New ffdshow build (?)

[Blkbird]

Since the ffdshow thread has been closed, I've having trouble finding information about new builds. Maybe this thread could serve that purpose.

[Sirber]

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

[CiNcH]

Guess he was asking for information on those builds, like changelogs aso.

[bourtzovlakas]

http://m17n.cool.ne.jp/freeware/mpc/

[Blkbird]

http://m17n.cool.ne.jp/freeware/mpc/

Thanks for that, but a list of comments from a version control system is about the worst kind of changelog - hardly better than none at all. No non-developer can read that kind of stuff. So what I look for is really those human-readable sentences about what has been changed (that affects user in a noticable way), as they were reported in the old ffdshow thread.

[bond]

there is no such changelog for development builds available afaik

[celtic_druid]

Most are pretty self explanatory anyway, like:
MMX libtheora postprocessing
experimental DV output
updated libavcodec

[Shirokuu]

Changes made to the updated libavcodec library (which is incorperated in ffdshow) can be found here (http://www1.mplayerhq.hu/cgi-bin/cvsweb.cgi/ffmpeg/libavcodec/?cvsroot=FFMpeg). Posted just for your convenience.

[Blkbird]

Appearantly there has been a recent fix for some subtitle issues (which I have). But I can't find any build newer than 20050803. Anyone?

[Tirges]

What the difference between:
ffdshow-20050822.exe & ffdshow-20050822MSVC71.exe

Does one (or the other) explicitly need MSVC71.dll ? Or not, as it's included in the build or something?

[Liisachan]

Appearantly there has been a recent fix for some subtitle issues (which I have). But I can't find any build newer than 20050803. Anyone?
I mirrored it already, right after celtic_druid compiled ffdshow-20050822.exe

I was too lazy to post about that, but you can check
http://ffdshow.faireal.net/

Afaik, ffdshow-20050822.exe & ffdshow-20050822MSVC71.exe = the same code, different compilers (ICL vs. VC), there might be speed difference.
Also, altho this should be rather exceptional or accidental, IIRC there was at least one file in the past that doesn't play with the ICL version but does play with the MSVC version of ffdshow.

[Koti]

Appearantly there has been a recent fix for some subtitle issues (which I have). But I can't find any build newer than 20050803. Anyone?Here (http://forum.doom9.org/showthread.php?p=698124#post698124)

What the difference between:
ffdshow-20050822.exe & ffdshow-20050822MSVC71.exe
Does one (or the other) explicitly need MSVC71.dll ? Or not, as it's included in the build or something?
Here (http://forum.doom9.org/showthread.php?p=702556#post702556)

[Tirges]

Ah, thanks!

As I take it: if errors occur, try the MSVC version, might be slower but probably more stable (especially on AMD's...)

[bob0r]

Ah, thanks!

As I take it: if errors occur, try the MSVC version, might be slower but probably more stable (especially on AMD's...)

Agreed, just tested ffdshow-20050822.exe and it crashes mpc with my x264 encodes (AMD XP 1800+ CPU).
With ffdshow-20050822MSVC71.exe no playback problems.

NOTE: ffdshow-20050822.exe on x264.nl is really ffdshow-20050822MSVC71.exe (i just renamed it for layout and general purposes, like stability comes first) !

[Liisachan]

Is the file in question AVC-in-AVI? Or does the same happen for MP4 too? Are there any known 'risky' x264 options? FAQ says Pyramid + Older decoder = doesn't work, but that should be fixed by now. Another report I read was about RDO + Exhausitive
http://forum.doom9.org/showthread.php?p=680780#post680780
Anyone knows any specific options in x264 that ffdshow doesn't like?

[Sharktooth]

high profile stuff...

[Liisachan]

Which options in High Profile? I've tested quite a few AVC files in MP4 with --analyse all --8x8dct (and --bframes 6 --b-pyramid --ref 15), but all of them play fine with the latest ffdshow ICL version. (Although, QuickTime 7 apparantly doesn't like High Profile if complicated)

[Koti]

Anyone knows any specific options in x264 that ffdshow doesn't like?

custom matrices , aside from that it does well I think

[Liisachan]

custom matrices But that's not a bug, but just a known limitation of the current version of ffdshow, isn't it?

aside from that it does well I think Indeed, that's my experience too.

But my experience is still limited, and there seem to be some more complicated options that advanced users are testing... x264.exe has really many switches.

[LigH]

But not enough! (Still missing "-interlaced".) ;) ;)

[Koti]

But that's not a bug, but just a known limitation of the current version of ffdshow, isn't it?
True - Nero does decode, but it's a option ffdshow doesnt like yet.
But not enough! (Still missing "-interlaced".) ;) ;)
Dang , forgot about interlaced since it's not in x264 encode options ( or is it and I missed it ? ) but is spec'd in avc hp.

