ffdshow development #2

[BlindWanderer]

Originally posted by oddball
Is there a way for ffdshow to crop black borders from uncropped encodes? CUrrently I use DiVXG400 to do it as it can crop into the overlay. But it would be great if ffdshow had this ability.

you mean to crop the frames automaticly? It would be an interesting feature with "Allow output format changes during playback" turned on. But i'm thinking of the decoder.

[oddball]

It's something I have asked for for some time. However I was told it's not possible which is somewhat odd since DiVXG400 can do it no problemo.

[vlada]

I believe it cannot be so hard to crop black borders from movie if FFDShow can add them.
But I have another 3 problems. The first one is displaying of itallic subtitles. The last letter is always cropped. Why?
The other problem I already mentioned before. I use tray icons for matroska splitter and for audio/video ffdshow decoder. So I have 3 icons there. If I stop a movie a start another one, without closing the player, I get another 3 tray icons. Btw. I unchecked "Make DS graph available to graphedit". What else could be the problem?
And the last thing is a problem with embedded subtitles and Haali's matroska splitter. FFDShow is trying to display the embeded subtitles end last used external once at the same time. Also switching subtitles in the splitter causes playback to pause, but this is not a big problem.
Vlada

[namchik]

Why the size of latest (???) ffdshow 2005.03.28 is so small compared to previous builds (1.8 Mb against 3+ Mb)... What features are removed? :confused:

[ernstblaauw]

I installed the latest build of celtic_druid (20050312) to use MakeAVIS with DVD2SVCD. But when I open a fake AVI made by MakeAVIS, DVD2SVCD sais in the preview window: 'could not decompress frame 0' (this is an Avisynth error). I never earlier used ffdshow, but a guy in another forum told me that this problem was introduced at the moment ffvfw.dll was integrated in ffdshow.
But maybe there is a workaround or solution available. Does anyone here uses DVD2SVCD with MakeAVIS? Is a solution available, or is this a known bug?

[timeismoney]

Originally posted by namchik
Why the size of latest (???) ffdshow 2005.03.28 is so small compared to previous builds (1.8 Mb against 3+ Mb)... What features are removed? :confused: Nothing removed, I compiled it via VC2003 in Release Win32 mode, not use ICL, so its size is small.

Here is a new version as ffdshow-20050405, of course unofficial version but fully cvs compiled, on default libfaad2 is removed, instead of realaac, I don't know if it can really fully replace libfaad2, but test some mp4/aac, sounds not bad. The tta decode function is pretty, but also need a splitter.

ftp://stormdl:stormdl@dl.yesure.com:3721/ffdshow/ffdshow-20050405.exe

[namchik]

timeismoney
but the size of installed ffdshow also differs: only 4.26 Mb .. previous builds were about 8 megs :confused:

[timeismoney]

Originally posted by namchik
timeismoney
but the size of installed ffdshow also differs: only 4.26 Mb .. previous builds were about 8 megs :confused: Yes, if i use a different compiler, the size will be increased, based on vs2003 itself, the performance maybe not as good as ICL, but the compatibility is not bad.

[samab]

timeismoney,

It's great that you make available your compiled versions for the public. One thing I like to add (from a brand new forum member) is that when your compiled versions hit the various codec websites (like free-codecs.com), those websites announced it as anonymously compiled with the big careful sign in capital letters.

Even though both you and Celtic compile "unofficial versions", the public treats the two versions very differently --- because they can't put a face on your anonymous versions.

I saw the 03-28 version of free-codecs.com and I decided against downloading it because I didn't know its origins. But now I know you compiled it and a quick search on this forum would immediately confirm that you are legit.

It's as simple as telling free-codecs.com that you compiled it and they checked your "credentials" by doing a quick search here in this forum.

[samab]

Milan is doing major changes to the source code, so he decided to suspend doing daily builds for a short while.

http://sourceforge.net/tracker/index.php?func=detail&aid=1170802&group_id=53761&atid=471490

[CruNcher]

Originally posted by timeismoney
Nothing removed, I compiled it via VC2003 in Release Win32 mode, not use ICL, so its size is small.

