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 "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 "avfilter.h"
32#include "audio.h"
33#include "formats.h"
34#include "internal.h"
35#include "video.h"
36
37#if CONFIG_AVCODEC
38#include "libavcodec/dsputil.h"
39#endif
40
41static const char *const var_names[] = {
42    "TB",                ///< timebase
43
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
48
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
53
54    "pict_type",         ///< the type of picture in the movie
55    "I",
56    "P",
57    "B",
58    "S",
59    "SI",
60    "SP",
61    "BI",
62    "PICT_TYPE_I",
63    "PICT_TYPE_P",
64    "PICT_TYPE_B",
65    "PICT_TYPE_S",
66    "PICT_TYPE_SI",
67    "PICT_TYPE_SP",
68    "PICT_TYPE_BI",
69
70    "interlace_type",    ///< the frame interlace type
71    "PROGRESSIVE",
72    "TOPFIRST",
73    "BOTTOMFIRST",
74
75    "consumed_samples_n",///< number of samples consumed by the filter (only audio)
76    "samples_n",         ///< number of samples in the current frame (only audio)
77    "sample_rate",       ///< sample rate (only audio)
78
79    "n",                 ///< frame number (starting from zero)
80    "selected_n",        ///< selected frame number (starting from zero)
81    "prev_selected_n",   ///< number of the last selected frame
82
83    "key",               ///< tell if the frame is a key frame
84    "pos",               ///< original position in the file of the frame
85
86    "scene",
87
88    NULL
89};
90
91enum var_name {
92    VAR_TB,
93
94    VAR_PTS,
95    VAR_START_PTS,
96    VAR_PREV_PTS,
97    VAR_PREV_SELECTED_PTS,
98
99    VAR_T,
100    VAR_START_T,
101    VAR_PREV_T,
102    VAR_PREV_SELECTED_T,
103
104    VAR_PICT_TYPE,
105    VAR_I,
106    VAR_P,
107    VAR_B,
108    VAR_S,
109    VAR_SI,
110    VAR_SP,
111    VAR_BI,
112    VAR_PICT_TYPE_I,
113    VAR_PICT_TYPE_P,
114    VAR_PICT_TYPE_B,
115    VAR_PICT_TYPE_S,
116    VAR_PICT_TYPE_SI,
117    VAR_PICT_TYPE_SP,
118    VAR_PICT_TYPE_BI,
119
120    VAR_INTERLACE_TYPE,
121    VAR_INTERLACE_TYPE_P,
122    VAR_INTERLACE_TYPE_T,
123    VAR_INTERLACE_TYPE_B,
124
125    VAR_CONSUMED_SAMPLES_N,
126    VAR_SAMPLES_N,
127    VAR_SAMPLE_RATE,
128
129    VAR_N,
130    VAR_SELECTED_N,
131    VAR_PREV_SELECTED_N,
132
133    VAR_KEY,
134    VAR_POS,
135
136    VAR_SCENE,
137
138    VAR_VARS_NB
139};
140
141typedef struct SelectContext {
142    const AVClass *class;
143    char *expr_str;
144    AVExpr *expr;
145    double var_values[VAR_VARS_NB];
146    int do_scene_detect;            ///< 1 if the expression requires scene detection variables, 0 otherwise
147#if CONFIG_AVCODEC
148    AVCodecContext *avctx;          ///< codec context required for the DSPContext (scene detect only)
149    DSPContext c;                   ///< context providing optimized SAD methods   (scene detect only)
150    double prev_mafd;               ///< previous MAFD                             (scene detect only)
151#endif
152    AVFrame *prev_picref; ///< previous frame                            (scene detect only)
153    double select;
154    int select_out;                 ///< mark the selected output pad index
155    int nb_outputs;
156} SelectContext;
157
158#define OFFSET(x) offsetof(SelectContext, x)
159#define DEFINE_OPTIONS(filt_name, FLAGS)                            \
160static const AVOption filt_name##_options[] = {                     \
