sex264 (updated almost daily by now)

[sysKin]

Thank you everyone for your testing - I've noted all the bugs you've found so far. Unfortunatley I'm busy again so I won't give you an updated build just yet - perhaps tommorow.

Originally posted by Sirber
the encoded clip (at 4FPS, CBR) is plauying with blinking gray frames :STryed, cannot decode. Bust I do the reg tweak above?
I've seen these frames with older directshow decoder. Yesterday's build works well, at least for me.

[Sirber]

How can I update? ffdshow is at it'S latest (Oct 12) :(Originally posted by sysKin
Thank you everyone for your testing - I've noted all the bugs you've found so far. Unfortunatley I'm busy again so I won't give you an updated build just yet - perhaps tommorow.Woot woot :D

[Ark]

A small question: what's the difference between Deblocking filter Strength (A) and (B)? One is aplicated before the other? Or is it some sort of spatial x(horizontal/y(vertical) implementation? Or temporal?

Thanks!

[Sirber]

Would be cool to set in red the "not working" options :)

[Ark]

Originally posted by Sirber
How can I update? ffdshow is at it'S latest (Oct 12) :(Woot woot :D

try celtic_druid's builds here (http://celticdruid.no-ip.com/xvid/ffdshow/), they're more recent (and can decode 2-bframes made videos...)

[Sirber]

Originally posted by Ark
try celtic_druid's builds here (http://celticdruid.no-ip.com/xvid/ffdshow/), they're more recent. URL does not work :(

[Ark]

Yes now it's down :confused: ...try this (http://ebola.gamersrevolt.it/celticdruid/ffdshow/) mirror...

[Sirber]

Thanks :D

[Sirber]

2pass failed: Error: Unable to find the state file.

[celtic_druid]

Site should be back up tomorrow some time.

[Sharktooth]

Fortunately i've mirrored the 2004.01.05 build just before the celtic druid website went down :)
As Ark said just use that mirror to get the latest build.

[hellfred]

Originally posted by virus
Well, thx for the suggestion :)
I've tried it, though I have no idea if this utility is updated or not and whether it should work or not (anyway, syskin uses msvc for compiling and not gcc like me - and in both cases, everything is very slow here).

Here's the output:

x264: MMXEXT against C
- pixel sad : [OK]
- pixel satd : [OK]
- pixel avg : [OK]
- sub_dctXxX : [OK]
- add_idctXxX : [OK]
- (i)dct4x4dc : [OK]
- (i)dct2x2dc : [OK]
mc[0][mv(2,1) 8x16] [FAILED]
mc[0][mv(2,1) 8x8 ] [FAILED]
mc[0][mv(2,1) 8x4 ] [FAILED]
mc[0][mv(2,1) 4x8 ] [FAILED]
mc[0][mv(2,1) 4x4 ] [FAILED]
mc[0][mv(2,3) 8x16] [FAILED]
mc[0][mv(2,3) 8x8 ] [FAILED]
mc[0][mv(2,3) 8x4 ] [FAILED]
mc[0][mv(2,3) 4x8 ] [FAILED]
mc[0][mv(2,3) 4x4 ] [FAILED]
- mc luma : [FAILED]
- mc chroma : [OK]

I wonder if those "failed" messages are relevant or not...

EDIT: maybe it's better to add that I use NASM version 0.98.38, compiled on Sep 12 2003

virus
Looks like you CPU has a notch, as the tests do work on both of my system. Akupenguin is maintaining that peace of code, So if he says the tests are up to date and working, i would believe him :D
But how this is related to your system being slow when encoding videos, i do not know, as i would expect defect CPUs to be as fast as sane ones, but producing damaged frames.
Hellfred

EDIT: Forgot to mention that i am using same version of nasm

[olnima]

Happy new Year to all,
one question looking into the future:
Is it possible someday in the nearer future to use this codec for capturing / realtime-encodings? I'm asking because I do not have any idea about possible speed-improvements and about "how beta is beta". Maybe the kind of encoding is too CPU-hungry for todays CPUs.
In VirtualDub I got ~10fps (Athlon 3000+).

BTW: I couldn't use the codec in VirtualVCR for capturing. Selection is possible, but recording doesn't start.

Maybe Syskin or other "knowing people" can do some "Forecasts"

Thanks alot

Olnima

[Koepi]

With available high-end systems it's not yet possible to achieve realtime encoding.

