[theora-dev] File size comparisons

[Jonathan Harker]

On 18/10/10 13:26, salsaman wrote:
> Hi all,
> I just did a brief test with 3 different codec combinations:
> 
> 1) ffmpeg sorenson/flv, mp3 audio in flv container
> 2) ffmpeg h264/vorbis audio/matroska
> 3) ogg/theora/vorbis using encoder_example
> 
> Here are the results:
> 
> 10721201 2010-10-16 14:46 origem1.flv
> 20731108 2010-09-13 23:04 origem1.mkv
> 33101703 2010-10-03 11:24 origem1.ogv
> 
> as you can see, flv wins hands down, and it doesn't look too good for .ogv :-(

Unfortunately this is completely meaningless, for many reasons.

1. Your test includes several independent and uncontrolled variables -
you are simultaneously comparing container format, video format, audio
format, audio codec and file size.

2. What source material did you use? At the very least, what was the
duration of the source material? What resolution was the source
material? Did you resize it and if so what resolution did you resize it to?

3. There is no indication what quality settings for the video or the
audio codecs you are using, therefore we have no way to know whether the
filesize difference is because (for example) your MP3 audio was at 64
kbps and the vorbis was at 128 kbps.

4. It's not entirely clear what you are comparing, or even how. Are you
are comparing file container formats (i.e. the Ogg container), or video
format (i.e. Theora)? You seem to be just going by file size, which
isn't going to tell you much at all, especially if you are trying to
compare quality.

If you're comparing video quality then I suggest you keep the audio and
container constant (i.e. use the Matroska container since that will
encode Flash video, H.264 and Theora), and use the same audio stream (or
leave it out altogether). Then make sure your video clips output the
same bitrate. This will probably require tweaking if using
VBR/ABR/constant rate factor etc. H.264 has many knobs which if you turn
up to 11 should produce the best result but at the expense of CPU
decoding. However Theora can now produce video better than MPEG4 Part 2
(DivX et al.) and at least the same and probably better than baseline H.264:

http://people.xiph.org/~greg/video/ytcompare/comparison.html
http://people.xiph.org/~maikmerten/youtube/

I'm certain other list members could chime in with even better
comparisons! :-)

J