On Thu, Jul 13, 2017 at 7:04 PM, Nicolas George <geo...@nsup.org> wrote:
> Hi. Thanks for the patch. See comments below. > > Le quintidi 25 messidor, an CCXXV, Ashish Singh a écrit : > > Hi, this is motion filter, one of the component filters of VMAF. > > It takes two videos as input. > > Run it using: ffmpeg -i main -i ref -lavfi motion -f null - > > Currently it outputs the average motion score over all frames. > > This is the commit message, it is there to stay. You probably should > have put these introductory comments between the triple-dash line and > the start of the diff itself, below the histogram. > > > > > --- > > Changelog | 1 + > > doc/filters.texi | 19 +++ > > libavfilter/Makefile | 2 + > > libavfilter/allfilters.c | 1 + > > > libavfilter/convolution.c | 144 +++++++++++++++++++++ > > libavfilter/convolution.h | 29 +++++ > > libavfilter/motion.h | 29 +++++ > > Does it really need to be in a separate file? > Hi, the convolution part will be used later by one more filter (vif) so I thought it's a good idea to keep it in a separate file to avoid duplication of code. > > > libavfilter/vf_motion.c | 316 ++++++++++++++++++++++++++++++ > ++++++++++++++++ > > 8 files changed, 541 insertions(+) > > create mode 100644 libavfilter/convolution.c > > create mode 100644 libavfilter/convolution.h > > create mode 100644 libavfilter/motion.h > > create mode 100644 libavfilter/vf_motion.c > > > > diff --git a/Changelog b/Changelog > > index 1778980..69657f4 100644 > > --- a/Changelog > > +++ b/Changelog > > @@ -10,6 +10,7 @@ version <next>: > > - config.log and other configuration files moved into ffbuild/ directory > > - update cuvid/nvenc headers to Video Codec SDK 8.0.14 > > - afir audio filter > > +- motion video filter > > > > version 3.3: > > - CrystalHD decoder moved to new decode API > > diff --git a/doc/filters.texi b/doc/filters.texi > > index 5985db6..011086b 100644 > > --- a/doc/filters.texi > > +++ b/doc/filters.texi > > @@ -9900,6 +9900,25 @@ Default method is @samp{fdiff}. > > Scene change detection threshold. Default is @code{5.0}. > > @end table > > > > > +@section motion > > I find the name of the filter too unspecific. > > > + > > +Obtain the average motion score between two input videos. > > + > > +This filter takes two input videos. > > + > > +Both input videos must have the same resolution and pixel format for > > +this filter to work correctly. Also it assumes that both inputs > > +have the same number of frames, which are compared one by one. > > + > > +The obtained average motion score is printed through the logging system. > > + > > +In the below example the input file @file{main.mpg} being processed is > compared > > +with the reference file @file{ref.mpg}. > > + > > +@example > > +ffmpeg -i main.mpg -i ref.mpg -lavfi motion -f null - > > +@end example > > + > > @section mpdecimate > > > > Drop frames that do not differ greatly from the previous frame in > > diff --git a/libavfilter/Makefile b/libavfilter/Makefile > > index f7dfe8a..446e367 100644 > > --- a/libavfilter/Makefile > > +++ b/libavfilter/Makefile > > @@ -13,6 +13,7 @@ OBJS = allfilters.o > \ > > avfiltergraph.o > \ > > buffersink.o > \ > > buffersrc.o > \ > > + convolution.o > \ > > drawutils.o > \ > > fifo.o > \ > > formats.o > \ > > @@ -227,6 +228,7 @@ OBJS-$(CONFIG_MESTIMATE_FILTER) += > vf_mestimate.o motion_estimation > > OBJS-$(CONFIG_METADATA_FILTER) += f_metadata.o > > OBJS-$(CONFIG_MIDEQUALIZER_FILTER) += vf_midequalizer.o > framesync.o > > OBJS-$(CONFIG_MINTERPOLATE_FILTER) += vf_minterpolate.o > motion_estimation.o > > +OBJS-$(CONFIG_MOTION_FILTER) += vf_motion.o dualinput.o > framesync.o > > OBJS-$(CONFIG_MPDECIMATE_FILTER) += vf_mpdecimate.o > > OBJS-$(CONFIG_NEGATE_FILTER) += vf_lut.o > > OBJS-$(CONFIG_NLMEANS_FILTER) += vf_nlmeans.o > > diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c > > index cd35ae4..7381799 100644 > > --- a/libavfilter/allfilters.c > > +++ b/libavfilter/allfilters.c > > @@ -239,6 +239,7 @@ static void register_all(void) > > REGISTER_FILTER(METADATA, metadata, vf); > > REGISTER_FILTER(MIDEQUALIZER, midequalizer, vf); > > REGISTER_FILTER(MINTERPOLATE, minterpolate, vf); > > + REGISTER_FILTER(MOTION, motion, vf); > > REGISTER_FILTER(MPDECIMATE, mpdecimate, vf); > > REGISTER_FILTER(NEGATE, negate, vf); > > REGISTER_FILTER(NLMEANS, nlmeans, vf); > > diff --git a/libavfilter/convolution.c b/libavfilter/convolution.c > > new file mode 100644 > > index 0000000..5c4520d > > --- /dev/null > > +++ b/libavfilter/convolution.c > > @@ -0,0 +1,144 @@ > > +/* > > + * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com> > > + * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com> > > + * > > + * This file is part of FFmpeg. > > + * > > + * FFmpeg is free software; you can redistribute it and/or > > + * modify it under the terms of the GNU Lesser General Public > > + * License as published by the Free Software Foundation; either > > + * version 2.1 of the License, or (at your option) any later version. > > + * > > + * FFmpeg is distributed in the hope that it will be useful, > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > > + * Lesser General Public License for more details. > > + * > > + * You should have received a copy of the GNU Lesser General Public > > + * License along with FFmpeg; if not, write to the Free Software > > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA > 02110-1301 USA > > + */ > > + > > +#include <stdbool.h> > > +#include "convolution.h" > > + > > +#define FORCE_INLINE __attribute__((always_inline)) > > +#define RESTRICT __restrict > > + > > > +static inline int floorn(int n, int m) > > +{ > > + return n - n % m; > > +} > > + > > +static inline int ceiln(int n, int m) > > +{ > > + return n % m ? n + (m - n % m) : n; > > +} > > These formulas are only valid if the parameters are positive. I suggest > you make the arguments unsigned and make very very sure the call sites > only give positive arguments. > > > + > > +FORCE_INLINE inline float convolution_edge(bool horizontal, const float > *filter, > > + int filt_width, const float > *src, > > > + int w, int h, int stride, > int i, > > The code base is currently moving all stride variables to ptrdiff_t. > > > + int j) > > +{ > > + int radius = filt_width / 2; > > + > > + float accum = 0; > > > + for (int k = 0; k < filt_width; ++k) { > > Declaring loop variables here is not accepted in FFmpeg's coding style. > > > + int i_tap = horizontal ? i : i - radius + k; > > + int j_tap = horizontal ? j - radius + k : j; > > + > > + if (horizontal) { > > + if (j_tap < 0) > > + j_tap = -j_tap; > > + else if (j_tap >= w) > > + j_tap = w - (j_tap - w + 1); > > + } else { > > + if (i_tap < 0) > > + i_tap = -i_tap; > > + else if (i_tap >= h) > > + i_tap = h - (i_tap - h + 1); > > + } > > + > > + accum += filter[k] * src[i_tap * stride + j_tap]; > > + } > > + return accum; > > +} > > + > > +static void convolution_x_c(const float *filter, int filt_width, > > + const float *src, float *dst, int w, int h, > > + int src_stride, int dst_stride, int step) > > +{ > > + int radius = filt_width / 2; > > + int borders_left = ceiln(radius, step); > > + int borders_right = floorn(w - (filt_width - radius), step); > > + > > + for (int i = 0; i < h; i++) { > > + for (int j = 0; j < borders_left; j += step) { > > + dst[i * dst_stride + j / step] = convolution_edge(true, > filter, > > + > filt_width, src, > > + w, h, > src_stride, > > + i, j); > > + } > > + > > + for (int j = borders_left; j < borders_right; j += step) { > > + float accum = 0; > > + for (int k = 0; k < filt_width; k++) { > > + accum += filter[k] * src[i * src_stride + j - radius + > k]; > > + } > > + dst[i * dst_stride + j / step] = accum; > > + } > > + > > + for (int j = borders_right; j < w; j += step) { > > + dst[i * dst_stride + j / step] = convolution_edge(true, > filter, > > + > filt_width, src, > > + w, h, > src_stride, > > + i, j); > > + } > > + } > > +} > > + > > +static void convolution_y_c(const float *filter, int filt_width, > > + const float *src, float *dst, int w, int h, > > + int src_stride, int dst_stride, int step) > > +{ > > + int radius = filt_width / 2; > > + int borders_top = ceiln(radius, step); > > + int borders_bottom = floorn(h - (filt_width - radius), step); > > + > > + for (int i = 0; i < borders_top; i += step) { > > + for (int j = 0; j < w; j++) { > > + dst[(i / step) * dst_stride + j] = convolution_edge(false, > filter, > > + > filt_width, src, > > + w, h, > src_stride, > > + i, j); > > + } > > + } > > + for (int i = borders_top; i < borders_bottom; i += step) { > > + for (int j = 0; j < w; j++) { > > + float accum = 0; > > + for (int k = 0; k < filt_width; k++) { > > + accum += filter[k] * src[(i - radius + k) * src_stride > + j]; > > + } > > + dst[(i / step) * dst_stride + j] = accum; > > + } > > + } > > + for (int i = borders_bottom; i < h; i += step) { > > + for (int j = 0; j < w; j++) { > > + dst[(i / step) * dst_stride + j] = convolution_edge(false, > filter, > > + > filt_width, src, > > + w, h, > src_stride, > > + i, j); > > + } > > + } > > +} > > + > > +void convolution_f32_c(const float *filter, int filt_width, const float > *src, > > + float *dst, float *tmp, int w, int h, int > src_stride, > > + int dst_stride) > > +{ > > + convolution_y_c(filter, filt_width, src, tmp, w, h, src_stride, > > + dst_stride, 1); > > + convolution_x_c(filter, filt_width, tmp, dst, w, h, src_stride, > > + dst_stride, 1); > > +} > > + > > diff --git a/libavfilter/convolution.h b/libavfilter/convolution.h > > new file mode 100644 > > index 0000000..34474f4 > > --- /dev/null > > +++ b/libavfilter/convolution.h > > @@ -0,0 +1,29 @@ > > +/* > > + * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com> > > + * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com> > > + * > > + * This file is part of FFmpeg. > > + * > > + * FFmpeg is free software; you can redistribute it and/or > > + * modify it under the terms of the GNU Lesser General Public > > + * License as published by the Free Software Foundation; either > > + * version 2.1 of the License, or (at your option) any later version. > > + * > > + * FFmpeg is distributed in the hope that it will be useful, > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > > + * Lesser General Public License for more details. > > + * > > + * You should have received a copy of the GNU Lesser General Public > > + * License along with FFmpeg; if not, write to the Free Software > > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA > 02110-1301 USA > > + */ > > + > > > +#ifndef CONVOLUTION_H_ > > +#define CONVOLUTION_H_ > > No need for the trailing underscore. > > > + > > > +void convolution_f32_c(const float *filter, int filter_width, const > float *src, > > + float *dst, float *tmp, int width, int height, > > + int src_stride, int dst_stride); > > Missing doxygen comment. > > Missing ff_ namespace. > > > + > > +#endif // CONVOLUTION_H_ > > diff --git a/libavfilter/motion.h b/libavfilter/motion.h > > new file mode 100644 > > index 0000000..0da41db > > --- /dev/null > > +++ b/libavfilter/motion.h > > @@ -0,0 +1,29 @@ > > +/* > > + * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com> > > + * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com> > > + * > > + * This