2 * Copyright (C) 2012 Mark Himsley
4 * get_scene_score() Copyright (c) 2011 Stefano Sabatini
5 * taken from libavfilter/vf_select.c
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * filter for upsampling or downsampling a progressive source
31 #include "libavutil/avassert.h"
32 #include "libavutil/imgutils.h"
33 #include "libavutil/internal.h"
34 #include "libavutil/opt.h"
35 #include "libavutil/pixdesc.h"
41 #include "framerate.h"
42 #include "scene_sad.h"
44 #define OFFSET(x) offsetof(FrameRateContext, x)
45 #define V AV_OPT_FLAG_VIDEO_PARAM
46 #define F AV_OPT_FLAG_FILTERING_PARAM
47 #define FRAMERATE_FLAG_SCD 01
49 static const AVOption framerate_options[] = {
50 {"fps", "required output frames per second rate", OFFSET(dest_frame_rate), AV_OPT_TYPE_VIDEO_RATE, {.str="50"}, 0, INT_MAX, V|F },
52 {"interp_start", "point to start linear interpolation", OFFSET(interp_start), AV_OPT_TYPE_INT, {.i64=15}, 0, 255, V|F },
53 {"interp_end", "point to end linear interpolation", OFFSET(interp_end), AV_OPT_TYPE_INT, {.i64=240}, 0, 255, V|F },
54 {"scene", "scene change level", OFFSET(scene_score), AV_OPT_TYPE_DOUBLE, {.dbl=8.2}, 0, 100., V|F },
56 {"flags", "set flags", OFFSET(flags), AV_OPT_TYPE_FLAGS, {.i64=1}, 0, INT_MAX, V|F, "flags" },
57 {"scene_change_detect", "enable scene change detection", 0, AV_OPT_TYPE_CONST, {.i64=FRAMERATE_FLAG_SCD}, INT_MIN, INT_MAX, V|F, "flags" },
58 {"scd", "enable scene change detection", 0, AV_OPT_TYPE_CONST, {.i64=FRAMERATE_FLAG_SCD}, INT_MIN, INT_MAX, V|F, "flags" },
63 AVFILTER_DEFINE_CLASS(framerate);
65 static double get_scene_score(AVFilterContext *ctx, AVFrame *crnt, AVFrame *next)
67 FrameRateContext *s = ctx->priv;
70 ff_dlog(ctx, "get_scene_score()\n");
72 if (crnt->height == next->height &&
73 crnt->width == next->width) {
74 AVDictionaryEntry *e_mafd = NULL;
76 double mafd = HUGE_VAL, diff;
79 ff_dlog(ctx, "get_scene_score() process\n");
80 e_mafd = av_dict_get(next->metadata, "lavfi.scd.mafd", NULL, AV_DICT_MATCH_CASE);
82 mafd = strtod(e_mafd->value, &tail);
83 if (*tail || mafd == HUGE_VAL) {
84 s->sad(crnt->data[0], crnt->linesize[0], next->data[0], next->linesize[0], crnt->width, crnt->height, &sad);
86 mafd = (double)sad * 100.0 / (crnt->width * crnt->height) / (1 << s->bitdepth);
88 diff = fabs(mafd - s->prev_mafd);
89 ret = av_clipf(FFMIN(mafd, diff), 0, 100.0);
92 ff_dlog(ctx, "get_scene_score() result is:%f\n", ret);
96 typedef struct ThreadData {
97 AVFrame *copy_src1, *copy_src2;
98 uint16_t src1_factor, src2_factor;
101 static int filter_slice(AVFilterContext *ctx, void *arg, int job, int nb_jobs)
103 FrameRateContext *s = ctx->priv;
104 ThreadData *td = arg;
105 AVFrame *work = s->work;
106 AVFrame *src1 = td->copy_src1;
107 AVFrame *src2 = td->copy_src2;
108 uint16_t src1_factor = td->src1_factor;
109 uint16_t src2_factor = td->src2_factor;
112 for (plane = 0; plane < 4 && src1->data[plane] && src2->data[plane]; plane++) {
113 const int start = (s->height[plane] * job ) / nb_jobs;
114 const int end = (s->height[plane] * (job+1)) / nb_jobs;
115 uint8_t *src1_data = src1->data[plane] + start * src1->linesize[plane];
116 uint8_t *src2_data = src2->data[plane] + start * src2->linesize[plane];
117 uint8_t *dst_data = work->data[plane] + start * work->linesize[plane];
119 s->blend(src1_data, src1->linesize[plane], src2_data, src2->linesize[plane],
120 dst_data, work->linesize[plane], s->line_size[plane], end - start,
121 src1_factor, src2_factor, s->blend_factor_max >> 1);
127 static int blend_frames(AVFilterContext *ctx, int interpolate)
129 FrameRateContext *s = ctx->priv;
130 AVFilterLink *outlink = ctx->outputs[0];
131 double interpolate_scene_score = 0;
133 if ((s->flags & FRAMERATE_FLAG_SCD)) {
135 interpolate_scene_score = s->score;
137 interpolate_scene_score = s->score = get_scene_score(ctx, s->f0, s->f1);
