[FFmpeg-devel] [PATCH 3/3] mov: support timecode extraction.

[Baptiste Coudurier]

On 1/3/12 12:28 AM, Clément Bœsch wrote:
> On Mon, Jan 02, 2012 at 09:36:19PM +0000, Baptiste Coudurier wrote:
>> On 1/2/12 4:23 PM, Clément Bœsch wrote:
>>> From: Clément Bœsch <clement.boesch at smartjog.com>
>>>
>>> ---
>>> I'm really not sure about this one. All the samples I have store the timecode
>>> in "frame number" format. But all of them have the Counter flag unset.
>>> Unfortunately, the specs[1] stats (p160+):
>>>
>>>     If the Counter flag is set to 1 in the timecode sample description, the
>>>     sample data is an unsigned 32-bit integer. The timecode counter value is
>>>     determined by dividing this unsigned 32-bit integer by the number of frames
>>>     field in the timecode sample description.
>>>
>>>     If the Counter flag is set to 0 in the timecode sample description, the
>>>     sample data format is a signed 32-bit integer and is used to calculate a
>>>     timecode record, defined as follows.
>>>
>>>     [...]
>>>
>>> So basically, I understand it as: cf=1 → frame number format, cf=0 → qt format.
>>>
>>> Also, I can't find a sample with a timecode stored in quicktime format (where
>>> counter flag should be set to 0).
>>
>> Just remove that code and display a message. It's not needed yet.
>>
> 
> Mmh, OK. That is certainly a naive question, but any idea why the specs
> are just being violated in that case?
> 
> Anyway, I dropped the QT format code, and assumed frame counter format all
> the time. I'm not so sure about printing a warning since it will likely
> warn for every single sample. OTOH, I think a comment on this might be
> appropriate for any future potential improvement.
> 
> New patch attached.
> 
> 
> 
> 0003-mov-support-timecode-extraction.patch
> 
> 
> From c816e8916c23f4d83c2d5ba0112ddc0df1513c1d Mon Sep 17 00:00:00 2001
> From: =?UTF-8?q?Cl=C3=A9ment=20B=C5=93sch?= <clement.boesch at smartjog.com>
> Date: Mon, 2 Jan 2012 11:56:06 +0100
> Subject: [PATCH 3/3] mov: support timecode extraction.
> 
> ---
>  Changelog         |    2 +-
>  libavformat/mov.c |   51 +++++++++++++++++++++++++++++++++++++++++++++++++--
>  2 files changed, 50 insertions(+), 3 deletions(-)
> 
> diff --git a/Changelog b/Changelog
> index 7b5d277..2dcc423 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -6,7 +6,7 @@ version next:
>  - v410 Quicktime Uncompressed 4:4:4 10-bit encoder and decoder
>  - SBaGen (SBG) binaural beats script demuxer
>  - OpenMG Audio muxer
> -- dv: add timecode to metadata
> +- Timecode extraction in DV and MOV
>  - thumbnail video filter
>  - XML output in ffprobe
>  - asplit audio filter
> diff --git a/libavformat/mov.c b/libavformat/mov.c
> index ee67a9f..57e7949 100644
> --- a/libavformat/mov.c
> +++ b/libavformat/mov.c
> @@ -36,6 +36,7 @@
>  #include "riff.h"
>  #include "isom.h"
>  #include "libavcodec/get_bits.h"
> +#include "libavcodec/timecode.h"
>  #include "id3v1.h"
>  #include "mov_chan.h"
>  
> @@ -2615,6 +2616,46 @@ finish:
>      avio_seek(sc->pb, cur_pos, SEEK_SET);
>  }
>  
> +static int parse_timecode_in_framenum_format(AVFormatContext *s, AVStream *st,
> +                                             uint32_t value)
> +{
> +    char buf[16];
> +    struct ff_timecode tc = {
> +        .drop  = st->codec->flags2 & CODEC_FLAG2_DROP_FRAME_TIMECODE,
> +        .rate  = (AVRational){st->codec->time_base.den,
> +                              st->codec->time_base.num},
> +    };
> +
> +    if (avpriv_check_timecode_rate(s, tc.rate, tc.drop) < 0)
> +        return AVERROR(EINVAL);
> +    av_dict_set(&st->metadata, "timecode",
> +                avpriv_timecode_to_string(buf, &tc, value), 0);

Overall the API is pretty ugly.
Why 2 params for check_timecode_rate when passing a struct tcblah would
be sufficient.
Why timecode_to_string when it's obviously a frame number instead of a timecode ?
Why a struct tcblah + value instead of one struct tc ?

> +    return 0;
> +}
> +
> +static int mov_read_timecode_track(AVFormatContext *s, AVStream *st)
> +{
> +    MOVStreamContext *sc = st->priv_data;
> +    int64_t cur_pos = avio_tell(sc->pb);
> +    uint32_t value;
> +
> +    if (!st->nb_index_entries)
> +        return -1;
> +
> +    avio_seek(sc->pb, st->index_entries->pos, SEEK_SET);
> +    value = avio_rb32(s->pb);
> +
> +    /* Assume Counter flag is set to 1 in tmcd track (even though it is likely
> +     * not the case) and thus assume "frame number format" instead of QT one.
> +     * No sample with tmcd track can be found with a QT timecode at the moment,
> +     * despite what the tmcd track "suggests" (Counter flag set to 0 means QT
> +     * format). */
> +    parse_timecode_in_framenum_format(s, st, value);
> +
> +    avio_seek(sc->pb, cur_pos, SEEK_SET);
> +    return 0;
> +}
> +
>  static int mov_read_header(AVFormatContext *s, AVFormatParameters *ap)
>  {
>      MOVContext *mov = s->priv_data;
> @@ -2640,8 +2681,14 @@ static int mov_read_header(AVFormatContext *s, AVFormatParameters *ap)
>      }
>      av_dlog(mov->fc, "on_parse_exit_offset=%"PRId64"\n", avio_tell(pb));
>  
> -    if (pb->seekable && mov->chapter_track > 0)
> -        mov_read_chapters(s);
> +    if (pb->seekable) {
> +        int i;
> +        if (mov->chapter_track > 0)
> +            mov_read_chapters(s);
> +        for (i = 0; i < s->nb_streams; i++)
> +            if (s->streams[i]->codec->codec_tag == MKTAG('t','m','c','d'))

AV_RL32("tmcd")


-- 
Baptiste COUDURIER
Key fingerprint                 8D77134D20CC9220201FC5DB0AC9325C5C1ABAAA
FFmpeg maintainer                                  http://www.ffmpeg.org