Here is a new version as ffdshow-20050405, of course unofficial version but fully cvs compiled, on default libfaad2 is removed, instead of realaac, I don't know if it can really fully replace libfaad2, but test some mp4/aac, sounds not bad. The tta decode function is pretty, but also need a splitter.

ftp://stormdl:stormdl@dl.yesure.com:3721/ffdshow/ffdshow-20050405.exe

Why do you replaced libfaad2 with Reals Decoder that one doesn't support parametric stereo, or does it?

[timeismoney]

Originally posted by CruNcher
Why do you replaced libfaad2 with Reals Decoder that one doesn't support parametric stereo, or does it? So milan tell it as:

And about realaac: I'm not sure, but for few sample files I tried (2 and 6 channel, LC and HE) it worked correctly and fast.
I already removed libfaad2 from installer (but I didn't enfore use of realaac if ff_libfaad2.dll is found)

Now the original nsis script of the installer has marked libfaad2 out of the package, and I tested some aac audio files or mp4 files I can find, realaac lib can support them correctly. But I don't know if my test is enough, so please try it, the report will be helpful to milan.

Originally posted by samab
timeismoney,

It's great that you make available your compiled versions for the public. One thing I like to add (from a brand new forum member) is that when your compiled versions hit the various codec websites (like free-codecs.com), those websites announced it as anonymously compiled with the big careful sign in capital letters.

Even though both you and Celtic compile "unofficial versions", the public treats the two versions very differently --- because they can't put a face on your anonymous versions.

I saw the 03-28 version of free-codecs.com and I decided against downloading it because I didn't know its origins. But now I know you compiled it and a quick search on this forum would immediately confirm that you are legit.

It's as simple as telling free-codecs.com that you compiled it and they checked your "credentials" by doing a quick search here in this forum. Thank you

[STTS]

Hi all.
Recently I discover the same feature (precision frame interpolation based on motion vector info) in hi-end LOEWE TV. TV owner suggest me to try WinDVD 6.0 with enabled DNM. I am be very impressed - DVD video look so smooth, so natural motion , I start to reviw all my DVD collection. Of course, it requires more CPU power (>>1.2 Hhz).
I am not familiar with mpeg1(2,4) codek internals but *feel* that if we have some motion information discovered during encoding and already stored in video stream. Is it hard to interpolate frame not onliy in I, P, B time, but between, in eventual time ? I think that any player can easy check for output device vertical blank interrupt and pass info to decoder. Decoder will use two nearest frames plus motion info, and produce frame that percision correlate with vb interrupt.
Sorry if it OT, but ffdshow seem to be best place for such features.

[namchik]

timeismoney
i can not download your new compiled build of ffdshow from your ftp... Maybe some other place?... Thanks.

[timeismoney]

Originally posted by namchik
timeismoney
i can not download your new compiled build of ffdshow from your ftp... Maybe some other place?... Thanks. Try here:

http://www.sbtop.com/UploadSoft/ffdshow-20050406.exe

Just updated, seems good. But as it is an unstable cvs compiled version, so use it as your own risk, thanks.

And I think more report will be helpful for milan.

[namchik]

timeismoney
Thanks... but this link is too slow... even on my dialup :(

[planet1]

http://ffdshow.sourceforge.net/tikiwiki/tiki-read_article.php?articleId=18


... :rolleyes:

[bond]

Originally posted by planet1
[B]http://ffdshow.sourceforge.net/tikiwiki/tiki-read_article.php?articleId=18that was the build where only the libav.dll was updated afaik and yes it was borked

[samab]

Originally posted by planet1
http://ffdshow.sourceforge.net/tikiwiki/tiki-read_article.php?articleId=18


... :rolleyes:

This is why I posted the above comments on why timeismoney should put a "face" on his compiled versions (i.e. telling milan and free-codecs.com who he is). At the very least, I think that you should name the file as ffdshow-20050406-vc2003. Celtic compiles various xvid versions with ICL 7 and ICL 8.1 --- and he names them differently.

It would be so much easier for everybody if there is more communication --- however I do understand that this is an open source project with volunteers contributing their free time. Milan should have said that he decided to stop daily build for a short while because of major code changes --- on his website, instead of burying it deep in some bug report comments.

[timeismoney]

To all points, I just compiled it for simply test as I think it is a great job by milan, to see many new functions have been there and no chance to use make me be anxious, so I simply build it for myself (a lazy woman so use the easiest VC71) and put it on my homepage with marked "Unofficial", "cvs build" & "VC71". I never announced it to be universal public until it has been transshipmented to other website like freecodecs.com by some friends, and someone here asked something for it. So I'd do something to cancel it, and tell milan my sorry

[samab]

Originally posted by timeismoney
So I'd do something to cancel it, and tell milan my sorry

You shouldn't feel sorry for it. You are nice enough to provide us with your build.

This is just a classic lack of communication by various volunteers.

I have said it twice before that it would be much easier for the rest of us if milan just announced on his website that he is temporarily suspending his daily build because of big code changes. Milan didn't explain his reasons to the public and we EAGERLY grab either celtic's 0312 release or your 0328 release without knowing that the source code is broken because of major code changes (and then of course all three of you gets the blame for unexplained crashes).

And milan has already stated that it's as much his fault of not writing down simple instructions on how to compile this open source project.

We appreciate your kindness in providing us with your builds.

[vlada]

Hello,
I have problems with MJPEG and DV encoders in FFDShow. I tried encoding to DV from VD (720x576@25, no audio) with latest build (20050406) and with some older ones. But the result is completely broken. It even sometimes crashes VD or player (FFDShow is used for decoding). Is this a known bug?
And now to MJPEG - it has a very poor performance. My computer (Athlon XP2000+) needs 80%-100% of CPU power for realtime capture in 768x576@25. It even sometimes drops frames. I remember I was able to capture full PAL video using Pegasus PicVideo MJPEG on Duron 600 MHz. This is a huge difference in performance. Could the MJPEG in FFDSHow be improved?
Thank you,
Vlada

[namchik]

I also noticed poor performance of MJPEG encoder in ffdshow 03.28... in milan's 03.03 it's ok and no dropped frames with my TV-tuner capture. By the way, milan explains this:The whole package is build in Visual Studio .NET 2003 with Microsoft compiler in Release target. While this isn't bad choice for ffdshow.ax itself and several other libraries included in VS.NET solution, it's inapropriate for building libavcodec and libmplayer.dll. Those two libraries are essential for good performance of ffdshow - libavcodec for decoding/encoding most of video and audio formats and libmplayer for software scaling, many colorspaces conversions and fast mp3 decoder. When those two libraries are build by Microsoft's compiler, all hand optimized code developed to effectively use MMX, 3DNow and SSE extensions is skipped resulting in much worse performance.

[timeismoney]

Originally posted by namchik
I also noticed poor performance of MJPEG encoder in ffdshow 03.28... in milan's 03.03 it's ok and no dropped frames with my TV-tuner capture. By the way, milan explains this: Thanks all who care it, I've contacted milan and luckily heard that two ffdshow builds in the daily builds section will be released soon: a SSE ICL build as usual and a build using uClibc++ as STL library.

[vlada]

namchik> I know about this incorrectly compiled build, but I'm not sure if this is my problem. I noticed the poor performance of MJPEG even in the last build (which should be O.K., I believe) and also in all previous builds I tested. So it seems to me as a problem with optimization of the codec.
Also could anyone else test the DV codec, please? Decoding works correctly, but the encoder is probably corrupted.
Regards,
Vlada

[madman1980]

I have two questions.

Is it possible to include all the different compiles that I see (sse, regular, sse2 etc) in one installation file? And just let it detect CPU itself. That would make things a lot easier for most people.

Will there be a compile for Windowx XP x64 soon?

[Terka]

which filter should i use to achieve high quality for interlaced video playback at 50fps?