138 ff_dlog(ctx, "blend_frames() interpolate scene score:%f\n", interpolate_scene_score);
140 // decide if the shot-change detection allows us to blend two frames
141 if (interpolate_scene_score < s->scene_score) {
143 td.copy_src1 = s->f0;
144 td.copy_src2 = s->f1;
145 td.src2_factor = interpolate;
146 td.src1_factor = s->blend_factor_max - td.src2_factor;
148 // get work-space for output frame
149 s->work = ff_get_video_buffer(outlink, outlink->w, outlink->h);
151 return AVERROR(ENOMEM);
153 av_frame_copy_props(s->work, s->f0);
155 ff_dlog(ctx, "blend_frames() INTERPOLATE to create work frame\n");
156 ctx->internal->execute(ctx, filter_slice, &td, NULL, FFMIN(FFMAX(1, outlink->h >> 2), ff_filter_get_nb_threads(ctx)));
162 static int process_work_frame(AVFilterContext *ctx)
164 FrameRateContext *s = ctx->priv;
166 int64_t interpolate, interpolate8;
171 if (!s->f0 && !s->flush)
174 work_pts = s->start_pts + av_rescale_q(s->n, av_inv_q(s->dest_frame_rate), s->dest_time_base);
176 if (work_pts >= s->pts1 && !s->flush)
180 s->work = av_frame_clone(s->f1);
182 if (work_pts >= s->pts1 + s->delta && s->flush)
185 interpolate = av_rescale(work_pts - s->pts0, s->blend_factor_max, s->delta);
186 interpolate8 = av_rescale(work_pts - s->pts0, 256, s->delta);
187 ff_dlog(ctx, "process_work_frame() interpolate: %"PRId64"/256\n", interpolate8);
188 if (interpolate >= s->blend_factor_max || interpolate8 > s->interp_end) {
189 s->work = av_frame_clone(s->f1);
190 } else if (interpolate <= 0 || interpolate8 < s->interp_start) {
191 s->work = av_frame_clone(s->f0);
193 ret = blend_frames(ctx, interpolate);
197 s->work = av_frame_clone(interpolate > (s->blend_factor_max >> 1) ? s->f1 : s->f0);
202 return AVERROR(ENOMEM);
204 s->work->pts = work_pts;
210 static av_cold int init(AVFilterContext *ctx)
212 FrameRateContext *s = ctx->priv;
213 s->start_pts = AV_NOPTS_VALUE;
217 static av_cold void uninit(AVFilterContext *ctx)
219 FrameRateContext *s = ctx->priv;
220 av_frame_free(&s->f0);
221 av_frame_free(&s->f1);
224 static int query_formats(AVFilterContext *ctx)
226 static const enum AVPixelFormat pix_fmts[] = {
228 AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUVJ411P,
229 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUVJ420P,
230 AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVJ422P,
231 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUVJ440P,
232 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
233 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV420P12,
234 AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV422P12,
235 AV_PIX_FMT_YUV444P9, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV444P12,
239 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
241 return AVERROR(ENOMEM);
242 return ff_set_common_formats(ctx, fmts_list);
245 #define BLEND_FRAME_FUNC(nbits) \
246 static void blend_frames##nbits##_c(BLEND_FUNC_PARAMS) \
249 uint##nbits##_t *dstw = (uint##nbits##_t *)dst; \
250 uint##nbits##_t *src1w = (uint##nbits##_t *)src1; \
251 uint##nbits##_t *src2w = (uint##nbits##_t *)src2; \
252 int bytes = nbits / 8; \
254 src1_linesize /= bytes; \
255 src2_linesize /= bytes; \
256 dst_linesize /= bytes; \
257 for (line = 0; line < height; line++) { \
258 for (pixel = 0; pixel < width; pixel++) \
259 dstw[pixel] = ((src1w[pixel] * factor1) + \
260 (src2w[pixel] * factor2) + half) \
261 >> BLEND_FACTOR_DEPTH(nbits); \
262 src1w += src1_linesize; \
263 src2w += src2_linesize; \
264 dstw += dst_linesize; \
270 void ff_framerate_init(FrameRateContext *s)
272 if (s->bitdepth == 8) {
273 s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH(8);
274 s->blend = blend_frames8_c;
276 s->blend_factor_max = 1 << BLEND_FACTOR_DEPTH(16);
277 s->blend = blend_frames16_c;
280 ff_framerate_init_x86(s);
283 static int config_input(AVFilterLink *inlink)
285 AVFilterContext *ctx = inlink->dst;
