Psy-RDO / Psy-trellis test results (clean CGI/game recording)

[Ranguvar]

Please vote (post)! Give details, and take a look at the full clips if you can. Once I get enough votes, I'll do another test with a grainy 1080p Blu-ray with those values that are favored.

Did a test with the short ~lossless Gears clip I made: here (http://ranguvar.ath.cx/GearsEncTest/Short_Brumak_Sample.7z) (Thanks to roo_ for hosting! Please only download if you will use; bandwidth does not grow on trees :)).

Tested with my r977 build (http://forum.doom9.org/showthread.php?p=1184352#post1184352).

Complete settings used (http://paste2.org/p/74902). Complete settings part 2 (http://paste2.org/p/74927). Complete settings part 3 (http://paste2.org/p/75323).

PLEASE USE THESE (one frame can only do so much):
ZIP of script, logs, stats, encoded streams, and DGA index files (a35): here (http://www.rapidspread.com/file.jsp?id=omsknn3qtl).
Second ZIP, which contains 0:0, 0.5:0.5, 0.8:0.8, 1.3:1.0, 1.3:1.3, 1.5:1.0, and 2.0:2.0 (the first one doesn't have them): here (http://www.rapidspread.com/file.jsp?id=aqgworruba). Another (sorry!) ZIP, with the last bunch of tests (1.15:0.3, 1.15:0.5, 1.15:0.7, 1.15:1.0, 1.3:0.3, 1.3:0.5, 1.3:0.7, 1.5:0.3, 1.5:0.5, and 1.5:0.7): here (http://www.rapidspread.com/file.jsp?id=p7a7vm6n3c).


Some example pics (frame 71):
(I recommend Firefox's Ctrl+num_tab feature for quick switching comparison)

Source (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/Source.png)

--psy-rd 0.0:0.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/0_0.png)

--psy-rd 0.5:0.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/05_0.png)

--psy-rd 0.5:0.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/05_05.png)

--psy-rd 0.8:0.8 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/08_08.png)

--psy-rd 1.0:0.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/10_0.png)

--psy-rd 1.0:0.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/10_05.png)

--psy-rd 1.0:1.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/10_10.png)

--psy-rd 1.0:1.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/10_15.png)

--psy-rd 1.15:0.3 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/115_03.png)

--psy-rd 1.15:0.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/115_05.png)

--psy-rd 1.15:0.7 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/115_07.png)

--psy-rd 1.15:1.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/115_10.png)

--psy-rd 1.3:0.3 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/13_03.png)

--psy-rd 1.3:0.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/13_05.png)

--psy-rd 1.3:0.7 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/13_07.png)

--psy-rd 1.3:1.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/13_10.png)

--psy-rd 1.3:1.3 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/13_13.png)

--psy-rd 1.5:0.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_0.png)

--psy-rd 1.5:0.3 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_03.png)

--psy-rd 1.5:0.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_05.png)

--psy-rd 1.5:0.7 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_07.png)

--psy-rd 1.5:1.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_10.png)

--psy-rd 1.5:1.5 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/15_15.png)

--psy-rd 2.0:0.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/20_0.png)

--psy-rd 2.0:2.0 (http://i482.photobucket.com/albums/rr186/Ranguvar13/BrumakPsyTest/20_20.png)


What are your thoughts? :) The only thing that's really sure is that psy-trellis owns, at least for this clip ;)

[Sharktooth]

15_15.png looks better to me (is it 1.5:1.5 or 1.5:0.5?)

[Ranguvar]

Sorry, it was 1.5:1.5 :) Fixed.

[Sharktooth]

well maybe the 1.0:1.0 is better overall... the 1.5:1.5 pic has a sort of noise around the edges.
any chance for a --psy-rd 1.5:1.0 screenshot?

[Ranguvar]

Sure. I'll do 0:0, 0.5:0.5, 0.8:0.8, 1.3:1.0, 1.3:1.3, 1.5:1.0, and 2.0:2.0 :)

[Sharktooth]

it seems from those pics that values higher than x.x:1.0 (psy-trellis > 1.0) tend to produce noise around the edges.
maybe the sweetspot is somewhere between 1.0 and 1.5.

[Ranguvar]

Updated.

EDIT: I think 1.3:1.0 looks the best, but it's close between that and 1.0:1.0.

[Sharktooth]

uhm it's hard... but i think 1.0:1.0 still looks better. 1.3:1.0 is loosing details here and there while it is sharper in other spots.
1.0:1.0 is more consistent but has some chroma noise in the weapon icon.
maybe 1.1:1.0 or 1.2:1.0 will be better but i have no time to do test encodes.

[Ranguvar]

OK, tomorrow I'll test in-depth between 1.0:1.0 and 1.3:1.3, and with 1.5:0.0 to 1.5:0.5, as 1.5:0.0 is quite good.

[LoRd_MuldeR]

uhm it's hard... but i think 1.0:1.0 still looks better. 1.3:1.0 is loosing details here and there while but is sharper in other spots.
1.0:1.0 is more consistent but has some chroma noise in the weapon icon.
maybe 1.1:1.0 or 1.2:1.0 will be better but i have no time to do test encodes.

Apparently the defaults were chosen wisely :)

[Ranguvar]

Apparently the defaults were chosen wisely :)

1.0:0.0 is default right now :)

[Sharktooth]

it used to be 1.0:1.0 but psy-trellis was disabled (0.0) since there were some negative reports

[G_M_C]

Sharktooth: Will you be updating Megui & presets to accommodate these changes, or will you be waiting untill we agree on some "best" values in this thread ?


PS, <<note: Just my opinion>>:
x264's options are getting really complex now; Some options are quite un-understandable to new users an/or are quite confusing. Might be time to re-evaluate all those options and device some new names or some other way to get all te options set.

for ex: --no fast p-skip to yes, could be made much easier to understand by simply calling it --fast p-skip (that you than have to set to no offcourse) ;) . The point beeing that it is more logical, and easier to explain / understand by new users.

[Avenger007]

for ex: --no fast p-skip to yes, could be made much easier to understand by simply calling it --fast p-skip (that you than have to set to no offcourse) ;) . The point beeing that it is more logical, and easier to explain / understand by new users.
Actually it's not more logical because that option was intended as a debugging option iirc, similarly for --no-dct-decimation.
Thus no one should have cause to disable those capabilities of the encoder without very good reason.

[G_M_C]

Actually it's not more logical because that option was intended as a debugging option iirc, similarly for --no-dct-decimation.
Thus no one should have cause to disable those capabilities of the encoder without very good reason.

If i'm not mistaken; I see many commandlines, and most of them have --no fast p-skip to yes, so actually your argument goes wrong there ;)

[Comatose]

1:1 vs 1.5:0 vs 1.5:1.0 vs 1.5:1.5, 1.5:1.0 looks best for me.
Compare the third window from the right in those. The top line is almost completely smoothed in all but 1.5:1.0, where it's very visible :)

[Sharktooth]

Sharktooth: Will you be updating Megui & presets to accommodate these changes, or will you be waiting untill we agree on some "best" values in this thread ?


PS, <<note: Just my opinion>>:
x264's options are getting really complex now; Some options are quite un-understandable to new users an/or are quite confusing. Might be time to re-evaluate all those options and device some new names or some other way to get all te options set.

for ex: --no fast p-skip to yes, could be made much easier to understand by simply calling it --fast p-skip (that you than have to set to no offcourse) ;) . The point beeing that it is more logical, and easier to explain / understand by new users.
--no-fast-pskip (as well as --no-dct-decimation) is an useless option unless you're debugging or want the supreme quality using insane values for other options.
however i like to mantain a cerrain consistency with the encoder options. if you want --fast-pskip, ask the x264 devs and i will do that as well on megui.
If i'm not mistaken; I see many commandlines, and most of them have --no fast p-skip to yes, so actually your argument goes wrong there ;)
if someone says to jump off the bridge, you jump off the bridge? ppl using --no-fast-pskip claiming a better quality should think about it twice or even more...
most of them dont even know what fast pskip is ...

[G_M_C]

--no-fast-pskip (as well as --no-dct-decimation) is an useless option unless you're debugging or want the supreme quality using insane values for other options.
however i like to mantain a cerrain consistency with the encoder options. if you want --fast-pskip, ask the x264 devs and i will do that as well on megui.

:)

Hmm, thats not what i meant to ask you; But i did make an unclear posting, I only wanted to give my optionion on all those options. Ive just recently started to use x264, and it has a very steep learning-curve partly because of all the options. Making it all more understadable/transparent might help the "lower steepness of the learning-curve". But as I said, its my opinion.

And all i wanted was to ask you if Megui will be updated to accommodate the new --psy-rd xx:xx options ;)

[Sharktooth]

nope. one picture or one source arent a good test.

[Episodio1]

Shartooth, wasnt "no-fast-pskip" to avoid blokiness in bluesky areas? Is it not useful anymore then?

[Sharktooth]

that was ages ago. the fast p-skip was then tweaked by the x264 devs coz there was that blocking issue. after the tweak, the blocking is almost non existent and --no-fast-pskip is pretty useless unless you're using insane options.

[Ranguvar]

nope. one picture or one source arent a good test.

I plan on doing a test with the 1080p Blu-ray of 30 Days of Night (quite grainy, dark, etc.) once we narrow down the possibilities with this :) This source is IMO as ideal as a single source can be, but still only 1 source.

And again, I don't think people are noticing, you can download the encoded videos, I put links in the first post. These frames were just examples.

[jethro]

And again, I don't think people are noticing, you can download the encoded videos, I put links in the first post. These frames were just examples.

...File does not exist!
Ok, nvm works now

[CruNcher]

nope. one picture or one source arent a good test.

yep it's very useless this way motion results and their speed/quality/complexity/size have to be taken into account to imho
very nice you can see what --no-fast-pskip does in a extreme low bitrate situation you gonna see that it creates a complete different look in such situations though for high bitrate you wont realize any difference :) you can either chose between more detail though also more motion artifacts or less detail no artifacts then :)

[Atak_Snajpera]

Did a test with the short ~lossless Gears clip I made: here
Why do you play@1280x768(?!?!?!) instead of 1280x720 (16:9) or 1280x800 (16:10). Does your monitor/TV have 1.6666666 aspect ratio???

[Ranguvar]

Yep, it does.

[Quark.Fusion]

Samsung R20 notebook for example have this resolution.

[Ranguvar]

Okay, lots of new --psy-rd values. Please cast your vote starting now, and reasons! (just post)
I highly advise downloading the actual encoded videos and taking a look, as one frame can only do so much.

Once I have some top values, I'll do a test with 1080p grainy Blu-ray footage.


My vote now goes to 1.15:0.7.

EDIT: Actually, while 1.3:0.5 has more edge crap and slight smoke banding, it does far better on the back wall IMO... check the shadows, compare those two against the source. 1.3:0.5 preserves the original shapes better.

[Sharktooth]

agreed, 1.15:0.7 or 1.0:1.0

[Quark.Fusion]

1.15:0.7 while looks sharper have more ringing and chroma artiafacts that 1.0:1.0. Maybe impression of sharpness is from ringing noise.
Edit: 1.15:0.7 actually saves more details, but I don't like that ringing.

[Ranguvar]

Keep in mind, as Sharktooth said in IRC, this is an uber-clean source. A little banding here will probably be killed in a source with any amount of grain/noise. But thanks for the vote :)

[audyovydeo]

Quick question : are these two settings equivalent ?

--psy-rd 0
--psy-rd 0.0:0.0

ie : how is one sure to disable it (apart from setting -m < 6 ?)

cheers
a/v

[kemuri-_9]

--psy-rd 0
--psy-rd 0.0:0.0

yes those are effective equivalents in disabling psy-rd and psy-trellis

[J_Darnley]

yes those are effective equivalents in disabling psy-rd and psy-trellis

--psy-rd x.y corresponds to --psy-rd x.y:x.y
that is both psy-rd and psy-trellis are set to the one value you give.

No, --psy-rd x is eqivalent to --psy-rd x:0

OPT("psy-rd")
{
if( 2 == sscanf( value, "%f:%f", &p->analyse.f_psy_rd, &p->analyse.f_psy_trellis ) ||
2 == sscanf( value, "%f,%f", &p->analyse.f_psy_rd, &p->analyse.f_psy_trellis ) )
{ }
else if( sscanf( value, "%f", &p->analyse.f_psy_rd ) )
{
p->analyse.f_psy_trellis = 0;
}
else
{
p->analyse.f_psy_rd = 0;
p->analyse.f_psy_trellis = 0;
}
}
Which means: if reading x:y or x,y is successful, use x and y and then do nothing; otherwise if reading x is successful, use x and set y to zero; otherwise if nothing can be read, set x and y both to zero.

But to answer audyovydeo's question, yes --psy-rd 0 is eqivalent to --psy-rd 0:0

[kemuri-_9]

ah, sorry my mistake then.
i didn't think of it being changed when it submitted to git. as that was the behavior when it was a patch.

[J_Darnley]

Ah, so it was. I had forgotten that already.