[Bug 732266] vaapidecode: h264: add support for decoding SVC base layers only

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=732266

--- Comment #43 from sreerenj <bsreerenj at gmail.com> ---
(In reply to Orestis Floros from comment #42)
> (In reply to sreerenj from comment #40)
> > (In reply to Orestis Floros from comment #38)
> > > (In reply to sreerenj from comment #35)
> > > > (In reply to Orestis Floros from comment #34)
> > > > > Created attachment 358103 [details] [review] [review] [review] [review] [review]
> > > > > libs: decoder: h264: decode SVC base view only
> > > > >
> > > > > (In reply to sreerenj from comment #33)
> > > > > > Review of attachment 358041 [details] [review] [review] [review] [review] [review] [review]:
> > > > > >
> > > > > > ::: gst-libs/gst/vaapi/gstvaapidecoder_h264.c
> > > > > > @@ +1793,3 @@
> > > > > >
> > > > > > +  if (priv->base_only && !GST_H264_IS_MVC_NALU (&pi->nalu))
> > > > > > +    result = gst_h264_parse_sps (&pi->nalu, sps, FALSE);
> > > > > >
> > > > > > Why does it only for SVC? svc and mvc base-only mode should behave
> > > > > > similarly. right?
> > > > >
> > > > > I don't know how I missed it, this part is buggy and breaks MVC streams with
> > > > > base-only=TRUE right now. This patch should do it.
> > > > >
> > > > > I don't know why they behave differently here, I made the distinction
> > > > > through trial and error.
> > > >
> > > > Hm, we need an explanation other than trial-and-error-fix-this.
> > >
> > > I think it is related to nalparser. I'd bet that it has something to do with
> > > lack of SVC support in codecparsers/gsth264parser.c. This was the change
> > > that fixed the issue from:
> >
> > Can you investigate for the real issue?
> 
> Why SVCHS-2.264 fails with
> "lt-gst-launch-1.0: gen75_mfd.c:556: gen75_mfd_avc_img_state: Assertion
> `pic_param->pic_fields.bits.field_pic_flag == 0' failed"
> when this check isn't enabled:
> 
> The stream has one sps and one subset sps unit with the following
> frame_mbs_only_flag values:
> 
> - sps: frame_mbs_only_flag=1
> - subset: frame_mbs_only_flag=0
> 
> according to 7.3.3 - Slice header syntax:
> 
> if( !frame_mbs_only_flag ) {
>     field_pic_flag
>     if( field_pic_flag )
>         bottom_field_flag
> }
> 
> 
> For the annex-A slices codecparsers/gsth264parser.c:2092 uses the subset sps
> while it should have used the normal sps, thus !frame_mbs_only_flag == 1 and
> field_pic_flag gets read
> 



If we already removed subset_sps and considering only base layer, none of the
slice header should reference to subset_sps.
Here the non-base view having field macroblocks.
If the slice heder in base view still reference to the non-base view nals
(subset-sps or pps) , there could be two possible reasons:

1: nalparser override the sps with subset sps since both have the same id.
Or
2: stream is wrongly encoded

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.