theora dev - Feb 2010 - Ogg files incompatible, was: Video encoded using libtheora-1.1.1 and libogg-1.1.4 plays too fast...

> Date: Thu, 11 Feb 2010 10:54:01 -0500
> From: tterribe at email.unc.edu
> CC: theora-dev at xiph.org
> 
> > This resulted in a slight larger file but no discernible difference in
> > the movie play length.  It''s certainly not playing the
expected duration.
> > What do you think?
> 
> Since you''re just compressing the same frame over and over again,
what
> you''re likely getting is a keyframe followed by a bunch of
zero-byte
> packets, which is Theora''s cheap way of saying "keep showing
the last
> frame". Some (broken) players treat these as semantically different,
> however, and will end playback early when they appear at the end of a
> stream. I suggest running ogginfo or oggzinfo on the file and see what
> it says the real length is. You may also be interested in oggz-validate,
> which can confirm that you are writing out all of the pages in the
> proper order.
	Thank you for your help!  According to oggz-validate, the previously generated
ogg file was missing an EOS packet.  With the new packet writing loop, there
appears to be nothing wrong with these ogg streams, again according to
oggz-validate.  It looks like you may be right about broken players.  When I
feed my code actual video frames, the files play at normal speed.	However,
something is still wrong.  My DirectShow filter, ffdshow, can''t play
these ogg files at all and, instead, show a very short sequence of random
colours, crashing most of the time.  VLC player can play these files but
I''m afraid that''s not good enough for my purposes.  My
ultimate goal is to upload these theora videos to YouTube but they have the same
problems as ffdshow.	The reason I still think the problem is with my code is
that both players and YouTube can play other ogg files not made by my software. 
Here are two examples:
http://www.osnews.com/img/19020/qvga.ogghttp://www.osnews.com/img/19020/720p-theora.ogg
	I haven''t actually uploaded these to YouTube but, in general, people
have had success uploading Ogg/Theora video files to YouTube and these play fine
with my DirectShow filter...	Should my code just work?
> By the way, please do not use a constant for the stream serial number.
> Multiple streams in the same file must have unique serial numbers, and
> if someone later wants to concatenate multiple streams you produce, they
> will have to rewrite the stream with a new serial number or (more
> likely) will generate an invalid file.
	Can you explain this further please?  Can''t they simply assign their
new stream a number unique to any existing one in the ogg file?	Thank you very
much...


 		 	   		  
_________________________________________________________________

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/theora-dev/attachments/20100212/15cc54ec/attachment.htm

Id Kong wrote:
> However, something is still wrong.  My DirectShow filter, ffdshow,
can''t
> play these ogg files at all and, instead, show a very short sequence of
> random colours, crashing most of the time.  VLC player can play these
Unless it''s using a very recent build of ffmpeg, ffdshow does not
properly support Theora. Its decoder was limited to the old VP3 subset,
and did not support any new things we included in the spec, despite the
fact that it was published in 2004. It only worked because the mainline
encoder didn''t use any of these extensions either, until the recent 1.1
release. Most of these issues have been fixed (thanks to David Conrad),
though ffmpeg still does not properly support cropping rectangles with a
non-zero offset for video that doesn''t have a multiple-of-16 frame size
(they don''t support this feature for H.264, either). Have you tried our
DirectShow filters: http://www.xiph.org/dshow/ ? If these do not work,
we would like to know.
> files but I''m afraid that''s not good enough for my
purposes.  My
> ultimate goal is to upload these theora videos to YouTube but they have
> the same problems as ffdshow.
Right. As far as I know, they are using some old build of ffmpeg to
transcode Theora. This situation is, shall we say, less than ideal.
> The reason I still think the problem is with my code is that both
> players and YouTube can play other ogg files not made by my software.
>  Here are two examples:
> 
> http://www.osnews.com/img/19020/qvga.ogg
> http://www.osnews.com/img/19020/720p-theora.ogg
Vendor: Xiph.Org libTheora I 20071025 3 2 1

These are using an encoder from before the 1.1 release, which is why
they work with the broken ffmpeg.
> Can you explain this further please?  Can''t they simply assign
their new
> stream a number unique to any existing one in the ogg file?
> Thank you very much...
I''m talking about tools that edit existing streams. Assigning a stream
a
new serial number means rewriting all of the pages, and computing a new
checksum for them.

