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/imgutils.h"
30 #include "libavutil/internal.h"
31 #include "libavutil/opt.h"
32 #include "libavutil/pixdesc.h"
38 #include "scene_sad.h"
40 static const char *const var_names[] = {
43 "pts", ///< original pts in the file of the frame
44 "start_pts", ///< first PTS in the stream, expressed in TB units
45 "prev_pts", ///< previous frame PTS
46 "prev_selected_pts", ///< previous selected frame PTS
48 "t", ///< timestamp expressed in seconds
49 "start_t", ///< first PTS in the stream, expressed in seconds
50 "prev_t", ///< previous frame time
51 "prev_selected_t", ///< previously selected time
53 "pict_type", ///< the type of picture in the movie
69 "interlace_type", ///< the frame interlace type
74 "consumed_samples_n",///< number of samples consumed by the filter (only audio)
75 "samples_n", ///< number of samples in the current frame (only audio)
76 "sample_rate", ///< sample rate (only audio)
78 "n", ///< frame number (starting from zero)
79 "selected_n", ///< selected frame number (starting from zero)
80 "prev_selected_n", ///< number of the last selected frame
82 "key", ///< tell if the frame is a key frame
83 "pos", ///< original position in the file of the frame
87 "concatdec_select", ///< frame is within the interval set by the concat demuxer
98 VAR_PREV_SELECTED_PTS,
122 VAR_INTERLACE_TYPE_P,
123 VAR_INTERLACE_TYPE_T,
124 VAR_INTERLACE_TYPE_B,
126 VAR_CONSUMED_SAMPLES_N,
139 VAR_CONCATDEC_SELECT,
144 typedef struct SelectContext {
145 const AVClass *class;
148 double var_values[VAR_VARS_NB];
153 int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise
154 ff_scene_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
155 double prev_mafd; ///< previous MAFD (scene detect only)
156 AVFrame *prev_picref; ///< previous frame (scene detect only)
158 int select_out; ///< mark the selected output pad index
162 #define OFFSET(x) offsetof(SelectContext, x)
163 #define DEFINE_OPTIONS(filt_name, FLAGS) \
164 static const AVOption filt_name##_options[] = { \
165 { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
166 { "e", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
167 { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
168 { "n", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
172 static int request_frame(AVFilterLink *outlink);
174 static av_cold int init(AVFilterContext *ctx)
176 SelectContext *select = ctx->priv;
179 if ((ret = av_expr_parse(&select->expr, select->expr_str,
180 var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
181 av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
185 select->do_scene_detect = !!strstr(select->expr_str, "scene");
187 for (i = 0; i < select->nb_outputs; i++) {
188 AVFilterPad pad = { 0 };
190 pad.name = av_asprintf("output%d", i);
192 return AVERROR(ENOMEM);
193 pad.type = ctx->filter->inputs[0].type;
194 pad.request_frame = request_frame;
195 if ((ret = ff_insert_outpad(ctx, i, &pad)) < 0) {
204 #define INTERLACE_TYPE_P 0
205 #define INTERLACE_TYPE_T 1
206 #define INTERLACE_TYPE_B 2
208 static int config_input(AVFilterLink *inlink)
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;
216 select->bitdepth = desc->comp[0].depth;
217 select->nb_planes = is_yuv ? 1 : av_pix_fmt_count_planes(inlink->format);
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;
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;
227 select->var_values[VAR_N] = 0.0;
228 select->var_values[VAR_SELECTED_N] = 0.0;
230 select->var_values[VAR_TB] = av_q2d(inlink->time_base);
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;
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;
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;
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;
262 select->var_values[VAR_SAMPLE_RATE] =
263 inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
265 if (CONFIG_SELECT_FILTER && select->do_scene_detect) {
266 select->sad = ff_scene_sad_get_fn(select->bitdepth == 8 ? 8 : 16);
268 return AVERROR(EINVAL);
273 static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
276 SelectContext *select = ctx->priv;
277 AVFrame *prev_picref = select->prev_picref;
280 frame->height == prev_picref->height &&
281 frame->width == prev_picref->width) {
286 for (int plane = 0; plane < select->nb_planes; plane++) {
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);
292 count += select->width[plane] * select->height[plane];
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);
302 select->prev_picref = av_frame_clone(frame);
306 static double get_concatdec_select(AVFrame *frame, int64_t pts)
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)
328 #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
329 #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
331 static void select_frame(AVFilterContext *ctx, AVFrame *frame)
333 SelectContext *select = ctx->priv;
334 AVFilterLink *inlink = ctx->inputs[0];
337 if (isnan(select->var_values[VAR_START_PTS]))
338 select->var_values[VAR_START_PTS] = TS2D(frame->pts);
339 if (isnan(select->var_values[VAR_START_T]))
340 select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
342 select->var_values[VAR_N ] = inlink->frame_count_out;
343 select->var_values[VAR_PTS] = TS2D(frame->pts);
