Log in

View Full Version : sex264 (updated almost daily by now)


Pages : 1 2 3 4 [5] 6

begu
13th January 2005, 15:09
Hi,
I just wanted to ask that what colorspace does the sex264 codec (and .264 codecs in general) use?
I presume that it is YV12.
Does the spec allow more bit depth than the previous h263?

I ask this because the current colordepth is not quite good enough for high quality encoding.

Isn't the DVD colorspace generally only 8 bit? And the chroma information is only half of the luma (ie. half the resolution of the luma).

As far as I understand, the DVD could be encoded as better colour resoluion, but it is waste of bitrate, since human eye is sensitive to luminance accuracy. BUT: the colour banding is clearly detectable, because it is only 8 bit.

We have currently much better color depth in our computer due to dx9 and new graphics hardware. Generally the colordepth is calimed to be 32 bit (in grapchics cards) but it is 8 bit / colorchannel. So it is only 8 bits for red, 8 for blue and 8 for green (etc.). But now we can have more colors with better bit depth (new HW). So the limiting factor is the DVD itself, not the HW or display. The current LCD TV sets have better bit depth. For example I have panasonic projector, that has 10 bit colordepth and it is way better than my previous sanyo 8 bit. Again only the DVD or Xvid files, etc. are the limiting factor.

So is it possible to use better colordepth with these new codecs? I assume that these 264 type codecs find their way to these upcoming HD-DVD and bluray discs as well. I know that if the source is good and the encoding is done using high bitrate, the natural dithering smooths up the colorbanding issue. But if the bitrate is low, the codec can't maintain the 'microstructure' of the colorgrades, resulting colorbanding (if using 8 bit depth). Now the better colordepth could improve this kind of banding a huge amount. Even if the 'microstructure' (dithering) is lost in encoding, the better colorspace (10+ bit) reults a picture without color bands.

Well, I'm not a coder, and my knowledge is very limited, but I assume that increasing a colourdepth doesn't take so much bitrate compared to the bitrate needed for the 'microstructure' to be visible to eliminate the colourbands. But I might be wrong. But I guess my point is understandable.

This post is quite messy, but maybe it is useful in someway, or creates some discussion. And finally thanks so much for this sex(y)264 thing :)

708145
13th January 2005, 15:30
I agree with you that 10bit per channel will lead to higher quality but the banding is not a 8bit issue!
It's an issue of too high quantization per post processing.

bis besser,
T0B1A5

akupenguin
13th January 2005, 19:11
Originally posted by JoeBGermany
How can I properly deinstall the programm. I still have blocks with b-frames in 2 pass mode and I want to check if itīs getting better after deinstall and reinstall again. /me shakes his head at the windows attitude of reinstall and hope the problem goes away...
If you really want it fixed, how about sending me a compressed clip exhibiting the bug?

Originally posted by Sirber
the celtic ffdshow I have has a bug with bframes, with AVC and sex264. which is squares So it might be a decoding bug? Anyway, I wrote the H.264 B-frame support in libavcodec too, so maybe I can still fix it. Unless it's a compiler/dshow issue.

Originally posted by gotaserena
anybody played with the 3-pass mode? Is it worth the extra time? I'd say not yet worth the time. (The improvement is around .1 dB PSNR ~= 2% bitrate.) It will provide a bigger gain when we get to using multipass for more than just ratecontrol.

stephanV
13th January 2005, 19:13
Originally posted by JoeBGermany
How can I properly deinstall the programm. I still have blocks with b-frames in 2 pass mode and I want to check if itīs getting better after deinstall and reinstall again.

You are not using PP in ffdshow right?

virus
14th January 2005, 02:21
There have been some updates to the VfW frontend added to the repository, both by me and akupenguin. The GUI is now in sync with the changes in the I/IDR frame management, there are new defaults (slower but top-quality) and there's a new "fast first pass" checkbox which will automatically disable a few flags during 1st pass. Should be effective especially if you encode with slow settings.

Also, you can no longer switch off direct mode prediction for B-frames ("None") so make sure you hit Load Defaults when using newer releases or you'll probably see that field go blank due to the changes :)

Besides, is there someone that would like to have 3-pass encoding added, even if right now it's not very effective? Please let me know.

cheers :)
virus

JoeBG
14th January 2005, 08:11
Originally posted by virus
There have been some updates to the VfW frontend added to the repository, both by me and akupenguin. The GUI is now in sync with the changes in the I/IDR frame management, there are new defaults (slower but top-quality) and there's a new "fast first pass" checkbox which will automatically disable a few flags during 1st pass. Should be effective especially if you encode with slow settings.

