[FFmpeg-devel] [PATCH 2/3] idet improvements: add reset_count feature

Michael Niedermayer michaelni at gmx.at
Sun Nov 2 18:43:37 CET 2014


On Sun, Nov 02, 2014 at 05:03:44AM -0800, Kevin Mitchell wrote:
> Removed the unrelated documentation changes, used uint64_t instead of float.

>  doc/filters.texi      |    8 +++++
>  libavfilter/version.h |    2 -
>  libavfilter/vf_idet.c |   79 ++++++++++++++++++++++++++++++++++----------------
>  libavfilter/vf_idet.h |    9 ++++-
>  4 files changed, 71 insertions(+), 27 deletions(-)
> d7f412c49b294b54535d4cfe48e54aa10f35ba08  0001-avfilter-vf_idet-add-a-half_life-option-for-statisti.patch
> From 0fc44ce601266e94fb8c6137091f3eb91616a235 Mon Sep 17 00:00:00 2001
> From: Kevin Mitchell <kevmitch at gmail.com>
> Date: Sun, 2 Nov 2014 04:37:13 -0800
> Subject: [PATCH 1/2] avfilter/vf_idet: add a "half_life" option for statistics
> 
> This can be useful for videos in which the interlacing pattern changes.
> 
> Also log the total number of frames as metadata and with avlog.
> ---
>  doc/filters.texi      |  8 ++++++
>  libavfilter/version.h |  2 +-
>  libavfilter/vf_idet.c | 79 +++++++++++++++++++++++++++++++++++----------------
>  libavfilter/vf_idet.h |  9 ++++--
>  4 files changed, 71 insertions(+), 27 deletions(-)
> 
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 33f842b..03bd1c6 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -5583,6 +5583,9 @@ Multiple frame detection incorporates the classification history of previous fra
>  The filter will log these metadata values:
>  
>  @table @option
> + at item frames_total
> +Total number of frames considered in current statistics.
> +
>  @item single.current_frame
>  Detected type of current frame using single-frame detection. One of:
>  ``tff'' (top field first), ``bff'' (bottom field first),
> @@ -5625,6 +5628,11 @@ The filter accepts the following options:
>  Set interlacing threshold.
>  @item prog_thres
>  Set progressive threshold.
> + at item half_life
> +Number of frames after which a given frame's contribution to the
> +statistics is halved (i.e., it contributes only 0.5 to it's
> +classification). The default of 0 means that all frames seen are given
> +full weight of 1.0 forever.
>  @end table
>  
>  @section il
> diff --git a/libavfilter/version.h b/libavfilter/version.h
> index 440c587..dab9b45 100644
> --- a/libavfilter/version.h
> +++ b/libavfilter/version.h
> @@ -31,7 +31,7 @@
>  
>  #define LIBAVFILTER_VERSION_MAJOR  5
>  #define LIBAVFILTER_VERSION_MINOR  2
> -#define LIBAVFILTER_VERSION_MICRO 101
> +#define LIBAVFILTER_VERSION_MICRO 102
>  
>  #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
>                                                 LIBAVFILTER_VERSION_MINOR, \
> diff --git a/libavfilter/vf_idet.c b/libavfilter/vf_idet.c
> index 6f99f39..4adadd4 100644
> --- a/libavfilter/vf_idet.c
> +++ b/libavfilter/vf_idet.c
> @@ -32,6 +32,7 @@
>  static const AVOption idet_options[] = {
>      { "intl_thres", "set interlacing threshold", OFFSET(interlace_threshold),   AV_OPT_TYPE_FLOAT, {.dbl = 1.04}, -1, FLT_MAX, FLAGS },
>      { "prog_thres", "set progressive threshold", OFFSET(progressive_threshold), AV_OPT_TYPE_FLOAT, {.dbl = 1.5},  -1, FLT_MAX, FLAGS },
> +    { "half_life", "half life of cumulative statistics", OFFSET(half_life),     AV_OPT_TYPE_FLOAT, {.dbl = 0.0},  -1, INT_MAX, FLAGS },
>      { NULL }
>  };
>  

> @@ -48,6 +49,14 @@ static const char *type2str(Type type)
>      return NULL;
>  }
>  
> +static int av_dict_set_float(AVDictionary **pm, const char *key, float value,
> +                int flags)
> +{
> +    char valuestr[22];
> +    snprintf(valuestr, sizeof(valuestr), "%.1f", value);
> +    return av_dict_set(pm, key, valuestr, flags);
> +}

i dont think "%f" is guranteed to produce the same value across
platforms, i think we had a issue in fate with win32 and a %f once
but maybe i misremember

something like
%d.%d", A/10, A%10
could be used


> +
>  int ff_idet_filter_line_c(const uint8_t *a, const uint8_t *b, const uint8_t *c, int w)
>  {
>      int x;
> @@ -74,6 +83,26 @@ int ff_idet_filter_line_c_16bit(const uint16_t *a, const uint16_t *b, const uint
>      return ret;
>  }
>  
> +static void log_cumulative_stats(AVFilterContext *ctx)
> +{
> +    IDETContext *idet = ctx->priv;
> +
> +    av_log(ctx, AV_LOG_INFO, "Single frame detection: TFF:%6.1f BFF:%6.1f Progressive:%6.1f Undetermined:%6.1f Total: %6.1f\n",
> +           idet->prestat[TFF] / PRECISION,
> +           idet->prestat[BFF] / PRECISION,
> +           idet->prestat[PROGRESSIVE] / PRECISION,
> +           idet->prestat[UNDETERMINED]  / PRECISION,
> +           idet->total / PRECISION
> +        );
> +    av_log(ctx, AV_LOG_INFO, "Multi frame detection: TFF:%6.1f BFF:%6.1f Progressive:%6.1f Undetermined:%6.1f Total: %6.1f\n",
> +           idet->poststat[TFF] / PRECISION,
> +           idet->poststat[BFF] / PRECISION,
> +           idet->poststat[PROGRESSIVE] / PRECISION,
> +           idet->poststat[UNDETERMINED] / PRECISION,
> +           idet->total / PRECISION
> +        );
> +}

same issue
also its at least to me a bit unexpected that the filter
prints final statistics which only represent the last frames instead
of all


> +
>  static void filter(AVFilterContext *ctx)
>  {
>      IDETContext *idet = ctx->priv;
> @@ -146,23 +175,31 @@ static void filter(AVFilterContext *ctx)
>          idet->cur->interlaced_frame = 0;
>      }
>  
> -    idet->prestat [           type] ++;
> -    idet->poststat[idet->last_type] ++;
> +    idet->total           = (uint64_t) round(idet->total        * idet->decay_coefficient);
> +    for(i=0; i<4; i++){
> +        idet->prestat [i] = (uint64_t) round(idet->prestat [i]  * idet->decay_coefficient);
> +        idet->poststat[i] = (uint64_t) round(idet->poststat[i]  * idet->decay_coefficient);
> +    }

i really think floats should not be used except for storing the user
provided parameters, which can be choosen to be exactly representable
for the case of a regression test



[...]



-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

If you think the mosad wants you dead since a long time then you are either
wrong or dead since a long time.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20141102/c194b5a7/attachment.asc>


More information about the ffmpeg-devel mailing list