161    { "expr", "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
162    { "e",    "set an expression to use for selecting frames", OFFSET(expr_str), AV_OPT_TYPE_STRING, { .str = "1" }, .flags=FLAGS }, \
163    { "outputs", "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
164    { "n",       "set the number of outputs", OFFSET(nb_outputs), AV_OPT_TYPE_INT, {.i64 = 1}, 1, INT_MAX, .flags=FLAGS }, \
165    { NULL }                                                            \
166}
167
168static int request_frame(AVFilterLink *outlink);
169
170static av_cold int init(AVFilterContext *ctx)
171{
172    SelectContext *select = ctx->priv;
173    int i, ret;
174
175    if ((ret = av_expr_parse(&select->expr, select->expr_str,
176                             var_names, NULL, NULL, NULL, NULL, 0, ctx)) < 0) {
177        av_log(ctx, AV_LOG_ERROR, "Error while parsing expression '%s'\n",
178               select->expr_str);
179        return ret;
180    }
181    select->do_scene_detect = !!strstr(select->expr_str, "scene");
182
183    for (i = 0; i < select->nb_outputs; i++) {
184        AVFilterPad pad = { 0 };
185
186        pad.name = av_asprintf("output%d", i);
187        if (!pad.name)
188            return AVERROR(ENOMEM);
189        pad.type = ctx->filter->inputs[0].type;
190        pad.request_frame = request_frame;
191        ff_insert_outpad(ctx, i, &pad);
192    }
193
194    return 0;
195}
196
197#define INTERLACE_TYPE_P 0
198#define INTERLACE_TYPE_T 1
199#define INTERLACE_TYPE_B 2
200
201static int config_input(AVFilterLink *inlink)
202{
203    SelectContext *select = inlink->dst->priv;
204
205    select->var_values[VAR_N]          = 0.0;
206    select->var_values[VAR_SELECTED_N] = 0.0;
207
208    select->var_values[VAR_TB] = av_q2d(inlink->time_base);
209
210    select->var_values[VAR_PREV_PTS]          = NAN;
211    select->var_values[VAR_PREV_SELECTED_PTS] = NAN;
212    select->var_values[VAR_PREV_SELECTED_T]   = NAN;
213    select->var_values[VAR_PREV_T]            = NAN;
214    select->var_values[VAR_START_PTS]         = NAN;
215    select->var_values[VAR_START_T]           = NAN;
216
217    select->var_values[VAR_I]  = AV_PICTURE_TYPE_I;
218    select->var_values[VAR_P]  = AV_PICTURE_TYPE_P;
219    select->var_values[VAR_B]  = AV_PICTURE_TYPE_B;
220    select->var_values[VAR_SI] = AV_PICTURE_TYPE_SI;
221    select->var_values[VAR_SP] = AV_PICTURE_TYPE_SP;
222    select->var_values[VAR_BI] = AV_PICTURE_TYPE_BI;
223    select->var_values[VAR_PICT_TYPE_I]  = AV_PICTURE_TYPE_I;
224    select->var_values[VAR_PICT_TYPE_P]  = AV_PICTURE_TYPE_P;
225    select->var_values[VAR_PICT_TYPE_B]  = AV_PICTURE_TYPE_B;
226    select->var_values[VAR_PICT_TYPE_SI] = AV_PICTURE_TYPE_SI;
227    select->var_values[VAR_PICT_TYPE_SP] = AV_PICTURE_TYPE_SP;
228    select->var_values[VAR_PICT_TYPE_BI] = AV_PICTURE_TYPE_BI;
229
230    select->var_values[VAR_INTERLACE_TYPE_P] = INTERLACE_TYPE_P;
231    select->var_values[VAR_INTERLACE_TYPE_T] = INTERLACE_TYPE_T;
232    select->var_values[VAR_INTERLACE_TYPE_B] = INTERLACE_TYPE_B;
233
234    select->var_values[VAR_PICT_TYPE]         = NAN;
235    select->var_values[VAR_INTERLACE_TYPE]    = NAN;
236    select->var_values[VAR_SCENE]             = NAN;
237    select->var_values[VAR_CONSUMED_SAMPLES_N] = NAN;
238    select->var_values[VAR_SAMPLES_N]          = NAN;
239
240    select->var_values[VAR_SAMPLE_RATE] =
241        inlink->type == AVMEDIA_TYPE_AUDIO ? inlink->sample_rate : NAN;
242
243#if CONFIG_AVCODEC
244    if (select->do_scene_detect) {
245        select->avctx = avcodec_alloc_context3(NULL);
246        if (!select->avctx)
247            return AVERROR(ENOMEM);
248        avpriv_dsputil_init(&select->c, select->avctx);
249    }
250#endif
251    return 0;
252}
253
254#if CONFIG_AVCODEC
255static double get_scene_score(AVFilterContext *ctx, AVFrame *frame)
256{
257    double ret = 0;
258    SelectContext *select = ctx->priv;
259    AVFrame *prev_picref = select->prev_picref;
260
261    if (prev_picref &&
262        frame->height    == prev_picref->height &&
263        frame->width    == prev_picref->width &&
264        frame->linesize[0] == prev_picref->linesize[0]) {
265        int x, y, nb_sad = 0;
266        int64_t sad = 0;
267        double mafd, diff;
268        uint8_t *p1 =      frame->data[0];
269        uint8_t *p2 = prev_picref->data[0];
270        const int linesize = frame->linesize[0];
271
272        for (y = 0; y < frame->height - 8; y += 8) {
273            for (x = 0; x < frame->width*3 - 8; x += 8) {
274                sad += select->c.sad[1](NULL, p1 + x, p2 + x,
275                                        linesize, 8);
276                nb_sad += 8 * 8;
277            }
278            p1 += 8 * linesize;
279            p2 += 8 * linesize;
280        }
281        emms_c();
282        mafd = nb_sad ? (double)sad / nb_sad : 0;
283        diff = fabs(mafd - select->prev_mafd);
284        ret  = av_clipf(FFMIN(mafd, diff) / 100., 0, 1);
285        select->prev_mafd = mafd;
286        av_frame_free(&prev_picref);
287    }
288    select->prev_picref = av_frame_clone(frame);
289    return ret;
290}
291#endif
292
293#define D2TS(d)  (isnan(d) ? AV_NOPTS_VALUE : (int64_t)(d))
294#define TS2D(ts) ((ts) == AV_NOPTS_VALUE ? NAN : (double)(ts))
295
296static void select_frame(AVFilterContext *ctx, AVFrame *frame)
297{
298    SelectContext *select = ctx->priv;
299    AVFilterLink *inlink = ctx->inputs[0];
300    double res;
301
302    if (isnan(select->var_values[VAR_START_PTS]))
303        select->var_values[VAR_START_PTS] = TS2D(frame->pts);
304    if (isnan(select->var_values[VAR_START_T]))
305        select->var_values[VAR_START_T] = TS2D(frame->pts) * av_q2d(inlink->time_base);
306
307    select->var_values[VAR_N  ] = inlink->frame_count;
308    select->var_values[VAR_PTS] = TS2D(frame->pts);
309    select->var_values[VAR_T  ] = TS2D(frame->pts) * av_q2d(inlink->time_base);
310    select->var_values[VAR_POS] = av_frame_get_pkt_pos(frame) == -1 ? NAN : av_frame_get_pkt_pos(frame);
311
312    switch (inlink->type) {
313    case AVMEDIA_TYPE_AUDIO:
314        select->var_values[VAR_SAMPLES_N] = frame->nb_samples;
315        break;
316
317    case AVMEDIA_TYPE_VIDEO:
318        select->var_values[VAR_INTERLACE_TYPE] =
319            !frame->interlaced_frame ? INTERLACE_TYPE_P :
320        frame->top_field_first ? INTERLACE_TYPE_T : INTERLACE_TYPE_B;
321        select->var_values[VAR_PICT_TYPE] = frame->pict_type;
322#if CONFIG_AVCODEC
323        if (select->do_scene_detect) {
324            char buf[32];
325            select->var_values[VAR_SCENE] = get_scene_score(ctx, frame);
326            // TODO: document metadata
327            snprintf(buf, sizeof(buf), "%f", select->var_values[VAR_SCENE]);
328            av_dict_set(avpriv_frame_get_metadatap(frame), "lavfi.scene_score", buf, 0);
329        }
330#endif
331        break;
332    }
333
334    select->select = res = av_expr_eval(select->expr, select->var_values, NULL);
335    av_log(inlink->dst, AV_LOG_DEBUG,
336           "n:%f pts:%f t:%f key:%d",
337           select->var_values[VAR_N],
338           select->var_values[VAR_PTS],
339           select->var_values[VAR_T],
340           (int)select->var_values[VAR_KEY]);
341
342    switch (inlink->type) {
343    case AVMEDIA_TYPE_VIDEO:
344        av_log(inlink->dst, AV_LOG_DEBUG, " interlace_type:%c pict_type:%c scene:%f",
345               select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_P ? 'P' :
346               select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_T ? 'T' :
347               select->var_values[VAR_INTERLACE_TYPE] == INTERLACE_TYPE_B ? 'B' : '?',
348               av_get_picture_type_char(select->var_values[VAR_PICT_TYPE]),
349               select->var_values[VAR_SCENE]);
350        break;
351    case AVMEDIA_TYPE_AUDIO:
352        av_log(inlink->dst, AV_LOG_DEBUG, " samples_n:%d consumed_samples_n:%d",
353               (int)select->var_values[VAR_SAMPLES_N],
354               (int)select->var_values[VAR_CONSUMED_SAMPLES_N]);
355        break;
356    }
357
358    if (res == 0) {
359        select->select_out = -1; /* drop */
360    } else if (isnan(res) || res < 0) {
361        select->select_out = 0; /* first output */
362    } else {
363        select->select_out = FFMIN(ceilf(res)-1, select->nb_outputs-1); /* other outputs */
364    }
365
366    av_log(inlink->dst, AV_LOG_DEBUG, " -> select:%f select_out:%d\n", res, select->select_out);
367
368    if (res) {
369        select->var_values[VAR_PREV_SELECTED_N]   = select->var_values[VAR_N];
370        select->var_values[VAR_PREV_SELECTED_PTS] = select->var_values[VAR_PTS];
371        select->var_values[VAR_PREV_SELECTED_T]   = select->var_values[VAR_T];
372        select->var_values[VAR_SELECTED_N] += 1.0;
373        if (inlink->type == AVMEDIA_TYPE_AUDIO)
374            select->var_values[VAR_CONSUMED_SAMPLES_N] += frame->nb_samples;
375    }
376
377    select->var_values[VAR_PREV_PTS] = select->var_values[VAR_PTS];
378    select->var_values[VAR_PREV_T]   = select->var_values[VAR_T];
379}
380
381static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
382{
383    AVFilterContext *ctx = inlink->dst;
384    SelectContext *select = ctx->priv;
385
386    select_frame(ctx, frame);
387    if (select->select)
388        return ff_filter_frame(ctx->outputs[select->select_out], frame);
389
390    av_frame_free(&frame);
391    return 0;
392}
393
394static int request_frame(AVFilterLink *outlink)
395{
396    AVFilterContext *ctx = outlink->src;
397    SelectContext *select = ctx->priv;
398    AVFilterLink *inlink = outlink->src->inputs[0];
399    int out_no = FF_OUTLINK_IDX(outlink);
400
401    do {
402        int ret = ff_request_frame(inlink);
403        if (ret < 0)
404            return ret;
405    } while (select->select_out != out_no);
406
407    return 0;
408}
409
410static av_cold void uninit(AVFilterContext *ctx)
411{
412    SelectContext *select = ctx->priv;
413    int i;
414
415    av_expr_free(select->expr);
416    select->expr = NULL;
417
418    for (i = 0; i < ctx->nb_outputs; i++)