It'll last a few more years until this will happen (and then we have HDTV instad of PAL/NTSC resolutions so the bigger image eats up all the speed gain - it'll last even longer).

Regards
Koepi

[olnima]

Thanks for reply.
That means to me: stay at good old xvid. And I do NOT have anything against that :-)

Olnima

[Sergejack]

Is the bitstream already defined ? (meaning what we encode now will still work on further releases)

[stephanV]

Originally posted by Sergejack
Is the bitstream already defined ? (meaning what we encode now will still work on further releases)

since they are trying to make an H264/AVC codec i would certainly hope so. :)

(of course, this excludes bugs)

[Latexxx]

Originally posted by Sergejack
Is the bitstream already defined ? (meaning what we encode now will still work on further releases)
It is already standardised. That means that every contemporary encoder which follows the standard can be decoded in the future.

[Sirber]

Anyone got the 2pass working?

[Lordlic]

Originally posted by Sirber
Anyone got the 2pass working?

Yup, it's working fine here. 2pass @ 900 kbit/s and subpixel refinement precision is 5. I didn't touch anything else. I used a dvd source (avs anyway).

(I used the pre-compiled version.)

[CruNcher]

be carefull it won't encode anything non mod16

[virus]

Originally posted by hellfred
Looks like you CPU has a notch
My CPU has no problems at all. In fact, doing a fresh checkout and recompiling everything makes checkasm go fine (but x264 is still very slow).

But considered the problems we (we all, not just me) had in the past using ffdshow (forcing to install alternative versions of ff_x264.dll compiled by celtic_druid to avoid blocking) I'd say that something's wrong in the ASM code, especially the MMXEXT routines which made gcc b0rk badly with optimizations enabled.

Oh, important detail I forgot to mention: here the speed of a build without ASM is similar to those with ASM enabled. Couple this with the amazing speed I get with celtic_druid's ff_x264.dll (and only with it) and maybe you'll have an hint about what's going on...

[easyfab]

Ok here is my little test about x264 options .
I want to know what exactly each options give for results.

The avi source was encode with vdm and sex64 05.01.05 at quant 10

The results:

( size ; ssim ; gain)
all options min (37.5 ; 96.55 ; reference)
all options min + cabac (35.6 ;96.55 ;5%)
all options min +intra 4x4 (37.2 ;96.62 ;~0% but more qual.)
all options min +sub 8x8 (37.5 ;96.55 ;0%)
all options min +sub 16x16 (37.1 ;96.53 ;1%)
all options min +srp 5 (35.7 ;96.56 ;5%)
all options min +max ref 15 (36.9 ;96.54 ;2%)
all options min + all above (32.0 ,96.61 ;15%)


all options min + bf1 (71.59 ??? )

I have prob with bframe so i exclude this option from my test.
The most gain are with cabac and sub ref precision.
And what impress me is that with all options enable the gain is superior of the sum each option alone.
For me, i will always enable all at max.
(Caution, this test was only done of one sample file)

[Ark]

I noticed that b-frames have a huge impact on quality for given filesize, but they don't like very well flames....in fact i encoded 1 minute of the intro of LOTR SEE, and when there're flames it's all a blocky fest (but only in b-frames)...

I can attach a clip if someone want to look...

Settings used were:

- 2-pass@800kbps
- CABAC ON
- Deblocking filter ON with:
- Strength (A) at -4
- Strength (B) at -3
- Max Reference frames 2
- Max B-frames 2
- B-frame prediction mode 1 (temporal)
- MSP with all options checked
- Subpixel refinement precision 5

all other settings at default.

[akupenguin]

Originally posted by easyfab
The avi source was encode with vdm and sex64 05.01.05 at quant 10 quant 10? that's lower than mpeg4 @ quant 1.
all options min + bf1 (71.59 ??? ) That's probably the B-frame delay matching up encoded frames to the wrong source frames.
And what impress me is that with all options enable the gain is superior of the sum each option alone. You can't test options alone. sub8x8 requires sub16x16, and only works well with good sub pixel refinement. multiple ref frames help much more with cabac and spr than they do alone.


@syskin, suggestions:
Disable psub8x8 checkbox if psub16x16 is unchecked.
Remove numbers from "bframe prediction" (the names are the only meaningful part).
While the two deblocking paremeters are theoretically separate (you could say A is strength, B is threshold), I haven't found any use for adjusting them separately; a single slider would be just as good. (Anyone have examples to the contrary?)
bugs: If I click "load defaults", it sets everything to 0.