file is part of FFmpeg. > > + * > > + * FFmpeg is free software; you can redistribute it and/or > > + * modify it under the terms of the GNU Lesser General Public > > + * License as published by the Free Software Foundation; either > > + * version 2.1 of the License, or (at your option) any later version. > > + * > > + * FFmpeg is distributed in the hope that it will be useful, > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > > + * Lesser General Public License for more details. > > + * > > + * You should have received a copy of the GNU Lesser General Public > > + * License along with FFmpeg; if not, write to the Free Software > > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA > 02110-1301 USA > > + */ > > + > > +#ifndef MOTION_TOOLS_H_ > > +#define MOTION_TOOLS_H_ > > + > > > +static int compute_motion(const float *ref, const float *dis, int w, > int h, > > + int ref_stride, int dis_stride, double *score, > > + void *ctx); > > + > > Missing doxygen comment. > > And a header file declaring a static function? Can you elaborate? > > > +#endif /* MOTION_TOOLS_H_ */ > > diff --git a/libavfilter/vf_motion.c b/libavfilter/vf_motion.c > > new file mode 100644 > > index 0000000..38568a9 > > --- /dev/null > > +++ b/libavfilter/vf_motion.c > > @@ -0,0 +1,316 @@ > > +/* > > + * Copyright (c) 2017 Ronald S. Bultje <rsbul...@gmail.com> > > + * Copyright (c) 2017 Ashish Pratap Singh <ashk43...@gmail.com> > > + * > > + * This file is part of FFmpeg. > > + * > > + * FFmpeg is free software; you can redistribute it and/or > > + * modify it under the terms of the GNU Lesser General Public > > + * License as published by the Free Software Foundation; either > > + * version 2.1 of the License, or (at your option) any later version. > > + * > > + * FFmpeg is distributed in the hope that it will be useful, > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU > > + * Lesser General Public License for more details. > > + * > > + * You should have received a copy of the GNU Lesser General Public > > + * License along with FFmpeg; if not, write to the Free Software > > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA > 02110-1301 USA > > + */ > > + > > +/** > > + * @file > > + * Calculate motion score between two input videos. > > + */ > > + > > +#include <inttypes.h> > > +#include "libavutil/avstring.h" > > +#include "libavutil/opt.h" > > +#include "libavutil/pixdesc.h" > > +#include "avfilter.h" > > +#include "dualinput.h" > > +#include "drawutils.h" > > +#include "formats.h" > > +#include "internal.h" > > +#include "motion.h" > > +#include "video.h" > > +#include "convolution.h" > > + > > +typedef struct MOTIONContext { > > + const AVClass *class; > > + FFDualInputContext dinput; > > + int width; > > + int height; > > + uint8_t type; > > + float *ref_data; > > + float *prev_blur_data; > > + float *blur_data; > > + float *temp_data; > > + double motion_sum; > > + uint64_t nb_frames; > > +} MOTIONContext; > > + > > > +#define OFFSET(x) offsetof(MOTIONContext, x) > > Unused. > > > +#define MAX_ALIGN 32 > > +#define ALIGN_CEIL(x) ((x) + ((x) % MAX_ALIGN ? MAX_ALIGN - (x) % > MAX_ALIGN : 0)) > > +#define OPT_RANGE_PIXEL_OFFSET (-128) > > + > > +static const AVOption motion_options[] = { > > + { NULL } > > +}; > > + > > +AVFILTER_DEFINE_CLASS(motion); > > + > > > +static const float FILTER_5[5] = { > > + 0.054488685, > > + 0.244201342, > > + 0.402619947, > > + 0.244201342, > > + 0.054488685 > > A comment telling where these constants come from would be nice. > > At a guess, I would say they are the values of the density function of > the normal distribution at 0, ±1 and ±2 renormalized to have their sum > 1. In that case, I think taking the cumulated values of the normal > distributions