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"
38 #include "libavcodec/dsputil.h"
41 static const char *const var_names[] = {
44 "pts", ///< original pts in the file of the frame
45 "start_pts", ///< first PTS in the stream, expressed in TB units
46 "prev_pts", ///< previous frame PTS
47 "prev_selected_pts", ///< previous selected frame PTS
49 "t", ///< first PTS in seconds
50 "start_t", ///< first PTS in the stream, expressed in seconds
51 "prev_t", ///< previous frame time
52 "prev_selected_t", ///< previously selected time
54 "pict_type", ///< the type of picture in the movie
63 "interlace_type", ///< the frame interlace type
68 "consumed_samples_n",///< number of samples consumed by the filter (only audio)
69 "samples_n", ///< number of samples in the current frame (only audio)
70 "sample_rate", ///< sample rate (only audio)
72 "n", ///< frame number (starting from zero)
73 "selected_n", ///< selected frame number (starting from zero)
74 "prev_selected_n", ///< number of the last selected frame
76 "key", ///< tell if the frame is a key frame
77 "pos", ///< original position in the file of the frame
90 VAR_PREV_SELECTED_PTS,
107 VAR_INTERLACE_TYPE_P,
108 VAR_INTERLACE_TYPE_T,
109 VAR_INTERLACE_TYPE_B,
111 VAR_CONSUMED_SAMPLES_N,
128 const AVClass *class;
131 double var_values[VAR_VARS_NB];
132 int do_scene_detect; ///< 1 if the expression requires scene detection variables, 0 otherwise
134 AVCodecContext *avctx; ///< codec context required for the DSPContext (scene detect only)
135 DSPContext c; ///< context providing optimized SAD methods (scene detect only)
136 double prev_mafd; ///< previous MAFD (scene detect only)
138 AVFrame *prev_picref; ///< previous frame (scene detect only)
140 int select_out; ///< mark the selected output pad index
144 #define OFFSET(x) offsetof(SelectContext, x)
145 #define DEFINE_OPTIONS(filt_name, FLAGS) \
146 static const AVOption filt_name##_options[] = { \
147 { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
148 { "e", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
149 { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
150 { "n", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
154 static int request_frame(AVFilterLink *outlink);
156 static av_cold int init(AVFilterContext *ctx)
158 SelectContext *select = ctx->priv;
161 if ((ret = av_expr_parse(&select->expr, select->expr_str,
162 var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
163 av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
167 select->do_scene_detect = !!strstr(select->expr_str, "scene");
169 for (i = 0; i < select->nb_outputs; i++) {
170 AVFilterPad pad = { 0 };
172 pad.name = av_asprintf("output%d", i);
174 return AVERROR(ENOMEM);
175 pad.type = ctx->filter->inputs[0].type;
176 pad.request_frame = request_frame;
177 ff_insert_outpad(ctx, i, &pad);
183 #define INTERLACE_TYPE_P 0
184 #define INTERLACE_TYPE_T 1
185 #define INTERLACE_TYPE_B 2
187 static int config_input(AVFilterLink *inlink)
189 SelectContext *select = inlink->dst->priv;
191 select->var_values[VAR_N] = 0.0;
192 select->var_values[VAR_SELECTED_N] = 0.0;
194 select->var_values[VAR_TB] = av_q2d(inlink->time_base);
196 select->var_values[VAR_PREV_PTS] = NAN;
197 select->var_values[VAR_PREV_SELECTED_PTS] = NAN;
198 select->var_values[VAR_PREV_SELECTED_T] = NAN;
199 select->var_values[VAR_PREV_T] = NAN;
200 select->var_values[VAR_START_PTS] = NAN;
201 select->var_values[VAR_START_T] = NAN;
203 select->var_values[VAR_PICT_TYPE_I] = AV_PICTURE_TYPE_I;
204 select->var_values[VAR_PICT_TYPE_P] = AV_PICTURE_TYPE_P;
205 select->var_values[VAR_PICT_TYPE_B] = AV_PICTURE_TYPE_B;
206 select->var_values[VAR_PICT_TYPE_SI] = AV_PICTURE_TYPE_SI;
207 select->var_values[VAR_PICT_TYPE_SP] = AV_PICTURE_TYPE_SP;
208 select->var_values[VAR_PICT_TYPE_BI] = AV_PICTURE_TYPE_BI;
