[FFmpeg-devel] [PATCH] avformat/mov: Add support for demuxing still HEIC images

[Vignesh Venkat]

On Tue, Oct 3, 2023 at 7:35 PM Leo Izen <leo.izen at gmail.com> wrote:
>
> On 9/26/23 13:37, Vignesh Venkatasubramanian via ffmpeg-devel wrote:
> > They are similar to AVIF images (both use the HEIF container).
> > The only additional work needed is to parse the hvcC box and put
> > it in the extradata.
> >
> > With this patch applied, ffmpeg (when built with an HEVC decoder)
> > is able to decode the files in
> > https://github.com/nokiatech/heif/tree/gh-pages/content/images
> >
> > Partially fixes trac ticket #6521.
> >
> > Signed-off-by: Vignesh Venkatasubramanian <vigneshv at google.com>
> > ---
> >   libavformat/isom.h |  2 ++
> >   libavformat/mov.c  | 38 +++++++++++++++++++++++++++++++++++++-
> >   2 files changed, 39 insertions(+), 1 deletion(-)
> >
> > diff --git a/libavformat/isom.h b/libavformat/isom.h
> > index 3d375d7a46..b30b9da65e 100644
> > --- a/libavformat/isom.h
> > +++ b/libavformat/isom.h
> > @@ -327,6 +327,8 @@ typedef struct MOVContext {
> >           int64_t extent_offset;
> >       } *avif_info;
> >       int avif_info_size;
> > +    int64_t hvcC_offset;
> > +    int hvcC_size;
> >       int interleaved_read;
> >   } MOVContext;
> >
> > diff --git a/libavformat/mov.c b/libavformat/mov.c
> > index 1996e0028c..cec9cb5fe1 100644
> > --- a/libavformat/mov.c
> > +++ b/libavformat/mov.c
> > @@ -1218,7 +1218,8 @@ static int mov_read_ftyp(MOVContext *c, AVIOContext *pb, MOVAtom atom)
> >           c->isom = 1;
> >       av_log(c->fc, AV_LOG_DEBUG, "ISO: File Type Major Brand: %.4s\n",(char *)&type);
> >       av_dict_set(&c->fc->metadata, "major_brand", type, 0);
> > -    c->is_still_picture_avif = !strncmp(type, "avif", 4);
> > +    c->is_still_picture_avif = !strncmp(type, "avif", 4) ||
> > +                               !strncmp(type, "mif1", 4);
>
> This appears to be an unrelated change. Is it?
>

No, "mif1" is the major brand reported by HEIC files. So this is
needed to detect HEIC files.

> >       minor_ver = avio_rb32(pb); /* minor version */
> >       av_dict_set_int(&c->fc->metadata, "minor_version", minor_ver, 0);
> >
> > @@ -4911,6 +4912,16 @@ static int avif_add_stream(MOVContext *c, int item_id)
> >       st->priv_data = sc;
> >       st->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
> >       st->codecpar->codec_id = AV_CODEC_ID_AV1;
> > +    if (c->hvcC_offset >= 0) {
> > +        int ret;
> > +        int64_t pos = avio_tell(c->fc->pb);
> > +        st->codecpar->codec_id = AV_CODEC_ID_HEVC;
> > +        avio_seek(c->fc->pb, c->hvcC_offset, SEEK_SET);
> > +        ret = ff_get_extradata(c->fc, st->codecpar, c->fc->pb, c->hvcC_size);
> > +        if (ret < 0)
> > +            return ret;
> > +        avio_seek(c->fc->pb, pos, SEEK_SET);
> > +    }
>
> Will this fail on non-seekable input? If it does, do we care?
>

I tried this with a pipe as an input (cat file.heic | ffmpeg -i - ...)
and it worked. I also see plenty of unconditional avio_seek calls in
this file. So I am not sure how it works.

Anyways, I added a check to fail here if the seek does not take us to
the desired position.

> >       sc->ffindex = st->index;
> >       c->trak_index = st->index;
> >       st->avg_frame_rate.num = st->avg_frame_rate.den = 1;
> > @@ -4953,6 +4964,8 @@ static int avif_add_stream(MOVContext *c, int item_id)
> >
> >   static int mov_read_meta(MOVContext *c, AVIOContext *pb, MOVAtom atom)
> >   {
> > +    c->hvcC_offset = -1;
> > +    c->hvcC_size = 0;
> >       while (atom.size > 8) {
> >           uint32_t tag;
> >           if (avio_feof(pb))
> > @@ -7826,6 +7839,28 @@ static int mov_read_iloc(MOVContext *c, AVIOContext *pb, MOVAtom atom)
> >       return atom.size;
> >   }
> >
> > +static int mov_read_iprp(MOVContext *c, AVIOContext *pb, MOVAtom atom)
> > +{
> > +    int size = avio_rb32(pb);
> > +    if (avio_rl32(pb) != MKTAG('i','p','c','o'))
> > +        return AVERROR_INVALIDDATA;
>
> Is there a reason you use ipco here instead of iprp? I'm not saying this
> is wrong, but I am confused while you're doing it this way.
>

The control flow will jump into this function only when an "iprp" box
has been seen (pb points to the start of the contents of the iprp
box). "ipco" is a sub-box of the "iprp" box that we care about. We
cannot register sub-boxes to mov_read_default without registering the
containing box, as the containing box will be skipped if it is
unknown. Some other sub-box reading is already implemented this way
(mov_read_meta for example).

> > +    size -= 8;
> > +    while (size > 0) {
> > +        int sub_size, sub_type;
> > +        sub_size = avio_rb32(pb);
> > +        sub_type = avio_rl32(pb);
> > +        sub_size -= 8;
> > +        size -= sub_size + 8;
> > +        if (sub_type == MKTAG('h','v','c','C')) {
> > +            c->hvcC_offset = avio_tell(pb);
> > +            c->hvcC_size = sub_size;
> > +            break;
> > +        }
>
> Are these permitted to use extended-size tags? i.e. size = 1, followed
> by a big-endian 64-bit size, then followed by the tag?
>

MP4 "type" fields are always 32-bits. So they cannot be of any other size.

Thank you for the review!

> > +        avio_skip(pb, sub_size);
> > +    }
> > +    return atom.size;
> > +}
> > +
> >   static const MOVParseTableEntry mov_default_parse_table[] = {
> >   { MKTAG('A','C','L','R'), mov_read_aclr },
> >   { MKTAG('A','P','R','G'), mov_read_avid },
> > @@ -7933,6 +7968,7 @@ static const MOVParseTableEntry mov_default_parse_table[] = {
> >   { MKTAG('p','c','m','C'), mov_read_pcmc }, /* PCM configuration box */
> >   { MKTAG('p','i','t','m'), mov_read_pitm },
> >   { MKTAG('e','v','c','C'), mov_read_glbl },
> > +{ MKTAG('i','p','r','p'), mov_read_iprp },
> >   { 0, NULL }
> >   };
> >
>
> - Leo Izen
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".



-- 
Vignesh