[magicclue]

http://100fps.com/

scroll down a little...

[bob0r]

http://mirror02.x264.nl/ffdshow_reg_error.jpg

Celtric Druid, can this be fixed? If so please put a new version on http://celticdruid.no-ip.com/xvid if you could.

This error has been confirmed by more people, running installer again will "solve" the problem.

[Vitos]

Originally posted by bob0r
Celtric Druid, can this be fixed? If so please put a new version on http://celticdruid.no-ip.com/xvid if you could.
This error has been confirmed by more people, running installer again will "solve" the problem. [/B]

Where did you read that's Celtic Druid's compile? You can't find it on his page. It's unofficial compile (you can read about it earlier in this thread) and isn't recommended at all.

[bob0r]

Originally posted by Vitos
Where did you read that's Celtic Druid's compile? You can't find it on his page. It's unofficial compile (you can read about it earlier in this thread) and isn't recommended at all.

Thats what i figured, since it was not on his main website.
A well know doom9 mod gave it to me :o

I got it from http://esby.free.fr/CelticDruid/ffdshow-20050406.exe
I figured it was his.... i PMed him also, so we can sort it out, thanks for the reply.

What is the best version you recommend now?

[namchik]

i use this:
http://cutka.szm.sk/files/ffdshow-20050303-sse.exe
works fine...

[celtic_druid]

The filesize is way too small to be my compile.

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

[namchik]

2 celtic_druid
in ffdshow-20050406 (http://m17n.cool.ne.jp/freeware/mpc/1/ffdshow-20050406.exe) there's a problem with russian fonts in the installer when choosing the components (though GNU license is OK)

Also i capture video with tv-tuner using ffdshow encoder (mjpeg). Builds after 20050303 are very CPU consuming during real-time capture... about 95-100%. With older builds CPU load was about 75%

ps. Sorry for my bad english :rolleyes:

[ExtraEye]

what's the difference between milan's ffdshow build to celtic druid's ffdshow build?

[namchik]

2 celtic_druid
i restarted my PC, now cpu load is normal... my apologies

and problem with russian font still exists, but i'll ask other russians, coz maybe something is wrong with my "MustDie" =)

[celtic_druid]

Other than adding msvcrt7.1 to the installer I didn't touch it so any problems with it you would need to contact Milan. Unless it has something to do with me compiling the nsi script?

The difference is that Milan's build is now over one month old and there have been lots of changes since then.

[bob0r]

@celtic_druid

Thanks for clearing everything up.

I have put your "new" version on http://x264.nl which should work again (when your DNS is updated)

Keep up the great work!

[ExtraEye]

is ffdshow in trouble too?
http://ffmpeg.sourceforge.net/
http://www.mplayerhq.hu/homepage/index.html

i sure hope things woun't change...

[esby]

Well with the european patent problem, every software that deal with patented material will be in trouble... meaning allmost every piece of software, including the open source ones.

And since we are in the multimedia field, everything is patented or something near that.

esby

[Liisachan]

I once asked Matroska devs why ffdshow in their packs were so old. and this (http://lists.matroska.org/pipermail/matroska-devel/2004-July/001828.html) is their reply:

The big problem is, all newer versions of ffdshow are coming ( which i
find VERY annoying to be honest ) with FFvfw bundled, and for licensing
reasons we definitely decided againt distributing a working MPEG4
encoder in our packs. Some of our team are family men or have a job, we
cant risk to have the MPEG-LA standing outside our houses and trying to
sue us for millions of US$'s of lost revenues, because matroska pack is
distributed several 10.000 times per months.


In this theory, I will be sued too eventually for distro'ing celtic_druid's builds on http://m17n.cool.ne.jp/freeware/mpc/

Software patents are getting really weird :(
libdca is another example of many
http://www.via.ecp.fr/via/ml/libdca-devel/2004-12/msg00001.html

I hope at least people can distribute source codes freely, as they are just plain texts after all.