Originally posted by Ark
i encoded 1 minute of the intro of LOTR SEE, and when there're flames it's all a blocky fest (but only in b-frames)... If you're decoding it in ffdshow, that's partly because we don't yet deblock B-frames.

[Ark]

Yes, ffdshow's decoding...

[eb]

olnima wroteIs it possible someday in the nearer future to use this codec for capturing / realtime-encodings? I'm asking because I do not have any idea about possible speed-improvements and about "how beta is beta". Maybe the kind of encoding is too CPU-hungry for todays CPUs.
In VirtualDub I got ~10fps (Athlon 3000+).


Thanks to Koepi and Xvid, and ffdshow it is possible to encode in real time from digital satelite broadcasting even with such not powerful as my Athlon 2200+ (at2000MHz).
Samples you can see on ftp://www.eb.enterpol.pl
user name www.eb.enterpol.pl
password eb

Latest two sample were recorded on live to Xvid with 1 b-frame with cropping to 720x320 at the same time audio was converted to 6 ch LPCM.
Video was not processed latter but audio was processed to 6 channel AC3, AAC,OGG and mp3 and mp2.
Two samples with this multiaudios are clearly named.

eb

[olnima]

I'm also capturing in realtime using xvid.
Can You tell me something about your hard/software and your configuration? xvid-settings, ffdshow-settings, digital or analog card, etc.

Would be nice if You open a thread in the capturing-section.

Greetz
Olnima

[eb]

Hi olnima,

It is not capturing this is recording, this names i am using to distinguish between video sources.
In my case it is SS2 DVB PC card for digital sat tv.
Programs that are useful for this are SkyView by marfi, MyTHeatre 2.76 by Saar.
I am going to present Graphedit drawings for this two samples and post there is short my post in http://forum.doom9.org/showthread.php?s=&postid=587202#post587202hem in satelite thread of this forum.

eb]

[Sirber]

Where does it store it's analisys file?

[SeeMoreDigital]

Sadly, just like Sirber, I'm only able to generate 1pass encodes too!


Cheers

[Ark]

Originally posted by Sirber
Where does it store it's analisys file?

...just where it stores the first-pass .avi file!
It should be named "x264.stats".

[Sirber]

Originally posted by SeeMoreDigital
Sadly, just like Sirber, I'm only able to generate 1pass encodes too!


Cheers Works now, was an error in my code :) Wasn't calling the right config file for second pass for avs2avi :rolleyes:

[SeeMoreDigital]

Originally posted by Sirber
Works now, was an error in my code :) Wasn't calling the right config file for second pass for avs2avi :rolleyes: I'm not using AVS files... but it (sort of) works for me too!

However, my 2pass encodes "hang" when played in ShowTime... and they don't seem to look as good as the 1pass :(

This reminds me of what used to happen with early versions of VP6 in 2pass mode!


Cheers

[Sirber]

Did you get latest ffdshow? Try in MPC :o

http://ebola.gamersrevolt.it/celticdruid/ffdshow/

So far my encodes are all blocky (big blocks), at 536kbps, anime content.

http://www.detritus.qc.ca/test.mkv

[akupenguin]

Originally posted by Sirber
So far my encodes are all blocky (big blocks), at 536kbps, anime content. First thing to try: don't use more than 1 B-frame, as they're not yet adaptive. (I'm pretty sure the beginning of your clip would be better off with no B-frames.) Also, ffdshow doesn't yet deblock B-frames.

[Sirber]

oh, good, trying :D

[Paced]

Nice work sysKin! But, I'm afraid I have to report something 'wierd.' Sex264 was working flawlessly for me up until I rebooted my computer - for the first time since I'd installed it. Now, VirtualDubMod just seems to abort the process/job 1/4 of the way through the first pass (I'm using the same settings I used before the reboot, and everything is being done the same way). Basically, VDubMod just exits, and when I re-open it to check the Job Control, it says that Job 1 was Aborted :confused: I then proceeded to try other codecs - XviD, VP6, etc. and they all worked fine on the same job.

Note: I also uninstalled sex264 then reinstalled it (without rebooting, not sure if that matters), but the same thing is still happening (Win2K Pro, SP4).

