[FFmpeg-devel] [PATCH] Video decoder and demuxer for AMV files

[Vladimir Voroshilov]

2007/9/27, Aurelien Jacobs <aurel at gnuage.org>:
> On Wed, 26 Sep 2007 17:18:27 +0700
> "Vladimir Voroshilov" <voroshil at gmail.com> wrote:
>
> > 2007/9/26, Aurelien Jacobs <aurel at gnuage.org>:
> > > Vladimir Voroshilov wrote:
> > >
> > [...]
> > > > I forget to say that resulting image is flipped upside-down.
> > > > Please anybody help me with fixing this.
> > >
> > > Short answer: negative stride
> > >
> > > Longer answer: Look at this code from mjpeg_decode_scan():
> > >    ptr = s->picture.data[c] +
> > >             (((s->linesize[c] * (v * mb_y + y) * 8) +
> > >             (h * mb_x + x) * 8) >> s->avctx->lowres);
> > >
> > > To flip the image upside-down, s->picture.data[c] need to be
> > > replaced by a pointer to (s->picture.data[c] + size of the image
> > > buffer - size of one line), and s->linesize[c] must be negative.
> > >
> > > That's the basic principle. You will probably need to do other
> > > small related changes. And ensure that the code still behave
> > > the same way as it does now for other mjpeg codecs.
> > >
> >
> > Thanks you for info!
> > I modified mjpeg_decode_scan routine.
> > Now decoder produces correct (not flipped) picture.
> >
> > All code enclosed into codec id check, thus there should not be any
> > problems with other mjpeg codecs.
> >
> > Updated patch is attached.
> >
> > Index: libavcodec/mjpegdec.c
> > ===================================================================
> > --- libavcodec/mjpegdec.c     (revision 10560)
> > +++ libavcodec/mjpegdec.c     (working copy)
> > @@ -662,6 +662,15 @@
> >      int EOBRUN = 0;
> >
> >      if(Ah) return 0; /* TODO decode refinement planes too */
> > +
> > +    if(s->avctx->codec->id==CODEC_ID_AMVVIDEO) {
> > +        //Modify pointers for getting upside-down flipped picture
> > +        for(i=0; i < nb_components; i++) {
> > +            int c = s->comp_index[i];
> > +            s->picture.data[c] += (s->linesize[c] * (s->v_scount[i]
> > * 8 * s->mb_height - 1));
> > +            s->linesize[c] *= -1;
> > +        }
> > +    }
> >      for(mb_y = 0; mb_y < s->mb_height; mb_y++) {
> >          for(mb_x = 0; mb_x < s->mb_width; mb_x++) {
> >              if (s->restart_interval && !s->restart_count)
> > @@ -717,6 +726,14 @@
> >              }
> >          }
> >      }
> > +    if(s->avctx->codec->id==CODEC_ID_AMVVIDEO) {
> > +        //Restore pointers to original values
> > +        for(i=0; i < nb_components; i++) {
> > +            int c = s->comp_index[i];
> > +            s->picture.data[c] += (s->linesize[c] * (s->v_scount[i]
> > * 8 * s->mb_height - 1));
> > +            s->linesize[c] *= -1;
> > +        }
> > +    }
> >      return 0;
> >  }
>
> This is ugly IMO.
> I would prefer not seeing any test related to one specific codec (ie. AMV)
> inside the generic internal decoding function used by a lot of codecs
> variations.
> At worst, you could move this code around the mjpeg_decode_scan() call
> in sp5xdec.c.
> At best you could init s->picture.data and s->picture.linesize directly
> to the proper value at amv codec init, but that may not be so simple.

AFAIK, mjpeg_decode_scan() is called from ff_mjpeg_decode_frame.
Moreover, s->picture is also created inside ff_mjpeg_decode_sof which
is called from ff_mjpeg_decode_frame.
In other words before ff_mjpeg_decode_frame picture does not exists yet.
After above call picture is already converted (fixing stride/pointers
after call to ff_mjpeg_decode_frame in sp5xdec.c not helps).

Thus i can  (without huge code duplication) use neigher first nor
second your suggection.

Fix me, if i am wrong.

-- 
Regards,
Vladimir Voroshilov     mailto:voroshil at gmail.com
JID: voroshil at gmail.com, voroshil at jabber.ru
ICQ: 95587719