[Mug Funky]

it's good to be careful, but (no offense to matroska project which i think is a good idea, albeit full off politics) i doubt anybody is using mkv for commercial purposes.

[iron2000]

Decided to try Celtic's 20050406 build.

The size is bloated to 10.43MB from the already big 8.1MB!!
Whats going on?

[PatchWorKs]

A possible solution ?

Decode -> all formats
Encode -> just free-patented ones

This can convince devs to migrate towards Theora/Dirac/SNOW/anyotherfree that is not so bad... ;)

[celtic_druid]

The installer is actually smaller than my last release.
5,537,792 ffdshow.ax
2,426,880 libavcodec.dll
536,576 ff_libfaad2.dll
385,024 ff_x264.dll
368,128 libmplayer.dll
348,160 msvcr71.dll
208,896 ff_libdts.dll
192,512 ff_theora.dll
188,416 TomsMoComp_ff.dll
159,744 ff_libmad.dll
159,232 ff_realaac.dll
131,072 ff_samplerate.dll
112,640 libmpeg2_ff.dll
95,744 ffvdub.vdf
91,648 ff_unrar.dll
76,288 ff_liba52.dll
74,240 ff_tremor.dll
35,840 ff_wmv9.dll
25,600 makeAVIS.exe
7,680 ffavisynth.dll
5,632 ff_vfw.dll
4,608 FLT_ffdshow.dll
4,096 ff_acm.acm

If I were to make it SSE2 only then I could shrink it a fair bit I would imagine. The way that it is at the moment it should work on all CPU's and the ICL parts should make use of SSE/SSE2, etc. if available, at least if you have an intel CPU. Second largest file is libavcodec.dll though which is just a generic gcc compile.

If you are worried about the size, stick with the MSVC built version that is floating around.

[esby]

Encode -> just free-patented ones

Well that's a nice goal, but not very useful, since you probably won't be able to use any of the codecs available...
Mpeg4 is patented, wmv & qt are also...,and I am not talking of possible sub licences affecting the whole.
And of course, even the codecs you are naming might be patented...

esby

[Liisachan]

Originally posted by PatchWorKs
A possible solution ?

Decode -> all formats
Encode -> just free-patented ones

Maybe not. as libdca (decoder for DTS) devs say: "The patent might be challengeable (at least for the decoder side) because it doesn't describe any invention as such but more a format of data ... However as everybody knows, it is a lengthy and expensive process to try to challenge (even a bogus) patent so ..." :(

This can convince devs to migrate towards Theora/Dirac/SNOW/anyotherfree that is not so bad... ;) That's exactly what I'm thinking, too. And let's not use MP4 or WMV or RV, let's prefer OggFile/OGM/MKV/NUT etc...as long as they can do the same job.

"The sad reality of free software is that it may take drastic measures such as these to make the four freedoms important enough to value."
(Your software rights or the best tools: often a sad choice (http://trends.newsforge.com/article.pl?sid=05/01/05/1842204))

[planet1]

prefer OGM over MP4 - now whats wrong with that ??? :rolleyes:

[Liisachan]

I just said patent-free things are preferable "as long as they can do the same job" that patented software can. Nothing more, nothing less.

In reality, I don't think (for instance) Ogg Theora can do what XviD can atm, but history may repeat itself. Think about MP3 vs Vorbis, or GIF vs PNG. Sometimes free software is, or will be, better in quality too, not just being open-sourced and patent-free.

[bond]

Originally posted by Liisachan
In reality, I don't think (for instance) Ogg Theora can do what XviD can atm, but history may repeat itself. Think about MP3 vs Vorbis, or GIF vs PNG. Sometimes free software is, or will be, better in quality too, not just being open-sourced and patent-free. [/B] you are mixing up formats with implementations of formats
i am sure you can write a vorbis encoder that gives worse quality than a mp3 encoder or a theora encoder that is better than a mpeg-4 asp encoder
apart from that theora has lots of features where its superior to mpeg-4 asp (which for example xvid is following)
apart from that theora uses patented technologies. the important point imho is not whether a technology is patented but whether its license free or not