> Date: Fri, 12 Feb 2010 08:30:50 -0500
> From: tterribe at email.unc.edu
> CC: theora-dev at xiph.org
> 
> Id Kong wrote:
> > However, something is still wrong.  My DirectShow filter, ffdshow,
can''t
> > play these ogg files at all and, instead, show a very short sequence
of
> > random colours, crashing most of the time.  VLC player can play these
> 
> Unless it''s using a very recent build of ffmpeg, ffdshow does not
> properly support Theora. Its decoder was limited to the old VP3 subset,
> and did not support any new things we included in the spec, despite the
> fact that it was published in 2004. It only worked because the mainline
> encoder didn''t use any of these extensions either, until the
recent 1.1
> release. Most of these issues have been fixed (thanks to David Conrad),
> though ffmpeg still does not properly support cropping rectangles with a
> non-zero offset for video that doesn''t have a multiple-of-16 frame
size
> (they don''t support this feature for H.264, either). Have you
tried our
> DirectShow filters: http://www.xiph.org/dshow/ ? If these do not work,
> we would like to know.
	Thank you for this information.  It''s an eye opener!
> > files but I''m afraid that''s not good enough for my
purposes.  My
> > ultimate goal is to upload these theora videos to YouTube but they
have
> > the same problems as ffdshow.
> 
> Right. As far as I know, they are using some old build of ffmpeg to
> transcode Theora. This situation is, shall we say, less than ideal.
	Indeed, it''s much less than ideal.  Do you know where I can find
source code for older versions of libtheora?  I can''t find any on the
site and a Google search wasn''t fruitful.  Since my ultimate goal is to
upload video to YouTube, what matters is that it actually work, so compatibility
is my priority...
> > The reason I still think the problem is with my code is that both
> > players and YouTube can play other ogg files not made by my software.
> >  Here are two examples:
> > 
> > http://www.osnews.com/img/19020/qvga.ogg
> > http://www.osnews.com/img/19020/720p-theora.ogg
> 
> Vendor: Xiph.Org libTheora I 20071025 3 2 1
> 
> These are using an encoder from before the 1.1 release, which is why
> they work with the broken ffmpeg.
	If I''m not mistaken, they were encoded using ffmpeg2theora so that
it''s compatible with (the broken) ffmpeg shouldn''t be
surprising.  So basically, no one''s using the "new"
specification...
> > Can you explain this further please?  Can''t they simply
assign their new
> > stream a number unique to any existing one in the ogg file?
> > Thank you very much...
> 
> I''m talking about tools that edit existing streams. Assigning a
stream a
> new serial number means rewriting all of the pages, and computing a new
> checksum for them.
	I was talking about editing existing streams.  Assuming whatever made the ogg
file gave each stream a unique but non-random serial number, if I wanted to add
to it, can''t I simply choose another non-random serial number
that''s unique to whatever numbers are already used in the ogg file? 
What am I missing here?	Thank you...
_________________________________________________________________
Check your Hotmail from your phone.
http://go.microsoft.com/?linkid=9708121
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/theora-dev/attachments/20100212/e702f342/attachment.htm

On 2/12/10, Id Kong <spam_receptacle_ at hotmail.com>
wrote:
> Do you know where I can find source code for older versions of
> libtheora?
https://svn.xiph.org/releases/theora

>> Right. As far as I know, they are using some old build of ffmpeg to
>> transcode Theora. This situation is, shall we say, less than ideal.
>
> Indeed, it''s much less than ideal. ?Do you know where I can find
source code
> for older versions of libtheora? ?I can''t find any on the site and
a Google
> search wasn''t fruitful. ?Since my ultimate goal is to upload video
to
> YouTube, what matters is that it actually work, so compatibility is my
> priority...
There''s a better way than using an old libtheora.  We anticipated the
need for continued compatability with VP3 and the new encoder can be
run in VP3 compat mode.  In this mode it will not use any features
that weren''t available in VP3.  This obviously penalizes efficiency,
but it does work.
> If I''m not mistaken, they were encoded using ffmpeg2theora so that
it''s
> compatible with (the broken) ffmpeg shouldn''t be surprising. ?So
basically,
> no one''s using the "new" specification...
Plenty of people are using the new spec, but ffmpeg was very very late
to the party.