344 select->var_values[VAR_T ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
345 select->var_values[VAR_POS] = frame->pkt_pos == -1 ? NAN : frame->pkt_pos;
346 select->var_values[VAR_KEY] = frame->key_frame;
347 select->var_values[VAR_CONCATDEC_SELECT] = get_concatdec_select(frame, av_rescale_q(frame->pts, inlink->time_base, AV_TIME_BASE_Q));
349 switch (inlink->type) {
350 case AVMEDIA_TYPE_AUDIO:
351 select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
354 case AVMEDIA_TYPE_VIDEO:
355 select->var_values[VAR_INTERLACE_TYPE] =
356 !frame->interlaced_frame ? INTERLACE_TYPE_P :
357 frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
358 select->var_values[VAR_PICT_TYPE] = frame->pict_type;
359 if (select->do_scene_detect) {
361 select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
362 // TODO: document metadata
363 snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
364 av_dict_set(&frame->metadata, "lavfi.scene_score", buf, 0);
369 select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
370 av_log(inlink->dst, AV_LOG_DEBUG,
371 "n:%f pts:%f t:%f key:%d",
372 select->var_values[VAR_N],
373 select->var_values[VAR_PTS],
374 select->var_values[VAR_T],
377 switch (inlink->type) {
378 case AVMEDIA_TYPE_VIDEO:
379 av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
380 (!frame->interlaced_frame) ? 'P' :
381 frame->top_field_first ? 'T' : 'B',
382 av_get_picture_type_char(frame->pict_type),
383 select->var_values[VAR_SCENE]);
385 case AVMEDIA_TYPE_AUDIO:
386 av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%f",
388 select->var_values[VAR_CONSUMED_SAMPLES_N]);
393 select->select_out = -1; /* drop */
394 } else if (isnan(res) || res < 0) {
395 select->select_out = 0; /* first output */
397 select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
400 av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
403 select->var_values[VAR_PREV_SELECTED_N] = select->var_values[VAR_N];
404 select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
405 select->var_values[VAR_PREV_SELECTED_T] = select->var_values[VAR_T];
406 select->var_values[VAR_SELECTED_N] += 1.0;
407 if (inlink->type == AVMEDIA_TYPE_AUDIO)
408 select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
411 select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
412 select->var_values[VAR_PREV_T] = select->var_values[VAR_T];
415 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
417 AVFilterContext *ctx = inlink->dst;
418 SelectContext *select = ctx->priv;
420 select_frame(ctx, frame);
422 return ff_filter_frame(ctx->outputs[select->select_out], frame);
424 av_frame_free(&frame);
428 static int request_frame(AVFilterLink *outlink)
430 AVFilterLink *inlink = outlink->src->inputs[0];
431 int ret = ff_request_frame(inlink);
435 static av_cold void uninit(AVFilterContext *ctx)
437 SelectContext *select = ctx->priv;
440 av_expr_free(select->expr);
443 for (i = 0; i < ctx->nb_outputs; i++)
444 av_freep(&ctx->output_pads[i].name);
446 if (select->do_scene_detect) {
447 av_frame_free(&select->prev_picref);
451 #if CONFIG_ASELECT_FILTER
453 DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
454 AVFILTER_DEFINE_CLASS(aselect);
456 static av_cold int aselect_init(AVFilterContext *ctx)
458 SelectContext *select = ctx->priv;
461 if ((ret = init(ctx)) < 0)
464 if (select->do_scene_detect) {
465 av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
466 return AVERROR(EINVAL);
472 static const AVFilterPad avfilter_af_aselect_inputs[] = {
475 .type = AVMEDIA_TYPE_AUDIO,
476 .config_props = config_input,
477 .filter_frame = filter_frame,
482 AVFilter ff_af_aselect = {
484 .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
485 .init = aselect_init,
487 .priv_size = sizeof(SelectContext),
488 .inputs = avfilter_af_aselect_inputs,
489 .priv_class = &aselect_class,
490 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
492 #endif /* CONFIG_ASELECT_FILTER */
494 #if CONFIG_SELECT_FILTER
496 static int query_formats(AVFilterContext *ctx)
498 SelectContext *select = ctx->priv;
500 if (!select->do_scene_detect) {
501 return ff_default_query_formats(ctx);
504 static const enum AVPixelFormat pix_fmts[] = {
505 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24, AV_PIX_FMT_RGBA,
506 AV_PIX_FMT_ABGR, AV_PIX_FMT_BGRA, AV_PIX_FMT_GRAY8,
507 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
508 AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVJ422P,
509 AV_PIX_FMT_YUV420P10,
512 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
515 return AVERROR(ENOMEM);
516 ret = ff_set_common_formats(ctx, fmts_list);
523 DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
524 AVFILTER_DEFINE_CLASS(select);
526 static av_cold int select_init(AVFilterContext *ctx)
530 if ((ret = init(ctx)) < 0)
536 static const AVFilterPad avfilter_vf_select_inputs[] = {
539 .type = AVMEDIA_TYPE_VIDEO,
540 .config_props = config_input,
541 .filter_frame = filter_frame,
546 AVFilter ff_vf_select = {
548 .description = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
551 .query_formats = query_formats,
552 .priv_size = sizeof(SelectContext),
553 .priv_class = &select_class,
554 .inputs = avfilter_vf_select_inputs,
555 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
557 #endif /* CONFIG_SELECT_FILTER */