210 select->var_values[VAR_INTERLACE_TYPE_P] = INTERLACE_TYPE_P;
211 select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T;
212 select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B;
214 select->var_values[VAR_PICT_TYPE] = NAN;
215 select->var_values[VAR_INTERLACE_TYPE] = NAN;
216 select->var_values[VAR_SCENE] = NAN;
217 select->var_values[VAR_CONSUMED_SAMPLES_N] = NAN;
218 select->var_values[VAR_SAMPLES_N] = NAN;
220 select->var_values[VAR_SAMPLE_RATE] =
221 inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
224 if (select->do_scene_detect) {
225 select->avctx = avcodec_alloc_context3(NULL);
227 return AVERROR(ENOMEM);
228 avpriv_dsputil_init(&select->c, select->avctx);
235 static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
238 SelectContext *select = ctx->priv;
239 AVFrame *prev_picref = select->prev_picref;
242 frame->height == prev_picref->height &&
243 frame->width == prev_picref->width &&
244 frame->linesize[0] == prev_picref->linesize[0]) {
245 int x, y, nb_sad = 0;
248 uint8_t *p1 = frame->data[0];
249 uint8_t *p2 = prev_picref->data[0];
250 const int linesize = frame->linesize[0];
252 for (y = 0; y < frame->height - 8; y += 8) {
253 for (x = 0; x < frame->width*3 - 8; x += 8) {
254 sad += select->c.sad[1](select, p1 + x, p2 + x,
262 mafd = nb_sad ? sad / nb_sad : 0;
263 diff = fabs(mafd - select->prev_mafd);
264 ret = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
265 select->prev_mafd = mafd;
266 av_frame_free(&prev_picref);
268 select->prev_picref = av_frame_clone(frame);
273 #define D2TS(d) (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
274 #define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
276 static void select_frame(AVFilterContext *ctx, AVFrame *frame)
278 SelectContext *select = ctx->priv;
279 AVFilterLink *inlink = ctx->inputs[0];
282 if (isnan(select->var_values[VAR_START_PTS]))
283 select->var_values[VAR_START_PTS] = TS2D(frame->pts);
284 if (isnan(select->var_values[VAR_START_T]))
285 select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
287 select->var_values[VAR_N ] = inlink->frame_count;
288 select->var_values[VAR_PTS] = TS2D(frame->pts);
289 select->var_values[VAR_T ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
290 select->var_values[VAR_POS] = av_frame_get_pkt_pos(frame) == -1 ? NAN : av_frame_get_pkt_pos(frame);
292 switch (inlink->type) {
293 case AVMEDIA_TYPE_AUDIO:
294 select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
297 case AVMEDIA_TYPE_VIDEO:
298 select->var_values[VAR_INTERLACE_TYPE] =
299 !frame->interlaced_frame ? INTERLACE_TYPE_P :
300 frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
301 select->var_values[VAR_PICT_TYPE] = frame->pict_type;
303 if (select->do_scene_detect) {
305 select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
306 // TODO: document metadata
307 snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
308 av_dict_set(avpriv_frame_get_metadatap(frame), "lavfi.scene_score", buf, 0);
314 select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
315 av_log(inlink->dst, AV_LOG_DEBUG,
316 "n:%f pts:%f t:%f key:%d",
317 select->var_values[VAR_N],
318 select->var_values[VAR_PTS],
319 select->var_values[VAR_T],
320 (int)select->var_values[VAR_KEY]);
322 switch (inlink->type) {
323 case AVMEDIA_TYPE_VIDEO:
324 av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
325 select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_P ? 'P' :
326 select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_T ? 'T' :
327 select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_B ? 'B' : '?',
328 av_get_picture_type_char(select->var_values[VAR_PICT_TYPE]),
329 select->var_values[VAR_SCENE]);
331 case AVMEDIA_TYPE_AUDIO:
332 av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%d",
333 (int)select->var_values[VAR_SAMPLES_N],