Monty

>> If I''m not mistaken, they were encoded using ffmpeg2theora so
that it''s
>> compatible with (the broken) ffmpeg shouldn''t be surprising.
?So basically,
>> no one''s using the "new" specification...
>
> Plenty of people are using the new spec, but ffmpeg was very very late
> to the party.
Oops, I should clarify---

There is no ''new spec'' and ''old spec''. 
There has always been only one
Theora spec, which has always included more features than the original
VP3.  Ffmpeg, however, has always mistakenly treated Theora as just a
rebadged VP3 and for a long time our own encoder defaulted to
outputting VP3 compatible bitstreams.  During this period, ffmpeg
refused to implement the full Theora spec as it wasn''t important to
them (traditionally, a large fraction of the ffmpeg community have
been h264 activitsts, and this faction considers Theora an irritating
distraction).

Monty

> Date: Fri, 12 Feb 2010 11:19:10 -0500
> From: xiphmont at xiph.org
> 
> > Indeed, it''s much less than ideal.  Do you know where I can
find source code
> > for older versions of libtheora?  I can''t find any on the
site and a Google
> > search wasn''t fruitful.  Since my ultimate goal is to upload
video to
> > YouTube, what matters is that it actually work, so compatibility is my
> > priority...
> 
> There''s a better way than using an old libtheora.  We anticipated
the
> need for continued compatability with VP3 and the new encoder can be
> run in VP3 compat mode.  In this mode it will not use any features
> that weren''t available in VP3.  This obviously penalizes
efficiency,
> but it does work.
	I was wondering what that VP3 compatibility constant was for!  It all makes
sense now...	Thank you very much.  My videos now work in every respect! 
It''s a shame about the efficiency but it''s better than not
having video at all...
	Actually, I have another (hopefully final) issue to deal with involving a
different kind of efficiency.  My application is encoding in real time and was
computationally expensive to begin with, even before libtheora!  Obviously
libtheora adds enormous CPU time, perhaps even more than the original
application alone!	How much effort was put into computational efficiency for
libtheora?  Do you think there''s much room for improvement?  Do you (or
anyone, please!) think you know somewhere that can improved in this regard?
Thank you so much...
> > If I''m not mistaken, they were encoded using ffmpeg2theora so
that it''s
> > compatible with (the broken) ffmpeg shouldn''t be surprising. 
So basically,
> > no one''s using the "new" specification...
> 
> Plenty of people are using the new spec, but ffmpeg was very very late
> to the party.
	I''m sorry, I thought that ffmpeg was "it."  I mean, does
anyone not use it?  What''s the alternative?  I get the impression that
integrated video encoding is not commonly needed so people just used these
established command line utilities...
	Of course, many thanks to Tim, who answered all my initial questions...
_________________________________________________________________

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/theora-dev/attachments/20100212/02d1a032/attachment.htm

On 02/12/2010 05:13 PM, Id Kong wrote:
> > Date: Fri, 12 Feb 2010 11:19:10 -0500
> > From: xiphmont at xiph.org
> >
> >
> > There''s a better way than using an old libtheora. We
anticipated the
> > need for continued compatability with VP3 and the new encoder can be
> > run in VP3 compat mode. In this mode it will not use any features
> > that weren''t available in VP3. This obviously penalizes
efficiency,
> > but it does work.
>
> I was wondering what that VP3 compatibility constant was for!  It all
> makes sense now...
> Thank you very much.  My videos now work in every respect!  It''s a
> shame about the efficiency but it''s better than not having video
at all...
>
> Actually, I have another (hopefully final) issue to deal with
> involving a different kind of efficiency.  My application is encoding
> in real time and was computationally expensive to begin with, even
> before libtheora!  Obviously libtheora adds enormous CPU time, perhaps
> even more than the original application alone!
> How much effort was put into computational efficiency for libtheora?
>  Do you think there''s much room for improvement?  Do you (or
anyone,
> please!) think you know somewhere that can improved in this regard?
> Thank you so much...
Performance optimization is definitely a priority for the Theora devs
(amongst which I am not...yet, but I don''t mind speaking for them!),
but
quality has been as well.  Real-time encoding of Theora is not currently
as important for most uses as final quality and efficiency are, because,
for the most part, videos are going to be encoded once and then used
many times thereafter, so encoding time isn''t such a big factor in
usability.