over [-0.5,+0.5], [0.5,1.5] and [1.5,+\infty[ may be more > elegant, but I do not know the theory behind that choice. > > > +}; > > + > > > +static inline double get_motion_avg(double motion_sum, uint64_t > nb_frames) > > +{ > > + return motion_sum / nb_frames; > > +} > > Is it really useful as a separate function? > > > + > > +static void offset(MOTIONContext *s, const AVFrame *ref, int stride) > > +{ > > + int w = s->width; > > + int h = s->height; > > + int i,j; > > + > > + int ref_stride = ref->linesize[0]; > > + > > + uint8_t *ref_ptr = ref->data[0]; > > + > > + float *ref_ptr_data = s->ref_data; > > + > > + for(i = 0; i < h; i++) { > > + for(j = 0; j < w; j++) { > > + ref_ptr_data[j] = (float) ref_ptr[j] + > OPT_RANGE_PIXEL_OFFSET; > > + } > > + ref_ptr += ref_stride / sizeof(uint8_t); > > + ref_ptr_data += stride / sizeof(float); > > + } > > +} > > + > > +static double image_sad_c(const float *img1, const float *img2, int w, > > + int h, int img1_stride, int img2_stride) > > +{ > > > + float accum = (float)0.0; > > 0.0f, or just 0.0. > > > + > > + for (int i = 0; i < h; i++) { > > + for (int j = 0; j < w; j++) { > > + float img1px = img1[i * img1_stride + j]; > > + float img2px = img2[i * img2_stride + j]; > > + > > + accum += fabs(img1px - img2px); > > + } > > + } > > + > > > + return (float) (accum / (w * h)); > > The cast looks completely wrong. > > > +} > > + > > +static int compute_motion(const float *ref, const float *dis, int w, > int h, > > + int ref_stride, int dis_stride, double *score, > > + void *ctx) > > +{ > > + *score = image_sad_c(ref, dis, w, h, ref_stride / sizeof(float), > > + dis_stride / sizeof(float)); > > + > > + return 0; > > +} > > + > > +static void set_meta(AVDictionary **metadata, const char *key, float d) > > +{ > > + char value[128]; > > + snprintf(value, sizeof(value), "%0.2f", d); > > + av_dict_set(metadata, key, value, 0); > > +} > > + > > +static AVFrame *do_motion(AVFilterContext *ctx, AVFrame *main, const > AVFrame *ref) > > +{ > > + MOTIONContext *s = ctx->priv; > > + AVDictionary **metadata = &main->metadata; > > + int stride; > > + size_t data_sz; > > + double score; > > + > > + stride = ALIGN_CEIL(s->width * sizeof(float)); > > + data_sz = (size_t)stride * s->height; > > + > > + offset(s, ref, stride); > > + > > + convolution_f32_c(FILTER_5, 5, s->ref_data, s->blur_data, > s->temp_data, > > + s->width, s->height, stride / sizeof(float), > stride / > > + sizeof(float)); > > + > > + if(!s->nb_frames) { > > + score = 0.0; > > + } else { > > + compute_motion(s->prev_blur_data, s->blur_data, s->width, > s->height, > > + stride, stride, &score, s); > > + } > > + > > + memcpy(s->prev_blur_data, s->blur_data, data_sz); > > + > > + set_meta(metadata, "lavfi.motion.score", score); > > + > > + s->nb_frames++; > > + > > + s->motion_sum += score; > > + > > + return main; > > +} > > + > > +static av_cold int init(AVFilterContext *ctx) > > +{ > > + MOTIONContext *s = ctx->priv; > > + > > + s->dinput.process = do_motion; > > + > > + return 0; > > +} > > + > > +static int query_formats(AVFilterContext *ctx) > > +{ > > + static const enum AVPixelFormat pix_fmts[] = { > > + AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P, > > + AV_PIX_FMT_YUV444P10LE, AV_PIX_FMT_YUV422P10LE, > AV_PIX_FMT_YUV420P10LE, > > + AV_PIX_FMT_NONE > > + }; > > + > > + AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); > > + if (!fmts_list) > > + return AVERROR(ENOMEM); > > + return ff_set_common_formats(ctx, fmts_list); > > +} > > + > > +static int config_input_ref(AVFilterLink *inlink) > > +{ > > + const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink-> > format); > > + AVFilterContext *ctx = inlink->dst; > > + MOTIONContext *s = ctx->priv; > > + int stride; > > + size_t data_sz; > > + > > + if (ctx->inputs[0]->w != ctx->inputs[1]->w || > > + ctx->inputs[0]->h != ctx->inputs[1]->h) { > > + av_log(ctx, AV_LOG_ERROR, "Width and height of input videos > must be same.