Log in

View Full Version : adaptive quantization, what is it and how do i do it?


deets
22nd May 2006, 20:26
hey,
Well im back messing around with the PSP as I have another flight in a month (YEY!!). im doing the usual messing about with bit rates and such, but last time i posted about low bit rates a lot of folk mention adaptive quantization or AQ to its buddies.

Now Ive googled it and searched on the forum, but i couldnt find much out about it. Was it just in an un official patch? if not, could someone give me a little bit of info about it and how i enable it :)

i know im risking sounding dumb and its probably blatantaly obvious in megui, but at least i can tick it off the list ;)

dan

Revgen
23rd May 2006, 00:23
AQ is a way to clear up blocking for "blue sky" areas. It's not like the AQ in Xvid.

It really isn't necessary for anything you're watching on your PSP.

Sagittaire
23rd May 2006, 01:37
AQ is a way to clear up blocking for "blue sky" areas. It's not like the AQ in Xvid.

It really isn't necessary for anything you're watching on your PSP.

In fact AQ is generaly pure HVS optimisation but you can use RDO for AQ too. You can use some mask for that and for example:
- Temporal masking : use higher ratio for motion part
- Spatial masking : use higher ratio for complexe texture part
- Lumismaking, Darkmasking ... etc

In fact AQ + RC done complete spatio-temporal adaptative quantisation for each independant macroblock. Try to find good HVS optimation for that is the Graal in little video encoding world ...

jellysandwich
23rd May 2006, 08:26
could someone give me a little bit of info about it and how i enable it :)

i know im risking sounding dumb and its probably blatantaly obvious in megui, but at least i can tick it off the list ;)

dan

I'd like to know this too... So you're not alone anymore. :)

js

foxyshadis
23rd May 2006, 11:12
Current x264 AQ, in sharktooth's builds:

--aq-tcplx <int>

1-10 being useful values. It compresses harder the areas that aren't referenced in later frames, so that those that are get a higher share.

The old AQ (--aq-strength & --aq-sensitivity) was to solve blocking problems that no-fast-pskip mostly solved. Internal changes have rendered them almost useless, requiring such a higher bitrate for same quality that it's better to skip them and just use the higher bitrate.

deets
23rd May 2006, 12:18
excellent thanks :)

ill give it a try!

Sharktooth
23rd May 2006, 13:03
my builds contain BOTH AQ methods.

desta
4th July 2006, 20:43
I'm trying to use "--aq-tcplx 8" in a recent encode. I'm using MeGUI 0.2.3.2177 and x264 532A. Every time I try to start the encode it throws an error in the queue and displays a non-descript error in the log...



Starting job job1-1 at 20:38:50
encoder commandline:
--pass 1 --bitrate 805 --stats "iceage.stats" --bframes 3 --b-pyramid --direct auto --filter -3,-1
--subme 1 --analyse none --me dia --cqmfile "mp4guys_lrm_v4.cfg" --progress --no-dct-decimate
--no-psnr --output NUL "iceage.avs" --aq-tcplx 8
successfully started encoding
Processing ended at 20:38:51
----------------------------------------------------------------------------------------------------------

Log for job job1-1

unknown option ( )

----------------------------------------------------------------------------------------------------------
The current job contains errors. Skipping chained jobs


... I tried using an older MeGUI (0.2.3.2132) but the same thing happened.

shon3i
4th July 2006, 21:00
Version of x264 532A has removed AQ patch, shartooth find some bug

desta
4th July 2006, 21:50
ahhh right, ok. Cheers for the info. :)

Sharktooth
5th July 2006, 02:46
yes the tcplx patch was causing problems with lossless encoding.

desta
5th July 2006, 03:40
Is it still effective to use the --aq-strength --aq-sensitivity commands?

I'm doing an encode of Ice Age, which is CGI animation. It has a lot of flat coloured areas, which tend to results in gradient lines or quite high blocking for darker areas. I started off using a bitrate of about 800kbps, with mp4 guys lrm v4 matrix, but it didn't seem to be any different than the standard matrix... I'm sure that's obviously not the case for every encode. I then tried bumping the bitrate up to 1200kbps, but still the same.

I wanted to use your eqm avc hr matrix, but i know you recommended not to use bitrates lower than 1400 for full res anamorphic.

So, after that brief life story, would you say the previous AQ commands help reduce the gradients and blocking?

Actually, while I'm here, for high detailed CGI animations, would it be best to use your AE profiles or the HQ profiles? Sorry, I know a question like that doesn't really belong in this thread.

riggits
5th July 2006, 06:50
AE will kill your nice CG details, the deblocking filter is (1, 1).
If you enjoy watching Ice Age with a modicum of detail present then try the CE or HQ presets.

desta
5th July 2006, 19:08
I've been using the HQ profiles anyway, but I was just curious as to when the AE profiles are no longer suitable for animation.