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"
31 #include "libavutil/pixelutils.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
94 VAR_PREV_SELECTED_PTS,
118 VAR_INTERLACE_TYPE_P,
119 VAR_INTERLACE_TYPE_T,
120 VAR_INTERLACE_TYPE_B,
122 VAR_CONSUMED_SAMPLES_N,
138 typedef struct SelectContext {
139 const AVClass *class;
142 double var_values[VAR_VARS_NB];
143 int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise
144 av_pixelutils_sad_fn sad; ///< Sum of the absolute difference function (scene detect only)
145 double prev_mafd; ///< previous MAFD (scene detect only)
146 AVFrame *prev_picref; ///< previous frame (scene detect only)
148 int select_out; ///< mark the selected output pad index
152 #define OFFSET(x) offsetof(SelectContext, x)
153 #define DEFINE_OPTIONS(filt_name, FLAGS) \
154 static const AVOption filt_name##_options[] = { \
155 { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
156 { "e", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
157 { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
158 { "n", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
162 static int request_frame(AVFilterLink *outlink);
164 static av_cold int init(AVFilterContext *ctx)
166 SelectContext *select = ctx->priv;
169 if ((ret = av_expr_parse(&select->expr, select->expr_str,
170 var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
171 av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
175 select->do_scene_detect = !!strstr(select->expr_str, "scene");
177 for (i = 0; i < select->nb_outputs; i++) {
178 AVFilterPad pad = { 0 };
180 pad.name = av_asprintf("output%d", i);
182 return AVERROR(ENOMEM);
183 pad.type = ctx->filter->inputs[0].type;
184 pad.request_frame = request_frame;
185 ff_insert_outpad(ctx, i, &pad);
191 #define INTERLACE_TYPE_P 0
192 #define INTERLACE_TYPE_T 1
193 #define INTERLACE_TYPE_B 2
195 static int config_input(AVFilterLink *inlink)
197 SelectContext *select = inlink->dst->priv;
199 select->var_values[VAR_N] = 0.0;
200 select->var_values[VAR_SELECTED_N] = 0.0;
202 select->var_values[VAR_TB] = av_q2d(inlink->time_base);
204 select->var_values[VAR_PREV_PTS] = NAN;
205 select->var_values[VAR_PREV_SELECTED_PTS] = NAN;
206 select->var_values[VAR_PREV_SELECTED_T] = NAN;
207 select->var_values[VAR_PREV_T] = NAN;
208 select->var_values[VAR_START_PTS] = NAN;
209 select->var_values[VAR_START_T] = NAN;
211 select->var_values[VAR_I] = AV_PICTURE_TYPE_I;
212 select->var_values[VAR_P] = AV_PICTURE_TYPE_P;
213 select->var_values[VAR_B] = AV_PICTURE_TYPE_B;
214 select->var_values[VAR_SI] = AV_PICTURE_TYPE_SI;
215 select->var_values[VAR_SP] = AV_PICTURE_TYPE_SP;
216 select->var_values[VAR_BI] = AV_PICTURE_TYPE_BI;
217 select->var_values[VAR_PICT_TYPE_I] = AV_PICTURE_TYPE_I;
218 select->var_values[VAR_PICT_TYPE_P] = AV_PICTURE_TYPE_P;
219 select->var_values[VAR_PICT_TYPE_B] = AV_PICTURE_TYPE_B;
220 select->var_values[VAR_PICT_TYPE_SI] = AV_PICTURE_TYPE_SI;
221 select->var_values[VAR_PICT_TYPE_SP] = AV_PICTURE_TYPE_SP;
222 select->var_values[VAR_PICT_TYPE_BI] = AV_PICTURE_TYPE_BI;
224 select->var_values[VAR_INTERLACE_TYPE_P] = INTERLACE_TYPE_P;
225 select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T;
226 select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B;
228 select->var_values[VAR_PICT_TYPE] = NAN;
229 select->var_values[VAR_INTERLACE_TYPE] = NAN;
230 select->var_values[VAR_SCENE] = NAN;
231 select->var_values[VAR_CONSUMED_SAMPLES_N] = NAN;
232 select->var_values[VAR_SAMPLES_N] = NAN;
234 select->var_values[VAR_SAMPLE_RATE] =
235 inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
237 if (select->do_scene_detect) {
238 select->sad = av_pixelutils_get_sad_fn(3, 3, 2, select); // 8x8 both sources aligned
240 return AVERROR(EINVAL);
245 static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
248 SelectContext *select = ctx->priv;
249 AVFrame *prev_picref = select->prev_picref;
252 frame->height == prev_picref->height &&
253 frame->width == prev_picref->width) {
254 int x, y, nb_sad = 0;
257 uint8_t *p1 = frame->data[0];
258 uint8_t *p2 = prev_picref->data[0];
259 const int p1_linesize = frame->linesize[0];
260 const int p2_linesize = prev_picref->linesize[0];
262 for (y = 0; y < frame->height - 7; y += 8) {
263 for (x = 0; x < frame->width*3 - 7; x += 8) {
264 sad += select->sad(p1 + x, p1_linesize, p2 + x, p2_linesize);
267 p1 += 8 * p1_linesize;
268 p2 += 8 * p2_linesize;
271 mafd = nb_sad ? (double)sad / nb_sad : 0;
272 diff = fabs(mafd - select->prev_mafd);
273 ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
274 select->prev_mafd = mafd;
275 av_frame_free(&prev_picref);
277 select->prev_picref = av_frame_clone(frame);
281 #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
282 #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
284 static void select_frame(AVFilterContext *ctx, AVFrame *frame)
286 SelectContext *select = ctx->priv;
287 AVFilterLink *inlink = ctx->inputs[0];
290 if (isnan(select->var_values[VAR_START_PTS]))
291 select->var_values[VAR_START_PTS] = TS2D(frame->pts);
292 if (isnan(select->var_values[VAR_START_T]))
293 select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
295 select->var_values[VAR_N ] = inlink->frame_count;