334 (int)select->var_values[VAR_CONSUMED_SAMPLES_N]);
339 select->select_out = -1; /* drop */
340 } else if (isnan(res) || res < 0) {
341 select->select_out = 0; /* first output */
343 select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
346 av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
349 select->var_values[VAR_PREV_SELECTED_N] = select->var_values[VAR_N];
350 select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
351 select->var_values[VAR_PREV_SELECTED_T] = select->var_values[VAR_T];
352 select->var_values[VAR_SELECTED_N] += 1.0;
353 if (inlink->type == AVMEDIA_TYPE_AUDIO)
354 select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
357 select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
358 select->var_values[VAR_PREV_T] = select->var_values[VAR_T];
361 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
363 AVFilterContext *ctx = inlink->dst;
364 SelectContext *select = ctx->priv;
366 select_frame(ctx, frame);
368 return ff_filter_frame(ctx->outputs[select->select_out], frame);
370 av_frame_free(&frame);
374 static int request_frame(AVFilterLink *outlink)
376 AVFilterContext *ctx = outlink->src;
377 SelectContext *select = ctx->priv;
378 AVFilterLink *inlink = outlink->src->inputs[0];
379 int out_no = FF_OUTLINK_IDX(outlink);
382 int ret = ff_request_frame(inlink);
385 } while (select->select_out != out_no);
390 static av_cold void uninit(AVFilterContext *ctx)
392 SelectContext *select = ctx->priv;
395 av_expr_free(select->expr);
398 for (i = 0; i < ctx->nb_outputs; i++)
399 av_freep(&ctx->output_pads[i].name);
402 if (select->do_scene_detect) {
403 av_frame_free(&select->prev_picref);
405 avcodec_close(select->avctx);
406 av_freep(&select->avctx);
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);
419 static const enum AVPixelFormat pix_fmts[] = {
420 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
423 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
428 #if CONFIG_ASELECT_FILTER
430 DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
431 AVFILTER_DEFINE_CLASS(aselect);
433 static av_cold int aselect_init(AVFilterContext *ctx)
435 SelectContext *select = ctx->priv;
438 if ((ret = init(ctx)) < 0)
441 if (select->do_scene_detect) {
442 av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
443 return AVERROR(EINVAL);
449 static const AVFilterPad avfilter_af_aselect_inputs[] = {
452 .type = AVMEDIA_TYPE_AUDIO,
453 .get_audio_buffer = ff_null_get_audio_buffer,
454 .config_props = config_input,
455 .filter_frame = filter_frame,
460 AVFilter avfilter_af_aselect = {
462 .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
463 .init = aselect_init,
465 .priv_size = sizeof(SelectContext),
466 .inputs = avfilter_af_aselect_inputs,
467 .priv_class = &aselect_class,
468 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
470 #endif /* CONFIG_ASELECT_FILTER */
472 #if CONFIG_SELECT_FILTER
474 DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
475 AVFILTER_DEFINE_CLASS(select);
477 static av_cold int select_init(AVFilterContext *ctx)
479 SelectContext *select = ctx->priv;
482 if ((ret = init(ctx)) < 0)
485 if (select->do_scene_detect && !CONFIG_AVCODEC) {
486 av_log(ctx, AV_LOG_ERROR, "Scene detection is not available without libavcodec.\n");
487 return AVERROR(EINVAL);
493 static const AVFilterPad avfilter_vf_select_inputs[] = {
496 .type = AVMEDIA_TYPE_VIDEO,
497 .get_video_buffer = ff_null_get_video_buffer,
498 .config_props = config_input,
499 .filter_frame = filter_frame,
504 AVFilter avfilter_vf_select = {
506 .description = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
509 .query_formats = query_formats,
511 .priv_size = sizeof(SelectContext),
512 .priv_class = &select_class,
514 .inputs = avfilter_vf_select_inputs,
515 .flags = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
517 #endif /* CONFIG_SELECT_FILTER */