ffdshow tryouts project: Discussion & Development

[Fizick]

Here is ffdshow_rev1022_avisynth2_20070312_igor1st.exe
with Leak Avisynth contribution (compilied VC7 ?):

http://www.mytempdir.com/1252011

It works well! I try DeGrainMedian filter (it use 3 frames).
Bug: first 5 frames are the same as very first (i.e. duplicates).

What is 10,10,10 buffers in settings?

[igor1st]

@all, are there any known bugs that are not yet in my opening post?
OSD item "Encoder info" doesn't work anymore.

[Leak]

It works well! I try DeGrainMedian filter (it use 3 frames).
Bug: first 5 frames are the same as very first (i.e. duplicates).
That's strange - with default settings the plugin should buffer the first 11 frames before even requesting the frame 0 from AviSynth, so AviSynth and DeGrainMedian should be able to request frames 0-10 right away...

Whenever AviSynth requests a frame, the nearest one that is buffered is used.

Wait a second - the very first frame is produced by buffering a single input frame followed by requesting frame 10-million-and-a-bit to avoid filters re-using anything they cached while processing that first frame. Maybe DeGrainMedian doesn't like this? (i.e. I'm requesting frames 10810800, 1, 2, 3, etc.)

This was the easiest solution for the problem of having to pass the first frame through so the filter graph starts to run. If I passed the first frame through as-is I'd have to resize it, change the colourspace and whatnot myself... sure, I could blow away AviSynth and re-initialize it after fetching the first frame, but it's already bad enough doing it once after every seek...

What is 10,10,10 buffers in settings?
Those settings specify how many buffers should be kept - while ffdshow is running, there's always a "current" frame. Buffering happens by taking the frames that come into the filter until there's enough frames buffered, and only then does anything get passed further up the filter chain and the DirectShow graph.

The first value specifies how many buffers should be kept before the "current" frame.

The second value specifies a limit for how many frames should be buffered for producing the "current" frame - i.e. SelectEven halves the number of frames, so the filter will buffer 2 input frames to produce one output frame; this is calculated from the number of frames the resulting AviSynth clip has but I needed some upper limit so doing a "Trim(0,1)" won't allocate buffers for the whole 10 million frames or so the fake input clip is long.

The third value specifies the number of frames to keep buffering ahead.

So with settings of x/y/z AviSynth filters can request frames from n-x to n+z when working on frame n, while the total number of frames AviSynth returns can be up to 1/y of the number of frames ffdshow hands over to AviSynth.

This all assumes that adding or removing frames happens linearly, of course.

In other news, I've managed to reproduce the installation problem (d'oh!) and fixed it. I'll post an updated build in a bit.

np: One Second Bridge - Alt. Ending (One Second Bridge)

[Leak]

Okay, here's a new build based on rev. 1028 (http://leak.no-ip.org/AviSynth/ffdshow_AviSynth_rev1028_20070313.exe) that should work even if the Visual C 8.0 redistributables aren't installed on your PC.

Could someone who had problems installing my earlier builds give it a test please? (I know it worked on a fresh Windows XP SP2 install, but I'd like to be sure...)

np: Console - Tunnelvision (Pan Or Ama)

[Liisachan]

Implementing this is important because vsfilter is no longer updated and it has some bugs.Maybe after beta2 if time is available.

Maybe mplayer's code or libASS (http://sourceforge.net/projects/libass/) is reusable for this?

[WorBry]

Thank you for the sample.
I found that the bug begins at rev 2411 of the original ffdshow.
Please uncheck "Quality control"->"Drop frame on delay" for the time being. Of course, more radical fix is needed.

Yes, the 'temporary remedy' appears to work. So, if I understand the situation correctly, the problem is entirely related to the FFDShow decoder per se and has nothing to do with the FFV1 encoding part. In other words, am I safe in committing my HuffYuv-YV12 archives to FFV1 (AC, Large Context KF10) without the possibility of changes to the FFV1 coder that might result in decoding incompatibilities with future FFDShow builds in which this bug has been fixed?

[haruhiko_yamagata]

Yes, the 'temporary remedy' appears to work. So, if I understand the situation correctly, the problem is entirely related to the FFDShow decoder per se and has nothing to do with the FFV1 encoding part. In other words, am I safe in committing my HuffYuv-YV12 archives to FFV1 (AC, Large Context KF10) without the possibility of changes to the FFV1 coder that might result in decoding incompatibilities with future FFDShow builds in which this bug has been fixed?
Probably no, because your computer (and most computers) is not fast enough to decode FFV1. FFV1 is not as fast as HuffYuv.

