[FFmpeg-devel] [PATCH] avcodec/svq1: fix interframe mean VLC symbols

[Peter Ross]

On Mon, Oct 17, 2022 at 10:38:53PM +0200, Michael Niedermayer wrote:
> On Mon, Oct 17, 2022 at 08:33:28PM +1100, Peter Ross wrote:
> > On Mon, Oct 17, 2022 at 05:04:29AM +0200, Andreas Rheinhardt wrote:
> > > Peter Ross:
> > > > Fixes ticket #128.
> > > > 
> > > > The SVQ1 interframe mean VLC symbols -128 and 128 are incorrectly swapped
> > > > in our SVQ1 implementation, resulting in visible artifacts for some videos.
> > > > This patch unswaps the order of these two symbols.
> > > > 
> > > > The most noticable example of the artiacts caused by this error can be observed in
> > > > https://trac.ffmpeg.org/attachment/ticket/128/svq1_set.7z '352_288_k_50.mov'.
> > > > The artifacts are not observed when using the reference decoder
> > > > (QuickTime 7.7.9 x86 binary).
> > > > 
> > > > As a result of this patch, the reference data for the fate-svq1 test
> > > > ($SAMPLES/svq1/marymary-shackles.mov) must be modified. For this file, our
> > > > decoder output is now bitwise identical to the reference decoder. I have
> > > > tested patch with various other samples and they are all now bitwise identical.
> > > 
> > > Seems like this is not the only test whose reference needs to be
> > > updated. There are also the fate-vsynth%-svq1 tests.
> > 
> > Right, the encoder. This change will cause previous videos created by the FFmpeg
> > encoder to playback *with* artifacts in newer builds of FFmpeg with this patch
> > applied. There is not much we can do about this, since there isn't place for a
> > version string in the SVQ3 bitstream (unlike MPEG-4 etc).
> 
> it is possible i think to avoid this
> 
> we just need anything that differs reasonably consistently between the binary
> encoder and the ffmpeg encoder
> the first thing that comes to mind is the temporal reference values, we seem
> to always store 0 but there may be other things we do that differ

for the temporal reference field, the official binary encoder emits an increasing
sequence of numbers for each frame, and they wrap at 8-bits. i don't think we can
use a simple '== 0 case is FFmpeg', as there's a good chance that a legitimate
interframe will occur whe the field is 0.

there's not much else to detect whether its a FFmpeg file, just the checksum and
unknown extradata bits, both which the buggy FFmpeg encoder set to zero. Problem
is, some legitimate samples also have both these fields set to zero.

i am open to what other ideas you have!

> This fix will need some caution so everything keeps working 
> 1. we need the decoder to detect the official and our old buggy encoder and 
>    handle each appropriately
> 2. when we fix our encoder there are at least 3 ways
> 2a. avoid the ambigous codes, these will decode fine in all cases

i like this option. encoder performance is already abysmal, so it won't hurt too much.

> 2b. whatever we used to detect the official encoder we can mimic that too
>     while fixing this. but that may be a one shot opertunity
> 2c. find a cleaner way to store our version first, implment that in the
>     decoder and then use it in the encoder when we fix this. This would
>     need some testing to make sure it works with the official decoder
>     I see a few things that might work for the version. the decoder decodes
>     some "embedded message", theres the temporal reference which seems not
>     mattering
>     there are various encoder choices that allow embeding some message if one
>     really wanted
> 
> thx

going forward i propose we set the mov container extradata field to LIBAVCODEC_IDENT.
"just" in case there is another error in the encoder that needs addressing in future.
my analysis of the binary decoder suggests it not inspect the container extradata field.

-- Peter
(A907 E02F A6E5 0CD2 34CD 20D2 6760 79C5 AC40 DD6B)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20221018/d868ba37/attachment.sig>