419        av_freep(&ctx->output_pads[i].name);
420
421#if CONFIG_AVCODEC
422    if (select->do_scene_detect) {
423        av_frame_free(&select->prev_picref);
424        if (select->avctx) {
425            avcodec_close(select->avctx);
426            av_freep(&select->avctx);
427        }
428    }
429#endif
430}
431
432static int query_formats(AVFilterContext *ctx)
433{
434    SelectContext *select = ctx->priv;
435
436    if (!select->do_scene_detect) {
437        return ff_default_query_formats(ctx);
438    } else {
439        static const enum AVPixelFormat pix_fmts[] = {
440            AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
441            AV_PIX_FMT_NONE
442        };
443        ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
444    }
445    return 0;
446}
447
448#if CONFIG_ASELECT_FILTER
449
450DEFINE_OPTIONS(aselect, AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
451AVFILTER_DEFINE_CLASS(aselect);
452
453static av_cold int aselect_init(AVFilterContext *ctx)
454{
455    SelectContext *select = ctx->priv;
456    int ret;
457
458    if ((ret = init(ctx)) < 0)
459        return ret;
460
461    if (select->do_scene_detect) {
462        av_log(ctx, AV_LOG_ERROR, "Scene detection is ignored in aselect filter\n");
463        return AVERROR(EINVAL);
464    }
465
466    return 0;
467}
468
469static const AVFilterPad avfilter_af_aselect_inputs[] = {
470    {
471        .name         = "default",
472        .type         = AVMEDIA_TYPE_AUDIO,
473        .config_props = config_input,
474        .filter_frame = filter_frame,
475    },
476    { NULL }
477};
478
479AVFilter ff_af_aselect = {
480    .name        = "aselect",
481    .description = NULL_IF_CONFIG_SMALL("Select audio frames to pass in output."),
482    .init        = aselect_init,
483    .uninit      = uninit,
484    .priv_size   = sizeof(SelectContext),
485    .inputs      = avfilter_af_aselect_inputs,
486    .priv_class  = &aselect_class,
487    .flags       = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
488};
489#endif /* CONFIG_ASELECT_FILTER */
490
491#if CONFIG_SELECT_FILTER
492
493DEFINE_OPTIONS(select, AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM);
494AVFILTER_DEFINE_CLASS(select);
495
496static av_cold int select_init(AVFilterContext *ctx)
497{
498    SelectContext *select = ctx->priv;
499    int ret;
500
501    if ((ret = init(ctx)) < 0)
502        return ret;
503
504    if (select->do_scene_detect && !CONFIG_AVCODEC) {
505        av_log(ctx, AV_LOG_ERROR, "Scene detection is not available without libavcodec.\n");
506        return AVERROR(EINVAL);
507    }
508
509    return 0;
510}
511
512static const AVFilterPad avfilter_vf_select_inputs[] = {
513    {
514        .name         = "default",
515        .type         = AVMEDIA_TYPE_VIDEO,
516        .config_props = config_input,
517        .filter_frame = filter_frame,
518    },
519    { NULL }
520};
521
522AVFilter ff_vf_select = {
523    .name          = "select",
524    .description   = NULL_IF_CONFIG_SMALL("Select video frames to pass in output."),
525    .init          = select_init,
526    .uninit        = uninit,
527    .query_formats = query_formats,
528    .priv_size     = sizeof(SelectContext),
529    .priv_class    = &select_class,
530    .inputs        = avfilter_vf_select_inputs,
531    .flags         = AVFILTER_FLAG_DYNAMIC_OUTPUTS,
532};
533#endif /* CONFIG_SELECT_FILTER */
534