[MPlayer-users] Re: divx 6
Corey Hickey
bugfood-ml at fatooh.org
Thu Apr 27 18:41:29 CEST 2006
Matthias Wieser wrote:
>>>>> hello
>>>>>
>>>>> is there any way to use the linux divx 6 binary codec with mencoder ?
>>>>> http://labs.divx.com/DivXLinuxCodec
>>>> No, though if someone were to add support it would be useful for
>>>> testing and comparison.
>>>>
>>>> In any case, you'll probably get better performance and/or quality by
>>>> using XviD or libavcodec MPEG-4, both of which are well-supported in
>>>> mencoder.
>>> Xvid and DivX 6.1 seem to be much faster at better quality than
>>> libavcodec. http://www.doom9.org/index.html?/codecs-main-105-1.htm
>> That depends, and a few things have changed since the Doom9 comparison.
>>
>> 1. Lavc has slightly better overall encoding than it used to.
>
> Same is true for Xvid.
Yeah. I didn't mean that wasn't so.
>> 2. XviD used to avoid B-frames where they would be detrimental much
>> better than lavc. With vb_strategy=2, lavc does just as well, though
>> much more slowly.
>
> "exactly as well" or "it tries to do as well"? Additionally last time I have
> tested vb_strategy=2 it has been buggy.
I examined frames in very high-motion scenes that looked quite bad
without vb_strategy=2. With vb_strategy=2 they weren't B-frames and
looked much much better. I have no idea how I would determine if the
decisions are _exactly_ as good as XviD's; maybe one is better than the
other, but the result is good for both.
If you have problems, did you report them?
>> 3. Lavc places I-frames slightly better than before, with sc_factor=6.
>
> Another option which has to be adjusted manually. Xvid just works.
Yeah. That's why I recommend XviD for anybody who doesn't want to tweak
parameters.
>> 4. XviD's ratecontrol adapts to motion much better.
>
> Very true.
>
>> Now, lavc can use XviD ratecontrol internally. Alternatively, a more
>> XviD-like ratecontrol equation for lavc is vrc_eq=(tex+10^8*mcVar)^0.6
>
> Then why not just use Xvid?
You can, but you don't have to in order to get about the same
ratecontrol performance. I'm not saying lavc is better in this respect
-- just that it can compete.
>> At this point, taking the above considerations into account, lavc will
>> produce video that is nearly indistinguishable from that encoded by XviD
>> (and certainly looks just as good), but at a much slower speed.
>
> That's why I have written "Xvid and DivX 6.1 seem to be much faster at
> better quality than libavcodec".
I'm sorry. I read that as "much faster _and_ better quality than
libavcodec". Either way I read it, though, you seemed to be saying that
XviD and DivX are always better than lavc, which isn't necessarily the case.
>> ...there's a bit more, though: I haven't mentioned qns. qns=2 makes
>> sharp high-contrast edges in encoded material look much better, but
>> slows down the encoding process dramatically.
>
> Much better? The doom9 comparison showed that lavc had problems with fine
> details. But high-contrast edges were no problem for any of the codecs.
I didn't recommend qns for the doom9 comparison; it was too slow.
From what I've seen, lavc without qns behaves the same as XviD in this
respect -- they both have nasty ringing on sharp edges. Presumably the
others do too, which means there was no difference to compare.
>> So, here's my current summary view on all this:
>>
>> - by default, lavc encodes much faster but looks somewhat worse than XviD
>
> "much faster"? The default options of lavc are extremly low quality. Making
> xvid as low quality makes it even faster, so I don't know which one is
> faster in the end. Adjusting lavc and xvid to produce comparable medium
> quality shows that xvid is faster (my own benchmarks) than lavc. With high
> quality options (doom9) xvid and divx are faster, too.
I may stand corrected on the fast-but-low-quality aspect. Last I tried,
I was only able to make XviD about 60% the speed of lavc.
I wouldn't doubt your medium-quality results, and your high-quality ones
match mine.
>> - if you enable most high-quality options, lavc can look as good as
>> XviD but encodes much more slowly
>
> Probably right. But do you have any proove for "as good"?
Only painstaking visual evaluation. That's all it comes down to, really.
> Does lavc support gmc?
For encoding, no. Whether it should or not is a topic I can't address.
> Does it support (a usability feature) like cartoon
> mode?
No, though you can select a different set of options to optimize for
animated material. I don't personally care about built-in presets like
that; there has been discussion from time to time, but nobody has made any.
>> - if you enable qns as well, lavc can look better than XviD but encodes
>> even more slowly
>
> The doom9 contest did not use Xvid's highest quality options.
No, but I use all the highest quality stuff for each codec I can.
-Corey
More information about the MPlayer-users
mailing list