Also, you can no longer switch off direct mode prediction for B-frames ("None") so make sure you hit Load Defaults when using newer releases or you'll probably see that field go blank due to the changes :)

Besides, is there someone that would like to have 3-pass encoding added, even if right now it's not very effective? Please let me know.

cheers :)
virus

sound great :)

Doom9
14th January 2005, 08:27
there's a new "fast first pass" checkbox which will automatically disable a few flags during 1st pass. Should be effective especially if you encode with slow settings. Does anybody know which settings those are?

akupenguin
14th January 2005, 08:33
fast 1st pass:
disables partitions P-sub8x8 and B-sub16x16
reduces number of reference frames by a factor of 2 (round up)
if subpel refinement was higher than 3, reduce it to 3. else reduce it by 1.

Sergejack
14th January 2005, 09:50
Could someone get in charge to make builds of the new releases ?
(I don't have much time to figure out every trouble I have with cygwin for now)

celtic_druid
14th January 2005, 12:12
I already put a compile up some hours ago. Even did an installer.

http://www.aziendeassociate.it/cd.asp?dir=/x264

Sharktooth
14th January 2005, 12:41
@celtic_druid: X264VFW.exe is always the latest rev. + the installer?
And what are the differences with sex.264?

P.S.: Mirror up @ http://ebola.gamersrevolt.it/celticdruid/x264/

celtic_druid
14th January 2005, 13:00
Yep, like the XviD builds, the installer is simply whatever the latest .7z is.

Guess sex264 was just pre svn and now that it (Syskin's VFW code) is in the svn it isn't really sex264 anymore.

Irwin
14th January 2005, 13:07
Why you use 7z?

celtic_druid
14th January 2005, 13:15
Why not use 7z? What else should I be using? As I have said before, suggest something that works better and I will switch.

Take x264vfw.revision97
rar = 131kb
zip = 146kb
7z = 112kb

Sirber
14th January 2005, 13:24
Originally posted by Irwin
Why you use 7z? Coz it's simply the best and it's free :D

Doom9
14th January 2005, 13:33
if you use winrar, update to the latest version and 7zip support is built-in (previous versions would make you think they can open 7zip, but then fail after the double click).

JoeBG
14th January 2005, 13:38
Originally posted by Irwin
Why you use 7z?

You donīt use 7zip as Standart? Why?

Irwin
14th January 2005, 14:34
I use zip - one click in Total Commander, it's fast on my machine (PIII 500mhz)
7z is very uncomfort for me - special program, many clicks, etc

celtic_druid
14th January 2005, 14:45
http://www.totalcmd.net/plugring/7zip_plugin.html

Irwin
14th January 2005, 14:48
Thanks!!!

sysKin
14th January 2005, 15:15
Originally posted by celtic_druid
http://www.totalcmd.net/plugring/7zip_plugin.html Ooooooooooo thank you so much celtic_druid :D I really really needed that.

Now, if only 7zip's own gui supported drag&drop (oh wait, maybe it does now?) my happines with 7z would be complete.

From now on, my website's x264 builds will be 7z as well. I considered it before (~10% better compression) but was worried about you not having 7zip. Now, it's too late ;p

Radek

Sirber
14th January 2005, 15:16
I don't think it does... D&D is not that complicated to implement :|

Shinobu
14th January 2005, 15:33
use tugzip.
free and support lot of uncomp&com setting (and 7z of course)

http://www.tugzip.com/

++

buzzqw
14th January 2005, 15:40
[OT on]
I build a little batch file to compress with 7z (without installing the ugly windows version)
---- start
@echo off
FOR %%i in (%1) DO SET extdir=%%~xi
FOR %%i in (%1) DO SET name=%%~ni
IF '%extdir%'=='' DO SET extdir=1
FOR %%i in (%1) DO SET where=%%~dpi
cls
IF '%extdir%'=='1' 7zip a -r -mx9 "%where%%name%.7z" %1/*
IF '%extdir%' neq '1' 7zip a -mx9 "%where%%name%.7z" %1
---- end

save it in your "send to" folders, and now you can compress with 7z by "send to 7z"

the decompression is even simplier

by by
BHH

IgorC
14th January 2005, 15:53
Originally posted by Shinobu
use tugzip.
free and support lot of uncomp&com setting (and 7z of course)

http://www.tugzip.com/

++

maybe this will be even better.
official http://www.7-zip.org/ new beta 2005-01-11 http://www.7-zip.org/dl/7z414b.exe

Irwin
14th January 2005, 16:05
This is x264 thread ! not 7z :) :p

Sergejack
14th January 2005, 16:35
Originally posted by celtic_druid
I already put a compile up some hours ago. Even did an installer.

http://celticdruid.no-ip.com/xvid/x264/

I suggest it is sticked in the first post.