286 FrameRateContext *s = ctx->priv;
287 const AVPixFmtDescriptor *pix_desc = av_pix_fmt_desc_get(inlink->format);
290 s->vsub = pix_desc->log2_chroma_h;
291 for (plane = 0; plane < 4; plane++) {
292 s->line_size[plane] = av_image_get_linesize(inlink->format, inlink->w, plane);
293 s->height[plane] = inlink->h >> ((plane == 1 || plane == 2) ? s->vsub : 0);
296 s->bitdepth = pix_desc->comp[0].depth;
298 s->sad = ff_scene_sad_get_fn(s->bitdepth == 8 ? 8 : 16);
300 return AVERROR(EINVAL);
302 s->srce_time_base = inlink->time_base;
304 ff_framerate_init(s);
309 static int activate(AVFilterContext *ctx)
312 AVFilterLink *inlink = ctx->inputs[0];
313 AVFilterLink *outlink = ctx->outputs[0];
314 FrameRateContext *s = ctx->priv;
318 FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
321 ret = process_work_frame(ctx);
325 return ff_filter_frame(outlink, s->work);
327 ret = ff_inlink_consume_frame(inlink, &inpicref);
332 if (inpicref->interlaced_frame)
333 av_log(ctx, AV_LOG_WARNING, "Interlaced frame found - the output will not be correct.\n");
335 if (inpicref->pts == AV_NOPTS_VALUE) {
336 av_log(ctx, AV_LOG_WARNING, "Ignoring frame without PTS.\n");
337 av_frame_free(&inpicref);
342 pts = av_rescale_q(inpicref->pts, s->srce_time_base, s->dest_time_base);
344 if (s->f1 && pts == s->pts1) {
345 av_log(ctx, AV_LOG_WARNING, "Ignoring frame with same PTS.\n");
346 av_frame_free(&inpicref);
351 av_frame_free(&s->f0);
356 s->delta = s->pts1 - s->pts0;
360 av_log(ctx, AV_LOG_WARNING, "PTS discontinuity.\n");
361 s->start_pts = s->pts1;
363 av_frame_free(&s->f0);
366 if (s->start_pts == AV_NOPTS_VALUE)
367 s->start_pts = s->pts1;
372 if (ff_inlink_acknowledge_status(inlink, &status, &pts)) {
377 ff_outlink_set_status(outlink, status, pts);
381 FF_FILTER_FORWARD_WANTED(outlink, inlink);
383 return FFERROR_NOT_READY;
386 static int config_output(AVFilterLink *outlink)
388 AVFilterContext *ctx = outlink->src;
389 FrameRateContext *s = ctx->priv;
392 ff_dlog(ctx, "config_output()\n");
395 "config_output() input time base:%u/%u (%f)\n",
396 ctx->inputs[0]->time_base.num,ctx->inputs[0]->time_base.den,
397 av_q2d(ctx->inputs[0]->time_base));
399 // make sure timebase is small enough to hold the framerate
401 exact = av_reduce(&s->dest_time_base.num, &s->dest_time_base.den,
402 av_gcd((int64_t)s->srce_time_base.num * s->dest_frame_rate.num,
403 (int64_t)s->srce_time_base.den * s->dest_frame_rate.den ),
404 (int64_t)s->srce_time_base.den * s->dest_frame_rate.num, INT_MAX);
406 av_log(ctx, AV_LOG_INFO,
407 "time base:%u/%u -> %u/%u exact:%d\n",
408 s->srce_time_base.num, s->srce_time_base.den,
409 s->dest_time_base.num, s->dest_time_base.den, exact);
411 av_log(ctx, AV_LOG_WARNING, "Timebase conversion is not exact\n");
414 outlink->frame_rate = s->dest_frame_rate;
415 outlink->time_base = s->dest_time_base;
418 "config_output() output time base:%u/%u (%f) w:%d h:%d\n",
419 outlink->time_base.num, outlink->time_base.den,
420 av_q2d(outlink->time_base),
421 outlink->w, outlink->h);
424 av_log(ctx, AV_LOG_INFO, "fps -> fps:%u/%u scene score:%f interpolate start:%d end:%d\n",
425 s->dest_frame_rate.num, s->dest_frame_rate.den,
426 s->scene_score, s->interp_start, s->interp_end);
431 static const AVFilterPad framerate_inputs[] = {
434 .type = AVMEDIA_TYPE_VIDEO,
435 .config_props = config_input,
440 static const AVFilterPad framerate_outputs[] = {
443 .type = AVMEDIA_TYPE_VIDEO,
444 .config_props = config_output,
449 AVFilter ff_vf_framerate = {
451 .description = NULL_IF_CONFIG_SMALL("Upsamples or downsamples progressive source between specified frame rates."),
452 .priv_size = sizeof(FrameRateContext),
453 .priv_class = &framerate_class,
456 .query_formats = query_formats,
457 .inputs = framerate_inputs,
458 .outputs = framerate_outputs,
459 .flags = AVFILTER_FLAG_SLICE_THREADS,
460 .activate = activate,