[bond]

two things:

- vd outputs error messages always at the end of each pass (not really an error message, but in the job control, next to each job there is a note "warning", which says "dub: i/o thread has not cycled for 10seconds - possible livelock" or so

- syskin, you might want to rename the subpartition options a little bit to make clear what they enable:
like "enable 8x8, 16x8, 8x16", or "enable 16-8 search..."

[Doom9]

"dub: i/o thread has not cycled for 10seconds - possible livelock"I've seen this a lot recently when I encoded for the codec comparison. It happened with all kinds of codecs and I have no idea what causes it.

[akupenguin]

Originally posted by Doom9
dub: i/o thread has not cycled for 10seconds - possible livelock That simply means that the codec is slow.

[Blue_MiSfit]

x264 is amazing on cartoon material.

I can achieve a 720p encode of Family Guy at less than 1000kbit.

I can do full frame 480p encodes with he-aac and do 94mb per episode to fit the entire series on one DVD-R (50 episodes at 94mb is 4.7 GB).

All this is with light spatial denoising only...

UnDot()
and
UnFilter(-5,-5)

and Lanczos as the resizer always.

Of course this is with all the quality enhancing features enabled, 2 passes, and 15 reference frames, so it runs very slowly encoding even on my overclocked athlon 64.

Works for me at any rate :)

btw, this may have been answered already and I apologize if it has been, but has x264's bitstream been finalized? I dont want to do a whole DVD-R of this stuff only to have future decoders b0rk...

~misfit

[akupenguin]

Originally posted by Blue_MiSfit
btw, this may have been answered already and I apologize if it has been, but has x264's bitstream been finalized? I dont want to do a whole DVD-R of this stuff only to have future decoders b0rk... Yes. x264 outputs H.264 Main Profile, which was finalized in 2003. There may well be bugs in x264, but nothing will intentionally break compatibility.

[Yong]

Crash virtualdub during first-pass:
An out-of-bounds memory access (access violation) occurred in module 'x264vfw'...
...while compressing frame 2468 from 022b0000 to 0a6b0020 using codec "x264 - H264/AVC encoder" (VideoSequenceCompressor.cpp:594)...
...while running thread "Processing" (thread.cpp:150).


Here's my encoding first-pass option:
bitrate=600, No CABAC, deblocking filter -3:-3, Max refrence frames 3, Max b-frames=1, prediction method= 2 spatial, max key frame interval= 300, subpixel refinement precision=0

Can i disable the CPU-hungry option at first-pass(just like the mencoder 2pass first-pass encoding)?

[708145]

constant quant for first pass gives better results than constant bitrate. Maybe you want to try. And very maybe it doesn't crash then as well.

bis besser,
T0B1A5

[JoeBG]

I don´t have any crash. Everything works fine with celticdruids ffdshow filters. I´m making a comparison in the moment between 1 b-frame 2-bframe and 3frame with 5 reference frames. Seems to be, that 3 b-frames brings the best results.

[PlazzTT]

The Last Samurai, 34 second comparison at ~500kbps

XviD (Cruncher's Extra Detail 1.3) (http://eirways.com/x264/xvid2.avi) (2.03MB)
(default settings, QpeL, 2 Bframes)

X264 (with sex264) (http://eirways.com/x264/x2642.avi) (2.09MB)
(defaults but with 1 BFrame (Spacial))

[SeeMoreDigital]

Originally posted by PlazzTT
The Last Samurai, 34 second comparison at ~500kbps

XviD (Cruncher's Extra Detail 1.3) (http://eirways.com/x264/xvid2.avi) (2.03MB)
(default settings, QpeL, 2 Bframes)

X264 (with sex264) (http://eirways.com/x264/x2642.avi) (2.09MB)
(defaults but with 1 BFrame (Spacial)) Are they 1pass or 2pass encodes?


Cheers

[PlazzTT]

2-pass.

X264 used ~511kbps, btw, XviD used 497kbps.

I aimed for 500kbps in both.

EDIT: Here's X264 @ 200kbps if anyones interested.
872kb (http://eirways.com/x264/x264_200kbps.avi) (same specs as X264 clip above, but at 200kbps)

[SeeMoreDigital]

Originally posted by PlazzTT
Here's X264 @ 200kbps if anyones interested.
872kb (http://eirways.com/x264/x264_200kbps.avi) (same specs as X264 clip above, but at 200kbps) Bloody good... And perfectly watchable.

Nice one!