MSlv
15th January 2005, 08:53
I encoded the first VOB (.99gb) from the Triple X DVD and I must say I'm impressed! I encoded with vdubmod using 1 Pass Quality at the default 26 value and all other settings unchanged and audio on Direct Stream Copy (ac-3 audio) and the final .avi was 5x smaller. The high motion scenes were unbelivable! Almost perfect! And no blocks!
What sucked though was the low motion scenes. The background lost a lot of detail and it was pretty blurry.

I can post some pics if you want.

I used VideoLAN to play the file.

The encoding speed was between 4-10 fps with vdubmod at Normal priority setting on my AMD 64 3000+ 512RAM.

All-in-all, it was a great experience. I can't wait for the 1st open-source codecs to become competitive to XviD and Nero's AVC. Or for the Nero codec to become usable outside the nero package, like Xvid or DivX, or x264 (;)) is.

akupenguin
15th January 2005, 09:58
Originally posted by MSlv
The high motion scenes were unbelivable! Almost perfect! And no blocks!
What sucked though was the low motion scenes. The background lost a lot of detail and it was pretty blurry.
Try a 2pass encode. It changes the bit distribution to be more in favor of low motion scenes.

Shinobu
15th January 2005, 10:42
this codec realy become competitive at average and hight bitrate.
do you plan to add a multi thread option ?
i'll test 2-pass in a few days, but 1-pass Q22 is realy great bitrate/quality.

++

Doom9
15th January 2005, 11:29
disables partitions P-sub8x8 and B-sub16x16Is there a way to disable those in mencoder? According to what bond told me about the features in this area, there are just 4x4mv and b8x8mv.

akupenguin
15th January 2005, 11:34
Originally posted by Shinobu
do you plan to add a multi thread option?
It's a possibility, since someone has kindly donated a dual opteron for testing. But I'm still working on algorithmic improvements. CPU optimization comes later (unless someone wants to send patches? ;))

akupenguin
15th January 2005, 11:46
Originally posted by Doom9
Is there a way to disable those in mencoder? According to what bond told me about the features in this area, there are just 4x4mv and b8x8mv.
The mencoder option "4x4mv" is identical to the vfw option "4x8, 8x4 and 4x4 P-frame search" which I have here abbreviated "P-sub8x8".
likewise, "b8x8mv" is the same as "B-sub16x16".

easyfab
15th January 2005, 11:58
I have a compile question.(i'm not a developper)

Default compile works fine with GCC under cygwin.
But I try to optimise the compile (cflags) for mu CPU (barton 2500+).
I have ~15% speed improvment of the available binaries for the moment.
When I add -march=athlon-xp the compile works (no errors) but VDM crash.
I searched and it comes from -msse option.
Must x264 have some sse codes to work with -march? or something else ?
For exemple xvid compile works fine when i use -msse cflags.

Thanks for your work.

akupenguin
15th January 2005, 12:23
I have a Barton 2500+, and I compile under Linux/GCC 3.3.1 with -march=athlon-4. no problems here. (march=athlon-xp works too, with no measurable difference in speed.)

Sergejack
15th January 2005, 12:33
I still have bugs in the videos I encode, some square of the animation just disapears sometime leaving black holes.
It isn't random, it happen the same way as I replay the video over and over again.

Paced
15th January 2005, 12:39
Originally posted by Sergejack
I still have bugs in the videos I encode, some square of the animation just disapears sometime leaving black holes.
It isn't random, it happen the same way as I replay the video over and over again.

Same here, I get black pixels also.

easyfab
15th January 2005, 12:40
Originally posted by akupenguin
I have a Barton 2500+, and I compile under Linux/GCC 3.3.1 with -march=athlon-4. no problems here. (march=athlon-xp works too, with no measurable difference in speed.)