296 select->var_values[VAR_PTS] = TS2D(frame->pts);
297 select->var_values[VAR_T ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
298 select->var_values[VAR_POS] = av_frame_get_pkt_pos(frame) == -1 ? NAN : av_frame_get_pkt_pos(frame);
299 select->var_values[VAR_KEY] = frame->key_frame;
301 switch (inlink->type) {
302 case AVMEDIA_TYPE_AUDIO:
303 select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
306 case AVMEDIA_TYPE_VIDEO:
307 select->var_values[VAR_INTERLACE_TYPE] =
308 !frame->interlaced_frame ? INTERLACE_TYPE_P :
309 frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
310 select->var_values[VAR_PICT_TYPE] = frame->pict_type;
311 if (select->do_scene_detect) {
313 select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
314 // TODO: document metadata
315 snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
316 av_dict_set(avpriv_frame_get_metadatap(frame), "lavfi.scene_score", buf, 0);
321 select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
322 av_log(inlink->dst, AV_LOG_DEBUG,
323 "n:%f pts:%f t:%f key:%d",
324 select->var_values[VAR_N],
325 select->var_values[VAR_PTS],
326 select->var_values[VAR_T],
329 switch (inlink->type) {
330 case AVMEDIA_TYPE_VIDEO:
331 av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
332 (!frame->interlaced_frame) ? 'P' :
333 frame->top_field_first ? 'T' : 'B',
334 av_get_picture_type_char(frame->pict_type),
335 select->var_values[VAR_SCENE]);
337 case AVMEDIA_TYPE_AUDIO:
338 av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%f",
340 select->var_values[VAR_CONSUMED_SAMPLES_N]);
345 select->select_out = -1; /* drop */
346 } else if (isnan(res) || res < 0) {
347 select->select_out = 0; /* first output */
349 select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
352 av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
355 select->var_values[VAR_PREV_SELECTED_N] = select->var_values[VAR_N];
356 select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
357 select->var_values[VAR_PREV_SELECTED_T] = select->var_values[VAR_T];
358 select->var_values[VAR_SELECTED_N] += 1.0;
359 if (inlink->type == AVMEDIA_TYPE_AUDIO)
360 select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
363 select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
364 select->var_values[VAR_PREV_T] = select->var_values[VAR_T];
367 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
369 AVFilterContext *ctx = inlink->dst;
370 SelectContext *select = ctx->priv;
372 select_frame(ctx, frame);
374 return ff_filter_frame(ctx->outputs[select->select_out], frame);
376 av_frame_free(&frame);
380 static int request_frame(AVFilterLink *outlink)
382 AVFilterContext *ctx = outlink->src;
383 SelectContext *select = ctx->priv;
384 AVFilterLink *inlink = outlink->src->inputs[0];
385 int out_no = FF_OUTLINK_IDX(outlink);
388 int ret = ff_request_frame(inlink);
391 } while (select->select_out != out_no);
396 static av_cold void uninit(AVFilterContext *ctx)
398 SelectContext *select = ctx->priv;
401 av_expr_free(select->expr);
404 for (i = 0; i < ctx->nb_outputs; i++)
405 av_freep(&ctx->output_pads[i].name);
407 if (select->do_scene_detect) {
408 av_frame_free(&select->prev_picref);
412 static int query_formats(AVFilterContext *ctx)
414 SelectContext *select = ctx->priv;
416 if (!select->do_scene_detect) {
417 return ff_default_query_formats(ctx);
420 static const enum AVPixelFormat pix_fmts[] = {
421 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
424 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
427 return AVERROR(ENOMEM);
428 ret = ff_set_common_formats(ctx, fmts_list);
435 #if CONFIG_ASELECT_FILTER
437 DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
438 AVFILTER_DEFINE_CLASS(aselect);
440 static av_cold int aselect_init(AVFilterContext *ctx)
442 SelectContext *select = ctx->priv;
445 if ((ret = init(ctx)) < 0)
448 if (select->do_scene_detect) {
449 av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
450 return AVERROR(EINVAL);
456 static const AVFilterPad avfilter_af_aselect_inputs[] = {
459 .type = AVMEDIA_TYPE_AUDIO,
460 .config_props = config_input,
461 .filter_frame = filter_frame,
466 AVFilter ff_af_aselect = {
468 .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
469 .init = aselect_init,
471 .priv_size = sizeof(SelectContext),
472 .inputs = avfilter_af_aselect_inputs,
473 .priv_class = &aselect_class,
474 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
476 #endif /* CONFIG_ASELECT_FILTER */
478 #if CONFIG_SELECT_FILTER
480 DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
481 AVFILTER_DEFINE_CLASS(select);
483 static av_cold int select_init(AVFilterContext *ctx)
487 if ((ret = init(ctx)) < 0)
493 static const AVFilterPad avfilter_vf_select_inputs[] = {
496 .type = AVMEDIA_TYPE_VIDEO,
497 .config_props = config_input,
498 .filter_frame = filter_frame,
503 AVFilter ff_vf_select = {
505 .description = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
508 .query_formats = query_formats,
509 .priv_size = sizeof(SelectContext),
510 .priv_class = &select_class,
511 .inputs = avfilter_vf_select_inputs,
512 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
514 #endif /* CONFIG_SELECT_FILTER */