x264's new AQ modes visual quality test

MasterNobody · 7th February 2012, 00:29

I think it is time to test few ideas of AQ on real people and to check if they worth it or not. So I would ask you to compare the visual quality of different AQ modes (--aq-mode option) and order them from the highest visual quality to the lowest on per sample base. So your result can look like:

Quote:

HBO_2000_10bit: aq1,aq2,aq3,aq4

which will mean that on HBO_2000_10bit sample you prefer aq1 the most and aq4 the least (and aq2 more than aq3).

Of course arbitrary comments what you liked or disliked in specific sample also welcomed. You don't need too make all samples but the more the better. Also not all of them are fully unique because for one source I made 4 variants: low bitrate 8bit / 10bit and middle..high bitrate 8bit and 10bit (so you can choose the one which you more interested or where you can see the difference).

Here are samples ("_bak" files inside archives differ only by 0.01 crf value from normal so that one of them have the size above chosen and another one below it for comparison without bitrate bias):
HBO_2000_10bit.zip ~24MB
HBO_2000_8bit.zip ~24MB
HBO_8000_10bit.zip ~81MB
HBO_8000_8bit.zip ~81MB
Lighthouses_1000_10bit.zip ~112MB
Lighthouses_1000_8bit.zip ~112MB
Lighthouses_8000_10bit.zip ~904MB
Lighthouses_8000_8bit.zip ~904MB
park_joy_4000_10bit.zip ~39MB
park_joy_4000_8bit.zip ~39MB
park_joy_10000_10bit.zip ~98MB
park_joy_10000_8bit.zip ~98MB
PlanetOfTheApes_500_10bit.zip ~50MB
PlanetOfTheApes_500_8bit.zip ~50MB
PlanetOfTheApes_2000_10bit.zip ~201MB
PlanetOfTheApes_2000_8bit.zip ~201MB
SourceCode_400_10bit.zip ~40MB
SourceCode_400_8bit.zip ~40MB
SourceCode_1500_10bit.zip ~152MB
SourceCode_1500_8bit.zip ~152MB
ThorSampleRemux_300_10bit.zip ~3MB
ThorSampleRemux_300_8bit.zip ~3MB
ThorSampleRemux_1500_10bit.zip ~15MB
ThorSampleRemux_1500_8bit.zip ~15MB

P.S. For those who want to test there own samples here are Windows binaries/sources: x264_aq_test.zip and patch. Base revision is little bit outdated by now but sample creation need a lot of time (and nothing extraordinary was changed after it [TV range algorithm for bit-depth conversions was patched in this builds also]).

jpsdr · 27th April 2012, 14:10

I thought there was only the mode 3 add, to try to make better results in dark area. But looking at the code patch out of curiosity, i've noticed that there is indeed mode 3 and 4 add.
Can you roughly explain, the idea/purpose behind mode 3 and 4 ? And if , still in the idea/purpose, was for exemple : mode 4 has been made more for xxx situation, and mode 3 more for yyy situation.
As i'm affraid of not being very clear (not fluant in english), i'm not asking "is mode x better than y", but truly the idea, the "what for" you had in mind for these new modes.
Thanks.

27th April 2012, 14:10	#2 \| Link
jpsdr Registered User Join Date: Oct 2002 Location: France Posts: 2,316	I thought there was only the mode 3 add, to try to make better results in dark area. But looking at the code patch out of curiosity, i've noticed that there is indeed mode 3 and 4 add. Can you roughly explain, the idea/purpose behind mode 3 and 4 ? And if , still in the idea/purpose, was for exemple : mode 4 has been made more for xxx situation, and mode 3 more for yyy situation. As i'm affraid of not being very clear (not fluant in english), i'm not asking "is mode x better than y", but truly the idea, the "what for" you had in mind for these new modes. Thanks.