The problem has been fixed at rev 1030.

[haruhiko_yamagata]

Maybe mplayer's code or libASS (http://sourceforge.net/projects/libass/) is reusable for this?
Thank you, I did not know libass. I'll look at it when I have time.
Can mplayer render SSA? ffdshow uses code from mplayer, which is not fully implemented.

[Milvus]

libASS is part of the mplayer project, so yes, mplayer can render SSA/ASS. Not as complete as VSFilter, but still quite nice.

Another link you should find very interesting : The ASA wiki (http://asa.diac24.net/) : Lot of informations about subtitling libraries, like this Compatibility Chart (http://asa.diac24.net/Compatibility).

[haruhiko_yamagata]

libASS is part of the mplayer project, so yes, mplayer can render SSA/ASS. Not as complete as VSFilter, but still quite nice.

Another link you should find very interesting : The ASA wiki (http://asa.diac24.net/) : Lot of informations about subtitling libraries, like this Compatibility Chart (http://asa.diac24.net/Compatibility).
Thank you for the links. I'll read later.

Do you really think mplayer is better than current ffdshow in handling SSA? I tested MPUI 20070310 for embedded SSA in mkv (mewmew-vorbis-ssa.mkv (http://www.matroska.org/samples/mewmew/index.html)), but I think ffdshow is better. Yes, I know it's meaningless to argue this, both of them are quite incomplete. Or is it the issue with Windows port or my installation?

[Episode]

On my OSX build of mplayer I need to use -ass -embeddedfonts -fontconfig flags to get subtitles rendered properly. Perhaps similar flags will work on Windows-version as well.

[haruhiko_yamagata]

On my OSX build of mplayer I need to use -ass -embeddedfonts -fontconfig flags to get subtitles rendered properly. Perhaps similar flags will work on Windows-version as well.
Adding these config, mplayer crashes. Sorry for OT, but how to make it play on Windows Xp?

[Episode]

@haruhiko_yamagata, how about trying it without the -fontconfig flag? I just tried mplayer 1.0rc1 on my Windows machine and "mplayer -ass -embeddedfonts filename" seems to work quite well.

[Eragon4ever]

On Windows MPlayer displays ASS with "-sid 0 -ass" however it crashes if the subtitle track is SSA or SRT.

[haruhiko_yamagata]

Thank you. ASS plays.
Code have already been imported from mplayer, though it is not updated. Main problem seems to be the porting.

[Peuj]

clsid2: Removed WavPack support. It was far to incomplete to be usable. Use CoreWavPack DS filter instead.

just for information: the latest CoreWavPack filter can be find here http://coreforge.org/frs/?group_id=28

[JnZ]

Could you please repair "proper" multi-channel decoding for OGG Vorbis files? (Due to Vorbis I speciffication). Official vorbis decoder decodes proper channels, but FFDShow decodes bat channel order.
Quote: Vorbis I uses a single mapping type (0), with implicit PCM channel mappings.

EDIT: OR FFDshow use right setting?

[_xxl]

Removed WavPack support. It was far to incomplete to be usable. Use CoreWavPack DS filter instead.
Libavcodec wavpack was removed.

[Fizick]

Leak,
your new build installed well:
1. Windows2000 without msvc8
2. WinXP with MS VC2005 express (NET, etc)

But bug still exist:
Number of extra very first frames= (buffer ahead)+1

Try script:

showframenumber()

and load decoded ffdshow video in virtualDub.

first frames returns number 1071079 (not 0).

[Px]

Use timecodec.
I prefer to watch video in video players ;), and I saw speed improvement (on 45 Mbps frames), but not too much, yes....

[_xxl]

but not too much, yes....
Nothing because libavcodec h264 decoder is not multithreaded.

[foxyshadis]

I prefer to watch video in video players ;), and I saw speed improvement (on 45 Mbps frames), but not too much, yes....

Any improvement you saw probably came from queuing keeping short cpu peaks from killing smoothness. Queuing can't thread the decoder, but it does keep the decoder working while waiting for the last frame to be displayed, which it won't without queuing.

[Leak]

But bug still exist:
Number of extra very first frames= (buffer ahead)+1

Try script:

showframenumber()

and load decoded ffdshow video in virtualDub.

first frames returns number 1071079 (not 0).
Please see my answer to this here (http://sourceforge.net/tracker/index.php?func=detail&aid=1676882&group_id=173941&atid=867362)...

[Px]

Any improvement you saw probably came from queuing keeping short cpu peaks from killing smoothness. Queuing can't thread the decoder, but it does keep the decoder working while waiting for the last frame to be displayed, which it won't without queuing.
I agree with you :)

[interfan]

Hi,

I have a problem to which I can't find a solution. I'm lost :(

After getting a new PC, I've installed XP SP2 and the latest ffdshow from www.free-codecs.com.
I then have noticed a problem, so I've reinstalled it, again, after d/l the latest version from the same source.
My current(which is 2nd) version is the FFDShow 2007-02-09 (rev. 899) SSE version.

But the problem persists.

In some sites the embedded video plays in strange colors(green, purple and other such colors). What bothers me most is the P2P based streaming videos like PPSTREAM.

I've updated my ATI driver to the latest as well- it didn't helped either.

I even don't know whether it's ffdshow-related problem, as I've not tried to play streaming videos before installing FFDSHOW. And, furthmore, except FFDSHOW, the only thing I install is real and quick-time alternatives.


I don't want to install and reinstall too much, especially as long as I'm enjoying a clean and fast system(you how things go in the beginning) and I like FFDSHOW very much, especially as it has given to my old P3 some another 2 years of usability :)


I'll appreciate help.

[clsid]

Those embedded videos are likely Windows Media files. By default ffdshow doesn't decode such files. Your problem can be solved by disabling DXVA for the Windows Media codecs. You should be able to find an option to disable it somewhere in your graphics card control panel.

[_xxl]

In some sites the embedded video plays in strange colors(green, purple and other such colors). What bothers me most is the P2P based streaming videos like PPSTREAM.

Is ffdshow decoding this?

[zambelli]

Can somebody else confirm these bugs:


AVIS decoding is not working at all in DirectShow mode - the filter fails to connect. In VfW mode it just decodes black.
When YV12 decoding is enabled only in VfW mode and YV12 video is played through DirectShow, the image gets mirrored
When YV12 decoding is enabled in both DShow and VfW, YV12 video is always converted to YUY2 when rendered to VMR (i.e. GraphEdit playback).

I've verified them with both beta1 and a clsid build from 3-14-2007. I'm fairly certain about the 1st one, but the 2nd & 3rd might be video hardware related.

[Episode]

I can confirm these bugs.

[Px]

In some sites the embedded video plays in strange colors(green, purple and other such colors). What bothers me most is the P2P based streaming videos like PPSTREAM.
That is not ffdshow problem, update your Windows Media Player at least to 10, better to 11 at http://windowsupdate.microsoft.com,
section "Optional"

[interfan]

That is not ffdshow problem, update your Windows Media Player at least to 10, better to 11 at http://windowsupdate.microsoft.com,
section "Optional"

I hate Microsofts' media players- they are so heavy...

But I'll try your suggestion, as I have some feelings that might help. I didn't understand clsids' suggestion anyway, because I don't even know what DXVA is.

I'll report whether it fixes the problem.

[clsid]

Installing just "Windows Media Format Runtime 11" instead of the whole Windows Media Player might be enough.

DXVA is related to hardware acceleration. Google it.

[LigH]

DXVA = DirectX Video Accelleration: The graphics chipset should support a little part of the MPEG decoding algorithm... Should. Not all models of a graphic card family provide it. Sometimes applications (like software DVD players) fail to use it even if the chip provides it. On top of it, Microsoft waited with the full support for Vista.

http://en.wikipedia.org/wiki/DXVA

[interfan]

Installing just "Windows Media Format Runtime 11" instead of the whole Windows Media Player might be enough.

How can I install it? it's a separated software to install or it's installed as a component in the WMP11 installation?

----

Never mind- I see these are just free codecs. I was hoping to avoid installing too much codecs, but don't really have an option...