\n"); > > + return AVERROR(EINVAL); > > + } > > + if (ctx->inputs[0]->format != ctx->inputs[1]->format) { > > + av_log(ctx, AV_LOG_ERROR, "Inputs must be of same pixel > format.\n"); > > + return AVERROR(EINVAL); > > + } > > + > > + s->width = ctx->inputs[0]->w; > > + s->height = ctx->inputs[0]->h; > > + > > + stride = ALIGN_CEIL(s->width * sizeof(float)); > > + data_sz = (size_t)stride * s->height; > > + > > + if (!(s->ref_data = av_malloc(data_sz))) { > > + av_log(ctx, AV_LOG_ERROR, "ref_buf allocation failed.\n"); > > > + return AVERROR(EINVAL); > > The correct return code would be ENOMEM. > > > + } > > + if (!(s->prev_blur_data = av_mallocz(data_sz))) { > > + av_log(ctx, AV_LOG_ERROR, "prev_blur_buf allocation failed.\n"); > > + return AVERROR(EINVAL); > > + } > > + if (!(s->blur_data = av_mallocz(data_sz))) { > > + av_log(ctx, AV_LOG_ERROR, "blur_buf allocation failed.\n"); > > + return AVERROR(EINVAL); > > + } > > + if (!(s->temp_data = av_mallocz(data_sz * 2))) { > > + av_log(ctx, AV_LOG_ERROR, "temp_buf allocation failed.\n"); > > + return AVERROR(EINVAL); > > + } > > + > > + s->type = desc->comp[0].depth > 8 ? 10 : 8; > > + > > + return 0; > > +} > > + > > +static int config_output(AVFilterLink *outlink) > > +{ > > + AVFilterContext *ctx = outlink->src; > > + MOTIONContext *s = ctx->priv; > > + AVFilterLink *mainlink = ctx->inputs[0]; > > + int ret; > > + > > + outlink->w = mainlink->w; > > + outlink->h = mainlink->h; > > + outlink->time_base = mainlink->time_base; > > + outlink->sample_aspect_ratio = mainlink->sample_aspect_ratio; > > + outlink->frame_rate = mainlink->frame_rate; > > + if ((ret = ff_dualinput_init(ctx, &s->dinput)) < 0) > > + return ret; > > + > > + return 0; > > +} > > + > > +static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref) > > +{ > > + MOTIONContext *s = inlink->dst->priv; > > + return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref); > > +} > > + > > +static int request_frame(AVFilterLink *outlink) > > +{ > > + MOTIONContext *s = outlink->src->priv; > > + return ff_dualinput_request_frame(&s->dinput, outlink); > > +} > > + > > +static av_cold void uninit(AVFilterContext *ctx) > > +{ > > + MOTIONContext *s = ctx->priv; > > + > > + ff_dualinput_uninit(&s->dinput); > > + > > + av_free(s->ref_data); > > + av_free(s->prev_blur_data); > > + av_free(s->blur_data); > > + av_free(s->temp_data); > > + > > + av_log(ctx, AV_LOG_INFO, "MOTION AVG: %.3f\n", > get_motion_avg(s->motion_sum, > > + > s->nb_frames)); > > +} > > + > > +static const AVFilterPad motion_inputs[] = { > > + { > > + .name = "main", > > + .type = AVMEDIA_TYPE_VIDEO, > > + .filter_frame = filter_frame, > > + },{ > > + .name = "reference", > > + .type = AVMEDIA_TYPE_VIDEO, > > + .filter_frame = filter_frame, > > + .config_props = config_input_ref, > > + }, > > + { NULL } > > +}; > > + > > +static const AVFilterPad motion_outputs[] = { > > + { > > + .name = "default", > > + .type = AVMEDIA_TYPE_VIDEO, > > + .config_props = config_output, > > + .request_frame = request_frame, > > + }, > > + { NULL } > > +}; > > + > > +AVFilter ff_vf_motion = { > > + .name = "motion", > > + .description = NULL_IF_CONFIG_SMALL("Calculate the MOTION > between two video streams."), > > + .init = init, > > + .uninit = uninit, > > + .query_formats = query_formats, > > + .priv_size = sizeof(MOTIONContext), > > + .priv_class = &motion_class, > > + .inputs = motion_inputs, > > + .outputs = motion_outputs, > > +}; > > Regards, > > -- > Nicolas George > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel