Psy RDO: Official testing thread (version 0.6 out!)

[jefrey]

yes it looks like the vbv error:(
damned

[Selur]

Psy RD strength is now a decimal value with default 1.0.
What's the range psyrd is ment to be in 0.0-2.0?

[cyberbeing]

What's the range psyrd is ment to be in 0.0-2.0?

The range for psyrd/psytrellis is currently 0.0-10.0 AFAIK.

[jase99]

I've used PSYRDO 0.6 for several builds but with the latest git (948) it won't apply. I've looked trhough the source and can't figure out whats wrong. Has anyone got a version of 0.6 that applies to 948 or is there a specific build I should stick with when using PSYRDO 0.6?

% git apply -v < ../PSYRDO0.6.diff
Checking patch common/common.c...
Checking patch common/common.h...
Checking patch common/dct.h...
Checking patch encoder/analyse.c...
Checking patch encoder/encoder.c...
Checking patch encoder/macroblock.c...
error: while searching for:
h->dctf.sub8x8_dct8( dct8x8, p_src, p_dst );

if( h->mb.b_trellis )
x264_quant_8x8_trellis( h, dct8x8, CQM_8IY, i_qscale, 1 );
else
h->quantf.quant_8x8( dct8x8, h->quant8_mf[CQM_8IY][i_qscale], h->quant8_b
error: patch failed: encoder/macroblock.c:121
error: encoder/macroblock.c: patch does not apply
Checking patch encoder/macroblock.h...
Checking patch encoder/rdo.c...
error: while searching for:

#define COPY_CABAC h->mc.memcpy_aligned( &cabac_tmp.f8_bits_encoded, &h->cabac.f8_bits_encoded, \
sizeof(x264_cabac_t) - offsetof(x264_cabac_t,f8_bits_encoded) )

static int ssd_mb( x264_t *h )
{
return h->pixf.ssd[PI
error: patch failed: encoder/rdo.c:50
error: encoder/rdo.c: patch does not apply
Checking patch x264.c...
Checking patch x264.h...

[Ranguvar]

Read the patches thread :) There were whitespace changes in the unpatched code since then. You need to patch with the -l option to ignore whitespace.

EDIT: That's for GNU patch, actually. I don't know if there's a similar parameter for git.

[jase99]

Thanks. "patch -p1 -l < ../PSYRDO0.6.diff" worked perfectly. There's the --whitespace option to "git apply" but I couldn't get any of it's available options (e.g., nowarn, fix, etc) to work correctly so I'll stick with "patch".

[CruNcher]

Btw people with Nvidia Cards and drivers (newest Physix released ones) do you also experience strange kernel peaks (allways bellow cpu peaks) @ Playback with VMR9 under Windows XP (SP3,Directx August runtime) ? (and no Kernel Peaks with VMR7) i find this rather strange tough i might open another thread about this.

[Sharktooth]

do it. that has nothing in common with psy rdo...

[Deinorius]

A little question.
Are there any builds for MacOSX with activated psy-rdo?

[refulgentis]

A little question.
Are there any builds for MacOSX with activated psy-rdo?
its trivial, just git pull bla bla bla,follow the steps above for patching, ./configure, make, sudo make install. i really recommend doing it yourself, its nice to have the knowledge, its definitely made a real difference in how I do encoding on OS X to be able to.

if you can't figure it out, PM me and I can lay it out simpler, or if you really need it I'll just upload a build.

[Deinorius]

It would be an interesting experience, but with less time it's difficult. ^^"
And it's not for me. I have no Mac, it's for a friend encoding videos of a convention and it would be too interesting, how these videos will gain quality with psy-rdo. I doubt he can compile.

[techouse]

When will people learn how to use OS X the UNIX way? :'(

[nerdpunk]

When will people learn how to use OS X the UNIX way? :'(

haha: once (quite some time ago) i was asked by a friend to compile something on his macbook... ~2 hours i spent looking where to get a working gcc from (of course i never used a mac before this time)

another thing: can we have a rediffed patch which applies to current git, dark shikari? i don't understand the changes in macroblock.c, so i fear touching it myself

[kemuri-_9]

the r953 version that's floating around the x264 patched builds thread works on the current revision.
here it is courtesy of files.x264.nl: x264_psy_rdo_0.6_r953.diff (http://files.x264.nl/x264_patches/x264_psy_rdo_0.6_r953.diff)

[bob0r]

the r953 version that's floating around the x264 patched builds thread works on the current revision.
here it is courtesy of files.x264.nl: x264_psy_rdo_0.6_r953.diff (http://files.x264.nl/force.php?file=./x264_patches/x264_psy_rdo_0.6_r953.diff)

Remove the force.php?file=./ part, thats for video files and windows only, it may also fuck up the patch.

http://files.x264.nl/x264_patches/x264_psy_rdo_0.6_r953.diff

[kemuri-_9]

well that's how it defaults to generating links for me, so no idea it had bugs like that, never had any problems with it.

[Comatose]

Judging by the file name it probably just flags it as an attachment (via headers) so that your browser always tries to download it (and not open it as a video/text file, f.e.)

[IgorC]

Lately I didn't follow progress around x264 so I loose some statements and I'm tired to read every page. Please, I would ask for some basic information for guys who know.
I use last revision 973 from Ranguvar. Does it mean with --trellis 2 I enable psy trellis 2 because psy rdo is by default.
To start I will try basic combinations
1) psy rdo on/off
2) psy trellis on/off
4 combinations : 00, 01,11, 10

I will also ask check this cli settings for 2 passes if I don't mess something:
x264.exe --threads 3 --pass 1 --progress --stats "x264_stat.log" --qcomp 0.6 --bframes 3 --b-adapt 2 --bime --weightb --subme 7 --keyint 300 --ref 16 --trellis 2 --mixed-refs --8x8dct --partitions all --b-rdo --direct auto --b-pyramid --bitrate 800 --no-fast-pskip --me umh --merange 16 --deblock -1:-1 --psy-rd -o 3x.mp4 Ma.avs

[Audionut]

With the latest revision psy rdo is enabled and psy trellis is disabled by default.

The correct command line is --psy-rd 1.0:1.0 or --psy-rd 0.0:0.0 etc.

[IgorC]

--psy-rd 1.0:1.0 --trellis 2 - means that psy rdo and psy trellis are enabled while
--psy-rd 1.0:0.0 --trellis 2 - means that psy rdo is enabled and normal (not psy) trellis 2 is enabled

Is it correct?

[Audionut]

Yep. And of course subme has to be 6 or 7 for --psy-rdo.

[IgorC]

ok, thank you very much for quick answer.

[dansus]

I do a lot of motorsport encodes which contains panning shots where the car is the focus and doesnt move around in the frame too much but has a fast moving un-encodable backround which is going to blur no matter what.

Am i right in thinking that a higher psy-rdo (+1.1) will allocate resources to the source in focus at expense for backround and i can reduce AQ strength to balance this out a bit if desired?

Also, and sorry if this is ot, but is it possible allocate resources to centre of the frame, ie 80% from centre gets 90% of bits and the rest bleeds out to the edge of frame?

thanks.

[Sagekilla]

You can't do hotspot type encoding, where you specify regions of interest that you'd like to get the majority of bits. It's not part of x264, but x264 should naturally allocate more bits to your in focus portion of the image. The blurred background should end up look like very low frequency details that will get less bits, but not block if you use a little bit of AQ.

[dansus]

What do you class as a little bit?

0.5?

[Sagekilla]

It depends on personal preferences. I consider 0.7 good for me since 1.0 sucks up too many bits while not providing enough of a quality boost, so I would go with 0.3 or 0.4. YMMV!

[mikeytown2]

@dansus
this avisynth script might help you with allocating more bits to the "important parts" of the video
http://forum.doom9.org/showthread.php?t=131920

[K0zi]

I have recently done some tests with various psy-rd and aq strengths. Here are the screens and the commandline:

--pass 2 --sar 64:45 --bitrate 1623 --level 5.1 --stats "[...].stats" --ref 5 --mixed-refs --no-fast-pskip --bframes 4 --b-adapt 2 --b-pyramid --weightb --direct spatial --deblock -3:-3 --subme 9 --analyse all --8x8dct --trellis 2 --psy-rd x:x --aq-strength x --me esa --threads auto --thread-input --progress --no-dct-decimate --no-psnr --no-ssim --output "[...].mkv" "[...].avs"


Source (http://img210.imageshack.us/img210/999/35205955eo4.png)

AQ strength 1.0
--psy-rd 1.0:1.0 (http://img10.imagehosting.gr/out.php/i343091_01AxA.png) --psy-rd 1.0:0.75 (http://img10.imagehosting.gr/out.php/i343092_01AxB.png) --psy-rd 1.0:0.5 (http://img10.imagehosting.gr/out.php/i343093_01AxC.png) --psy-rd 1.0:0.0 (http://img10.imagehosting.gr/out.php/i343094_01AxD.png) --psy-rd 0.75:0.5 (http://img10.imagehosting.gr/out.php/i343095_01AxE.png) --psy-rd 0.5:0.5 (http://img10.imagehosting.gr/out.php/i343096_01AxF.png) --psy-rd 0.5:0.25 (http://img10.imagehosting.gr/out.php/i343097_01AxG.png)

AQ strength 0.75
--psy-rd 1.0:1.0 (http://img10.imagehosting.gr/out.php/i343098_01BxA.png) --psy-rd 1.0:0.75 (http://img10.imagehosting.gr/out.php/i343099_01BxB.png) --psy-rd 1.0:0.5 (http://img10.imagehosting.gr/out.php/i343100_01BxC.png) --psy-rd 1.0:0.0 (http://img10.imagehosting.gr/out.php/i343101_01BxD.png) --psy-rd 0.75:0.5 (http://img10.imagehosting.gr/out.php/i343102_01BxE.png) --psy-rd 0.5:0.5 (http://img10.imagehosting.gr/out.php/i343103_01BxF.png) --psy-rd 0.5:0.25 (http://img10.imagehosting.gr/out.php/i343104_01BxG.png)

AQ strength 0.5
--psy-rd 1.0:1.0 (http://img10.imagehosting.gr/out.php/i343105_01CxA.png) --psy-rd 1.0:0.75 (http://img10.imagehosting.gr/out.php/i343106_01CxB.png) --psy-rd 1.0:0.5 (http://img10.imagehosting.gr/out.php/i343107_01CxC.png) --psy-rd 1.0:0.0 (http://img10.imagehosting.gr/out.php/i343108_01CxD.png) --psy-rd 0.75:0.5 (http://img10.imagehosting.gr/out.php/i343109_01CxE.png) --psy-rd 0.5:0.5 (http://img10.imagehosting.gr/out.php/i343110_01CxF.png) --psy-rd 0.5:0.25 (http://img10.imagehosting.gr/out.php/i343111_01CxG.png)

And then I realized I don't understand how it works :rolleyes:
I can't find any post or guideline here that explains how to use these new options, and therefore, I'd like to request such an explanation :)
I mean, what are higher strengths supposed to do? Keep the source grain or add even more? :)

[LoRd_MuldeR]

I mean, what are higher strengths supposed to do? Keep the source grain or add even more? :)

I think Psy RDO neither tries to get as close as possible to the original nor does it intentionally add grain.

Instead it tries to retain the same "complexity" as the original, making the image look more pleasant for the human eye.

Higher values make the effect stronger: It helps to retain more detail, but also tends to add a new kind of artifacts.

(BTW: The second parameter of "--psy-rd" is for Psy Trellis, which is disabled by default)

[K0zi]

So, should I use higher strengths for grainy movies then?
Or the amount of grain isn't the only factor here and to find the "most appropriate" settings I need to run hundreds of tests each time?

And given the screenshots (settings) above, which one would you pick? :)

[LoRd_MuldeR]

So, should I use higher strengths for grainy movies then?
Or the amount of grain isn't the only factor here and to find the "most appropriate" settings I need to run hundreds of tests each time?

Well, you'll have to find the value that preserves as much grain as possible without adding new annoying artifacts. It's a trade-off...

[vucloutr]

Probably deadzones are used even with trellis=2 because trellis is used only for luma plane compression, and not used for chroma planes.This is correct.

Also note that deadzones are used for RDO when trellis=1, and even in trellis=2, deadzones are used for B-frames unless b-rdo is enabled. Finally, even if RDO is on all frametypes, deadzones are still used for fast P-skip decision.

regarding to this http://git.videolan.org/?p=x264.git;a=commit;h=79194caffdc216e338674d88e50adca2f4ea8fa2 trellis is nowadaystrellis also used for chroma plane compression. (?)

so with --trellis 2, --subme 9 (RDO on all frames) and --no-fast-pskip deadzones is not used, correct ?

[Dark Shikari]

regarding to this http://git.videolan.org/?p=x264.git;a=commit;h=79194caffdc216e338674d88e50adca2f4ea8fa2 trellis is nowadaystrellis also used for chroma plane compression. (?)

so with --trellis 2, --subme 9 (RDO on all frames) and --no-fast-pskip deadzones is not used, correct ?Sounds correct.

This means that if I combine that with QPRD and lambda-based AQ, the entire encoder from top to bottom can be based off lambda values... :sly:

[vucloutr]

seems like a mighty plot although i dont know what it means^^

thanks for that quick answer. :)

[Soshen]

To you think that a low psy like 0.60 are good for anime?

[LoRd_MuldeR]

To you think that a low psy like 0.60 are good for anime?

You can assume that there's a reason why "--tune animation" will lower Psy-RD strength to 0.4 ;)

[Soshen]

Yeah but i made a test and at bitrate 1400 the output was better with 1.0 than 0.4.

I've changed only psy value in the tests and there are so mutch differences: the test with psy-rd 0.4 have so mutch more compression artifacts.

(offcourse i use trellis = 2, ME=multi hex and subpixels refinem. = 07 RD on all frames)

[kemuri-_9]

just because in one case the value you choose is better does not mean it's better for all cases.

[creamyhorror]

I've changed only psy value in the tests and there are so mutch differences: the test with psy-rd 0.4 have so mutch more compression artifacts.
Is your animation very grainy/detailed? Maybe you can post a few comparison frames?

[Soshen]

Well is a new anime dvd version. I encode from original vobs.

I made test with 3 different dvd source.

Real Drive, Ponyo, Nodame Cantabile.

In all 3 test psy rd 0.40 increase the number of compression artifacts specially in dark area or in areas with high contrast, like clots borders.

Maybe with sar encoding have it some problems?

[poisondeathray]

In all 3 test psy rd 0.40 increase the number of compression artifacts specially in dark area or in areas with high contrast, like clots borders.


increase over what? --psy-rd 1:0 ? or --psy-rd 0:0

what settings and bitrate?

what filters & avs script?