Incidentally, libtheora is still faster than, for example, x264 is for
encoding, and decoding Theora video with libtheora is likewise faster
than decoding H.264 video.  At least, this has been my personal
experience with my own personal test samples.
>
> >
> > Plenty of people are using the new spec, but ffmpeg was very very late
> > to the party.
>
> I''m sorry, I thought that ffmpeg was "it."  I mean, does
anyone not
> use it?  What''s the alternative?  I get the impression that
integrated
> video encoding is not commonly needed so people just used these
> established command line utilities...
I think Monty''s reply before this post addressed your concern.  ffmpeg
is "it" for certain things, but Theora development and support have
not
been as much a priority for the core ffmpeg developers as has been, say,
support for H.264.  I do not think they are against it, so to speak,
they just don''t care much for it.  I''m sure if someone else
put in the
work, they''d accept it, though, which has been the case for most of the
recent Theora improvements, if I''m not
mistaken.
>
> Of course, many thanks to Tim, who answered all my initial questions...
And thank you for choosing a free-as-in-freedom format for your video
needs!  The world appreciates your decision!

On Sat, Feb 13, 2010 at 2:22 AM, Id Kong <spam_receptacle_ at hotmail.com>
wrote:
>
>
>> Date: Fri, 12 Feb 2010 08:30:50 -0500
>> From: tterribe at email.unc.edu
>> CC: theora-dev at xiph.org
>>
>> Id Kong wrote:
>> > However, something is still wrong. My DirectShow filter, ffdshow,
can''t
>> > play these ogg files at all and, instead, show a very short
sequence of
>> > random colours, crashing most of the time. VLC player can play
these
>>
>> Unless it''s using a very recent build of ffmpeg, ffdshow does
not
>> properly support Theora. Its decoder was limited to the old VP3 subset,
>> and did not support any new things we included in the spec, despite the
>> fact that it was published in 2004. It only worked because the mainline
>> encoder didn''t use any of these extensions either, until the
recent 1.1
>> release. Most of these issues have been fixed (thanks to David Conrad),
>> though ffmpeg still does not properly support cropping rectangles with
a
>> non-zero offset for video that doesn''t have a multiple-of-16
frame size
>> (they don''t support this feature for H.264, either). Have you
tried our
>> DirectShow filters: http://www.xiph.org/dshow/ ? If these do not work,
>> we would like to know.
>
> Thank you for this information. ?It''s an eye opener!
>
>> > files but I''m afraid that''s not good enough for
my purposes. My
>> > ultimate goal is to upload these theora videos to YouTube but they
have
>> > the same problems as ffdshow.
>>
>> Right. As far as I know, they are using some old build of ffmpeg to
>> transcode Theora. This situation is, shall we say, less than ideal.
>
> Indeed, it''s much less than ideal. ?Do you know where I can find
source code
> for older versions of libtheora? ?I can''t find any on the site and
a Google
> search wasn''t fruitful. ?Since my ultimate goal is to upload video
to
> YouTube, what matters is that it actually work, so compatibility is my
> priority...
>
>> > The reason I still think the problem is with my code is that both
>> > players and YouTube can play other ogg files not made by my
software.
>> > Here are two examples:
>> >
>> > http://www.osnews.com/img/19020/qvga.ogg
>> > http://www.osnews.com/img/19020/720p-theora.ogg
>>
>> Vendor: Xiph.Org libTheora I 20071025 3 2 1
>>
>> These are using an encoder from before the 1.1 release, which is why
>> they work with the broken ffmpeg.
> If I''m not mistaken, they were encoded using ffmpeg2theora so that
it''s
> compatible with (the broken) ffmpeg shouldn''t be surprising. ?So
basically,
> no one''s using the "new" specification...
ffmpeg2theora doesn''t use the ffmpeg theora encoder, but the xiph
theora encoder. And the new ffmpeg also uses the full feature set. So,
all new encoding software uses the "new" specification (i.e. full
specification). If YouTube haven''t updated their encoding pipeline,
that''s a different issue.

Cheers,
Silvia.