPE_B,
124 VAR_CONSUMED_SAMPLES_N,
137 VAR_CONCATDEC_SELECT,
142 typedef struct SelectContext {
143 const AVClass *class;
146 double var_values[VAR_VARS_NB];
147 int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise
148 ff_scene_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
149 double prev_mafd; ///< previous MAFD (scene detect only)
150 AVFrame *prev_picref; ///< previous frame (scene detect only)
152 int select_out; ///< mark the selected output pad index
156 #define OFFSET(x) offsetof(SelectContext, x)
157 #define DEFINE_OPTIONS(filt_name, FLAGS) \
158 static const AVOption filt_name##_options[] = { \
159 { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
160 { "e", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
161 { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
162 { "n", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
166 static int request_frame(AVFilterLink *outlink);
168 static av_cold int init(AVFilterContext *ctx)
170 SelectContext *select = ctx->priv;
173 if ((ret = av_expr_parse(&select->expr, select->expr_str,
174 var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
175 av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
179 select->do_scene_detect = !!strstr(select->expr_str, "scene");
181 for (i = 0; i < select->nb_outputs; i++) {
182 AVFilterPad pad = { 0 };
184 pad.name = av_asprintf("output%d", i);
186 return AVERROR(ENOMEM);
187 pad.type = ctx->filter->inputs[0].type;
188 pad.request_frame = request_frame;
189 if ((ret = ff_insert_outpad(ctx, i, &pad)) < 0) {
198 #define INTERLACE_TYPE_P 0
199 #define INTERLACE_TYPE_T 1
200 #define INTERLACE_TYPE_B 2
202 static int config_input(AVFilterLink *inlink)
204 SelectContext *select = inlink->dst->priv;
206 select->var_values[VAR_N] = 0.0;
207 select->var_values[VAR_SELECTED_N] = 0.0;
209 select->var_values[VAR_TB] = av_q2d(inlink->time_base);
211 select->var_values[VAR_PREV_PTS] = NAN;
212 select->var_values[VAR_PREV_SELECTED_PTS] = NAN;
213 select->var_values[VAR_PREV_SELECTED_T] = NAN;
214 select->var_values[VAR_PREV_T] = NAN;
215 select->var_values[VAR_START_PTS] = NAN;
216 select->var_values[VAR_START_T] = NAN;
218 select->var_values[VAR_I] = AV_PICTURE_TYPE_I;
219 select->var_values[VAR_P] = AV_PICTURE_TYPE_P;
220 select->var_values[VAR_B] = AV_PICTURE_TYPE_B;
221 select->var_values[VAR_SI] = AV_PICTURE_TYPE_SI;
222 select->var_values[VAR_SP] = AV_PICTURE_TYPE_SP;
223 select->var_values[VAR_BI] = AV_PICTURE_TYPE_BI;
224 select->var_values[VAR_PICT_TYPE_I] = AV_PICTURE_TYPE_I;
225 select->var_values[VAR_PICT_TYPE_P] = AV_PICTURE_TYPE_P;
226 select->var_values[VAR_PICT_TYPE_B] = AV_PICTURE_TYPE_B;
227 select->var_values[VAR_PICT_TYPE_SI] = AV_PICTURE_TYPE_SI;
228 select->var_values[VAR_PICT_TYPE_SP] = AV_PICTURE_TYPE_SP;
229 select->var_values[VAR_PICT_TYPE_BI] = AV_PICTURE_TYPE_BI;
231 select->var_values[VAR_INTERLACE_TYPE_P] = INTERLACE_TYPE_P;
232 select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T;
233 select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B;
235 select->var_values[VAR_PICT_TYPE] = NAN;
236 select->var_values[VAR_INTERLACE_TYPE] = NAN;
237 select->var_values[VAR_SCENE] = NAN;
238 select->var_values[VAR_CONSUMED_SAMPLES_N] = NAN;
239 select->var_values[VAR_SAMPLES_N] = NAN;
241 select->var_values[VAR_SAMPLE_RATE] =
242 inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
244 if (CONFIG_SELECT_FILTER && select->do_scene_detect) {
245 select->sad = ff_scene_sad_get_fn(8);
247 return AVERROR(EINVAL);
252 static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
255 SelectContext *select = ctx->priv;
256 AVFrame *prev_picref = select->prev_picref;
259 frame->height == prev_picref->height &&
260 frame->width == prev_picref->width) {
264 select->sad(prev_picref->data[0], prev_picref->linesize[0], frame->data[0], frame->linesize[0], frame->width * 3, frame->height, &sad);
266 mafd = (double)sad / (frame->width * 3 * frame->height);
267 diff = fabs(mafd - select->prev_mafd);
268 ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
269 select->prev_mafd = mafd;
270 av_frame_free(&prev_picref);
272 select->prev_picref = av_frame_clone(frame);
276 static double get_concatdec_select(AVFrame *frame, int64_t pts)
278 AVDictionary *metadata = frame->metadata;
279 AVDictionaryEntry *start_time_entry = av_dict_get(metadata, "lavf.concatdec.start_time", NULL, 0);
280 AVDictionaryEntry *duration_entry = av_dict_get(metadata, "lavf.concatdec.duration", NULL, 0);
281 if (start_time_entry) {
282 int64_t start_time = strtoll(start_time_entry->value, NULL, 10);
283 if (pts >= start_time) {
284 if (duration_entry) {
285 int64_t duration = strtoll(duration_entry->value, NULL, 10);
286 if (pts < start_time + duration)
298 #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
299 #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
301 static void select_frame(AVFilterContext *ctx, AVFrame *frame)
303 SelectContext *select = ctx->priv;
304 AVFilterLink *inlink = ctx->inputs[0];
307 if (isnan(select->var_values[VAR_START_PTS]))
308 select->var_values[VAR_START_PTS] = TS2D(frame->pts);
309 if (isnan(select->var_values[VAR_START_T]))
310 select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