[LigH]

AFAIK (and bond confirmed), x264 does not yet provide interlaced encoding (Ateme does). But that's already mentioned in the thread about "x264 development"...

[Mug Funky]

interlaced would be a cool thing to have...

[celtic_druid]

Just noticed this: http://cutka.szm.sk/files/ffdshow-20050828.exe
I'll try and put up a new build tonight though, Been a lot of changes since the last one.

[azsd]

downloaded~~thx
the upcoming one with x264 decoder's CQM support?

[celtic_druid]

x264 doesn't have a decoder, it is for encoding only and it already supports CQM's in ffdshow for some time. AVC decoding in ffdshow is done via libavcodec which doesn't support CQM decoding.

http://www1.mplayerhq.hu/cgi-bin/cvsweb.cgi/ffmpeg/libavcodec/h264.c?cvsroot=FFMpeg
When you see CQM decoding support there, then libavcodec supports it and it can be added to ffdshow.

[Liisachan]

Just noticed this: http://cutka.szm.sk/files/ffdshow-20050828.exe
This build is buggy for me.
(1) MP4 (x264) plays too fast. Unusable.
(2) BSPlayer crashes for MKV: BSPlayer is buggy anyway, I don't like it, but I just keep testing it. It should play MKV if you use Haali's splitter (not Gabest's), but it doesn't.
--
Those 2 problems are gone if I reinstall 20050822 back.

[Egh]

Well, according to changes in last weeks milan cutka was mainly doing 64bit enhancements. So i guess that 28th august build is testing-only indeed, unless he used an older source for it.

See that: http://sourceforge.net/tracker/index.php?func=detail&aid=1274652&group_id=53761&atid=471489

Date: 2005-08-29 05:06
Sender: milan_cutka
Logged In: YES
user_id=547197

Disabling YV12 works for me. Try this build: http://cutka.szm.
sk/files/ffdshow-20050828.exe

and more importantly, from the same day:

Date: 2005-08-29 16:36
Sender: milan_cutka
Logged In: YES
user_id=547197

Sorry for the h.264 problems. I forgot to remove code intended
only for development purposes. Fixed now.

[celtic_druid]

Well cvs builds are always potentially buggy. Not sure if I will have time to put up a new build tonight. New mplayer/mencoder builds are up though.

[Liisachan]

Okay, I commented out the link on the 'mirror' site.
This build doesn't seem to be meant for wide use, and nothing good will come out of it if betanews.com/free-codecs.com etc. find it.
Thanks for the cautions.

[azsd]

sorry,I had read hellfred's posts yesterday and misunderstand it's added in mplayer (mplayer use libavcodec as x264's decoder,the same of ffdshow)

[celtic_druid]

[2005-09-09]

SIMD softlight
UINT8->uint8_t
VYUY support in XviD colorspace conversion routines
better support for 8 bit palettized images input in VFW
correct MMX SAD
don't use shuffle instruction in 3dnow! code
dotproduct.asm
successfully compiled 64-bit version - untested
updated libavcodec, x264
don't always enable YV12 output in VFW mode
fallback C idct and fdct for DCT filter
few 64-bit compatibility fixes
a step closer to x64 build
don't crash with zero delay in dolby decoder
few vobsub fixes,
fixed YUY2 levels crash
updated libtheora to alpha5
updated vs 2005 project
working on avc in mpeg support
x264 patches from Sharktooth's builds
2xSai and hq2x resizers
Don't try to always get output buffer size for audio samples - use last known value. Should fix compatibility problems with some applications, hopefully won't break other.
libtheora decoder supports radlight parser
maybe not very useful and not fully correct but: h.263 and mpeg quantization and dequantization in DCT filter
aspect ratios in OSD
display current quantization matrices when available
use xcm extension for supplied quantization matrices,
cqm extension support
updated libavcodec
updated VS 2005 projects

[Inventive Software]

Anybody know where I can get the sources for the builds Celtic_druid so kindly provides? I'm keen to look into them and see what's there...

[celtic_druid]

View here:
http://cvs.sourceforge.net/viewcvs.py/ffdshow/ffdshow/

To get:
cvs -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/ffdshow login
cvs -z3 -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/ffdshow co -P ffdshow

or cvs -z3 -d:pserver:anonymous@cvs.sourceforge.net:/cvsroot/ffdshow co -P -d"2005-09-09" ffdshow if you wanted the exact sources for 2005-09-09 and the cvs had been updated since.

[moonman]

ffdshow 2005-09-09 mirrored here (http://svartling.hopto.org/index.php?q=2005/sep/ffdshow-2005-09-09) .

[Chaos Creator]

It seems to be something wrong in ffdshow with theora! I am doing a 2 pass encode and no matter what value I choose for the filesize in the second pass, I always end up with videos that have the same exact filesize! There is no problem with previous releases of ffdshow!!! The only reason that I use the latest ffdshow 20050909 is because it has theora-alpha-5 in it!!!