[Px]

Installing just "Windows Media Format Runtime 11" instead of the whole Windows Media Player might be enough.
Sometimes, there are strange problems with standalone runtime, which doesn't appear when full pack with player installed, so I prefer to recommend player update, nor runtime. Also, I must mention, that in some cases you need to additionally install WMV9 VCM (last time - game "You are empty")

[Px]

How can I install it? it's a separated software to install or it's installed as a component in the WMP11 installation?
As a component
Never mind- I see these are just free codecs. I was hoping to avoid installing too much codecs, but don't really have an option...
You have these codecs already installed, but they outdated, and because of that, produces color artifacts on video...

[interfan]

It's working!!! :)

Those Windows Media Format Runtime 11 really helped and it's definitely the better way(who the hell need WMP?!) for me.

I'm thankful as I've gone through 3 forums before this one and nobody could help me there.

One last question- I like controlling my system and I dislike having garbages which I can't erase purely. Will I be able to uninstall "Windows Media Format Runtime 11"? If yes- how?

[zambelli]

I can confirm these bugs.
Thanks!

Can anybody from the ffdshow team look into this? I can file an official bug through Sourceforge too if necessary.

[clsid]

I'll add them to the buglist. See opening post. Thanks for submitting them.

[foxyshadis]

MPC build discussion moved to new thread (http://forum.doom9.org/showthread.php?t=123537).
DVD navigator bug moved to new thread (http://forum.doom9.org/showthread.php?t=123538).

[Blinx123]

Has anybody here used ffdshow with the Lord of the Rings DVDs? How does it look in the fact of detail and beauty?

Could anybody post some Comparision Pictures between the normal and the scaled/post processed version?

[foxyshadis]

That might be a more appropriate question for Jeremy's thread (realtime postprocessing with ffdshow and avisynth (http://forum.doom9.org/showthread.php?t=115727)) than here.

[skystrife]

Again, this has probably been asked several times before, but are there any plans to compile a 64-bit version of ffdshow-tryouts? The only 64-bit build out there right now is the celtic_druid one, and IIRC it's fairly old.

[_xxl]

Now ffdshow can't be compiled for 64-bit.Also libavcodec.This is not a priority.

[Dr Pizza]

Again, this has probably been asked several times before, but are there any plans to compile a 64-bit version of ffdshow-tryouts? The only 64-bit build out there right now is the celtic_druid one, and IIRC it's fairly old.

I wish I could at least find the source to the celtic druid 64-bit build. It's apparently a build of 2546 (which is the most recent ffdshow source on sf.net), but it must have been modified, because 2546 won't build as an x64 binary (indeed, there aren't even project settings for it). If I could get that working then patching with updates from ffdshow-tryouts might be possible, to produce an up-to-date 64-bit build. But unfortunately, I cant.

The big stumbling block is the assembler portions. C and C++ I can do. At a pinch, SSE intrinsics are possible. Assembler, however, I can't do a damn thing with. If yasm won't build it, I can't fix it.

I wish x64 were regarded as more important than it currently is.

[Episode]

Is there really a need for x64 build of ffdshow? I mean, it won't be any faster than normal 32-bit build and x64 is compatible with normal builds.

[haruhiko_yamagata]

AVIS decoding is not working at all in DirectShow mode - the filter fails to connect. In VfW mode it just decodes black.


AVIS decoding in DirectShow mode works for me.
In vfw mode, ffdshow fails to connect.:confused:

[Dr Pizza]

Is there really a need for x64 build of ffdshow? I mean, it won't be any faster than normal 32-bit build and x64 is compatible with normal builds.
64-bit software can't use 32-bit codecs (so if you're using 64-bit WMP, for example, you need 64-bit codecs).
And it probably should be a bit faster, too, due to the extra architectural registers (as long as the compiler does a decent job, at least).

[lilhobo]

why cant i using ffdshow for RV10 in sony vegas???

or do i need to remove realplayer first??

[skystrife]

Is there really a need for x64 build of ffdshow? I mean, it won't be any faster than normal 32-bit build and x64 is compatible with normal builds.

There are several x64 players out there that won't use 32-bit codecs. If you take a x64 build of Vista, your Media Center, Movie Maker, and DVD Maker are all 64-bit, so the current 32-bit ffdshow won't work with those applications.

I'm just really hoping that this will happen, because I love ffdshow for it's immense flexibility, and I'd like to be able to use an updated 64-bit version of it (celtic_druid's build has some issues with H.264 decoding- artifacts galore).