312 select->var_values[VAR_N ] = inlink->frame_count_out;
313 select->var_values[VAR_PTS] = TS2D(frame->pts);
314 select->var_values[VAR_T ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
315 select->var_values[VAR_POS] = frame->pkt_pos == -1 ? NAN : frame->pkt_pos;
316 select->var_values[VAR_KEY] = frame->key_frame;
317 select->var_values[VAR_CONCATDEC_SELECT] = get_concatdec_select(frame, av_rescale_q(frame->pts, inlink->time_base, AV_TIME_BASE_Q));
319 switch (inlink->type) {
320 case AVMEDIA_TYPE_AUDIO:
321 select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
324 case AVMEDIA_TYPE_VIDEO:
325 select->var_values[VAR_INTERLACE_TYPE] =
326 !frame->interlaced_frame ? INTERLACE_TYPE_P :
327 frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
328 select->var_values[VAR_PICT_TYPE] = frame->pict_type;
329 if (select->do_scene_detect) {
331 select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
332 // TODO: document metadata
333 snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
334 av_dict_set(&frame->metadata, "lavfi.scene_score", buf, 0);
339 select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
340 av_log(inlink->dst, AV_LOG_DEBUG,
341 "n:%f pts:%f t:%f key:%d",
342 select->var_values[VAR_N],
343 select->var_values[VAR_PTS],
344 select->var_values[VAR_T],
347 switch (inlink->type) {
348 case AVMEDIA_TYPE_VIDEO:
349 av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
350 (!frame->interlaced_frame) ? 'P' :
351 frame->top_field_first ? 'T' : 'B',
352 av_get_picture_type_char(frame->pict_type),
353 select->var_values[VAR_SCENE]);
355 case AVMEDIA_TYPE_AUDIO:
356 av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%f",
358 select->var_values[VAR_CONSUMED_SAMPLES_N]);
363 select->select_out = -1; /* drop */
364 } else if (isnan(res) || res < 0) {
365 select->select_out = 0; /* first output */
367 select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
370 av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
373 select->var_values[VAR_PREV_SELECTED_N] = select->var_values[VAR_N];
374 select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
375 select->var_values[VAR_PREV_SELECTED_T] = select->var_values[VAR_T];
376 select->var_values[VAR_SELECTED_N] += 1.0;
377 if (inlink->type == AVMEDIA_TYPE_AUDIO)
378 select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
381 select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
382 select->var_values[VAR_PREV_T] = select->var_values[VAR_T];
385 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
387 AVFilterContext *ctx = inlink->dst;
388 SelectContext *select = ctx->priv;
390 select_frame(ctx, frame);
392 return ff_filter_frame(ctx->outputs[select->select_out], frame);
394 av_frame_free(&frame);
398 static int request_frame(AVFilterLink *outlink)
400 AVFilterLink *inlink = outlink->src->inputs[0];
401 int ret = ff_request_frame(inlink);
405 static av_cold void uninit(AVFilterContext *ctx)
407 SelectContext *select = ctx->priv;
410 av_expr_free(select->expr);
413 for (i = 0; i < ctx->nb_outputs; i++)
414 av_freep(&ctx->output_pads[i].name);
416 if (select->do_scene_detect) {
417 av_frame_free(&select->prev_picref);
421 #if CONFIG_ASELECT_FILTER
423 DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
424 AVFILTER_DEFINE_CLASS(aselect);
426 static av_cold int aselect_init(AVFilterContext *ctx)
428 SelectContext *select = ctx->priv;
431 if ((ret = init(ctx)) < 0)
434 if (select->do_scene_detect) {
435 av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
436 return AVERROR(EINVAL);
442 static const AVFilterPad avfilter_af_aselect_inputs[] = {
445 .type = AVMEDIA_TYPE_AUDIO,
446 .config_props = config_input,
447 .filter_frame = filter_frame,
452 AVFilter ff_af_aselect = {
454 .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
455 .init = aselect_init,
457 .priv_size = sizeof(SelectContext),
458 .inputs = avfilter_af_aselect_inputs,
459 .priv_class = &aselect_class,
460 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
462 #endif /* CONFIG_ASELECT_FILTER */
464 #if CONFIG_SELECT_FILTER
466 static int query_formats(AVFilterContext *ctx)
468 SelectContext *select = ctx->priv;
470 if (!select->do_scene_detect) {
471 return ff_default_query_formats(ctx);
474 static const enum AVPixelFormat pix_fmts[] = {
475 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
478 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
481 return AVERROR(ENOMEM);
482 ret = ff_set_common_formats(ctx, fmts_list);
489 DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
490 AVFILTER_DEFINE_CLASS(select);
492 static av_cold int select_init(AVFilterContext *ctx)
496 if ((ret = init(ctx)) < 0)
502 static const AVFilterPad avfilter_vf_select_inputs[] = {
505 .type = AVMEDIA_TYPE_VIDEO,
506 .config_props = config_input,
507 .filter_frame = filter_frame,
512 AVFilter ff_vf_select = {
514 .description = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
517 .query_formats = query_formats,
518 .priv_size = sizeof(SelectContext),
519 .priv_class = &select_class,
520 .inputs = avfilter_vf_select_inputs,
521 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
523 #endif /* CONFIG_SELECT_FILTER */