I have GCC 3.4.1 .
I will test with GCC 3.3 version
Thanks

mezzanine
15th January 2005, 12:54
I've changed and old gui i made some months ago to support x264.
The gui is using avs2avi and other tools to encode dvd's. If anyone wants to test it send me a pv. Here is a screenshot

http://fyv12.homelinux.net/fusionyv12.gif

Output container is avi but it can easily changed to mkv:p

hellfred
15th January 2005, 13:18
Originally posted by akupenguin
It's a possibility, since someone has kindly donated a dual opteron for testing. But I'm still working on algorithmic improvements. CPU optimization comes later (unless someone wants to send patches? ;))
I praise that person for his courtesy, as that system will prove to be very useful for you while coding and therefor provide us all with a fast evolving x264 codec.
So Tank You a lot!

Hellfred

hellfred
15th January 2005, 13:27
Originally posted by mezzanine
I've changed and old gui i made some months ago to support x264.
The gui is using avs2avi and other tools to encode dvd's. If anyone wants to test it send me a pv. Here is a screenshot

http://mezzanine1.dyndns.org/fusionyv12.gif

Output container is avi but it can easily changed to mkv:p
That GUI looks promising. Did you already implement all features displayed in the GUI? That it would indeed be an impressing pice of software, and the best way to get a next to one click backup of a DVD that usese x264. But sorry, i have just the leisure to read the forum in my rare spare time, no time for extensive testing. :(
Therefore I hope you find beta tester and release a version for us to play with.
Hellfred

mezzanine
15th January 2005, 13:46
Originally posted by hellfred
That GUI looks promising. Did you already implement all features displayed in the GUI? That it would indeed be an impressing pice of software, and the best way to get a next to one click backup of a DVD that usese x264.

Yes all the features are implemented.

http://fyv12.homelinux.net/screenshots.htm

JoeBG
15th January 2005, 15:54
Originally posted by mezzanine
Yes all the features are implemented.

http://mezzanine1.dyndns.org/screenshots.htm

Looks great, but I miss some functions which would be nice to get integrated when possible :)

1) AAC Audio. Best would be a function for BeSweet like in RipRobot. There you can add your own commandline (I added commandlines for AAC)

2) SRT Subpictures

3) Muxing to MP4 and MKV
Mp4 is the natural Kontainer of AVC and MKV is open source

Muxing for MP4 should work with mp4box.

SirCanealot
15th January 2005, 19:54
Is anyone else having problems with the builds from http://syskin.is.dreaming.org/x264/?

I'm getting the error from VirtualDubMod:

"Cannot start video compression:

An unknown error occured (may be corrupt data).
(error code -100)"

Anyone got any ideas?
I've tried restarting and reinstalling, but the same thing happens.
The older builds work fine, but they don't seem to be mirrored on there any more. -_-

This does it nomatter WHAT settings I have on or off.

Anyone got any ideas?

Thanks...

jdobbs
15th January 2005, 20:34
Are you running Windows 2000? I noticed that it ran ok on my XP machine, but I got that error under my Win2000 machine... don't have a clue why, though.

unmei
15th January 2005, 20:47
SirCanealot, i get this error when i try to encode a video that is not mod 16.

Ark
15th January 2005, 21:48
I had the very same issue, in an attempt of encoding with some settings i wanted to test, like 3 b-frames and 5 max ref. frames (source was mod16).

I changed them to default value and everything was fine, maybe the solution is the simplest thing: hit the default button after a reinstall, maybe you haven't tried it yet :confused:

SirCanealot
16th January 2005, 02:46
Originally posted by unmei
SirCanealot, i get this error when i try to encode a video that is not mod 16.

Yup, that's it.
My source is 704x396 :/
Any idea on when this will be "fixed"?
Edit: acording to my calculator, this doesn't leave me with many 16:9 resolutions to use :/

Sergejack
16th January 2005, 07:04
Originally posted by SirCanealot
Yup, that's it.
My source is 704x396 :/
Any idea on when this will be "fixed"?
Edit: acording to my calculator, this doesn't leave me with many 16:9 resolutions to use :/

When you'll resize your source to 704*384 wich indeed IS mod 16.

Yong
16th January 2005, 07:19
Originally posted by SirCanealot
Is anyone else having problems with the builds from http://syskin.is.dreaming.org/x264/?

I'm getting the error from VirtualDubMod:

"Cannot start video compression:

An unknown error occured (may be corrupt data).
(error code -100)"

I received this message too, don't know why,
If i entered encoding bitrate too low(below 650) for some movies...
ratecontrol problem?
Same problem with mencoder,
can't start video encoding(crash at second pass...) if the maximum quantizer set too high(depend of movies).