Sorry for my bad english!

[azsd]

IMHO the -d"2005-09-09" options will only put newest (2005-09-15) source into overided "2005-09-09" folder,it can't do a snapshot from 2005-09-09

[celtic_druid]

Yeah, should be -D2005-09-09.

[hellfred]

It seems to be something wrong in ffdshow with theora! I am doing a 2 pass encode and no matter what value I choose for the filesize in the second pass, I always end up with videos that have the same exact filesize! There is no problem with previous releases of ffdshow!!! The only reason that I use the latest ffdshow 20050909 is because it has theora-alpha-5 in it!!!

Sorry for my bad english!
For encoding videos with Theora codec on win32, one can use ffmpeg2theora, too. It will take any input video supported by libavcodec in quite a lot of containers. If you want to give it at try, get it here (http://www.v2v.cc/~j/ffmpeg2theora/) .

Hellfred

[Chaos Creator]

I want to use ffmpeg2theora but it doesn't accept AVS files and it doesn't support 2-pass!!! :( :( :(
AviSynth support in our days is a must have, I believe!

[celtic_druid]

2nd pass filesize set to 5,000kB
results:
avi = 5,014kB
ogg = 5,042kB

2nd pass filesize set to 2,000kB
results:
avi = 2,154kB
ogg = 2,161kB

Seems to be working ok to me.

[Chaos Creator]

For me:

2nd pass filesize set to 8,192kB
results:
8,164

2nd pass filesize set to 12,288kB
results:
9,116

2nd pass filesize set to 10,240kB
results:
9,116

2nd pass filesize set to 20,480kB
results:
9,116

2nd pass filesize set to 5,120kB
results:
5,136

I tried many different filesizes! It seems that if you set a filesize that ffdshow thinks is much more than needed, then it cut it down to what it believes that is better, but the quality of video is always bad!!!! :confused: :confused: :confused:

[LigH]

Do you understand the meaning of "saturation"? 9,116 KB seems to be the size for a constant lowest-possible quantization, therefore maximum quality possible. So, try to compare with CQ more if the results are similar -- if not, then it may indeed be wrong.

[Chaos Creator]

Do you understand the meaning of "saturation"? 9,116 KB seems to be the size for a constant lowest-possible quantization, therefore maximum quality possible. So, try to compare with CQ more if the results are similar -- if not, then it may indeed be wrong.

I will post some frames to see your self!!! ;)

[LigH]

Would be nice.

And, BTW: Please, don't misunderstand me - sometimes I sound a bit demanding, but just accidently, I'm not a native english speaker...

[Chaos Creator]

Hey, I am not a native english speaker either!!! :D
I didn't misunderstood you!

Ok, let's see!!!

Source
Source (http://img310.imageshack.us/img310/8117/source9gg.png)

--------------------------------------
Filesize Set to 20,480
Actual filesize 20,496
encode1 (http://img310.imageshack.us/img310/5294/204802vj.png)

--------------------------------------
Filesize Set to 25,600
Actual filesize 23,368
encode2 (http://img310.imageshack.us/img310/3777/256002tp.png)

--------------------------------------
Filesize Set to 30,720
Actual filesize 23,368
encode3 (http://img310.imageshack.us/img310/1899/307205bt.png)

Yes, the last two results (which have the same exact actual filesize) are similar! It seems that we have the maximum quality possible!
Then the only thing I've got to say is that I am really dissapointed from the quality! :( :( :(
The first time I saw all this detail to go away, I thought that I've done something wrong!!!!

[movax]

That loss might be what it looks at max quality or whatnot...you still don't want to saturate the codec. Try using some filters to fix it up instead.

[Chaos Creator]

That loss might be what it looks at max quality or whatnot...you still don't want to saturate the codec. Try using some filters to fix it up instead.

I wanted to see how good can this codec compress a video! Using filters to saturate more, is not an option for me! I don't like to put filters in my dvd backups for any reason! I really prefer to wait a year or two of development and see the codec to encode better and to keep more detail on its own!! I know, it might sound strange what I am saying, but that's me!!! ;)

[Chaos Creator]

I tried to make some videos with ffmpeg2theora 0.15 with (--videoquality 10) and the video quality I got was really great! I think now that there must be something wrong with ffdshow. Maybe in the 2-pass mode it provides! Ffdshow doesn't keep any quality, even at highest filesize I set to it. Ffmpeg2theora gave me great results!!!

[Liisachan]

celtic_druid's 20050920 is out.

I got 500GB/mo bw for this mirror, but (it's a darn cheap server and) might be slower.
http://ffdshow.faireal.net/

This server might be faster, tho the bw is more limited here:
http://m17n.cool.ne.jp/freeware/mpc/