ffdshow tryouts project: Discussion & Development

[CiNcH]

Ok, problem solved with 1021.

Now I have the problem, that with anamorphic material, picture seems zoomed and with wrong aspect ratio (flat heads...), but again, only with Overlay Mixer.

[Inventive Software]

With the amount of bug fixes from beta1, would the devs like to put forward a beta2 build?

[sillKotscha]

With the amount of bug fixes from beta1, would the devs like to put forward a beta2 build?

already answered a few posts above ;)

I'm messed up with subtitle and some other bugs, but as soon as possible :D .
Recent changes have to be tested a little more.

[Atak_Snajpera]

FFDshow still uses only one core for decoding H264. 1080p movies with bitrate at 15mbit are killing my C2D@2.8Ghz

[Leak]

FFDshow still uses only one core for decoding H264. 1080p movies with bitrate at 15mbit are killing my C2D@2.8Ghz
Yeah, well - libavcodec (http://en.wikipedia.org/wiki/Libavcodec) still uses only one core for it. That's what ffdshow uses - you'd need to bring that up with the libavcodec developers (aka the mplayer team)...

np: Swod - Walz 57 (Gehen)

[Fizick]

Leak,
Extented Avisynth support is very interesting!
But I can not install any of your builds :(
Setup stop at 90 percent of line with message window:
Internal error: Expression error 'Runtime Error (at 93:412): RegSvr32 failed with exit code 0x3'.

I try use filemon - regmon to catch. may be it is with last lines:
regsvr32 C:\Program Files\ffdshow\ffdshow.ax.2.manifest FILE NOT FOUND
regsvr32 C:\Program Files\ffdshow\ffdshow.ax.2.config FILE NOT FOUND

AthlonXP, WinXP Home Russian SP2, MS VC2005 express, etc

generic clsid and xxl builds are installed OK. But seems they do not use VC80.
(do not install msvcr80.dll)

[WorBry]

Having archived more than 500GB of deinterlaced/denoised home video in FFDShow-HuffYuv (YV12) format, I've been looking of late at alternative lossless codecs with a higher compression ratio. FFV1 certainly fits the bill. However, whilst the encodes open OK in VDub, if I try and play-back in other media players (TCMP, MPC etc) they play for a few seconds and then disintegrate into a mosaic of coloured blocks before either freezing or repeating the same pattern. I'm currently using the latest clsid build (Rev 1008, 20070308) with the following FFV1 settings - AC, Large Content, KF interval 10 (default) - but the same playback problem results no matter what settings I use. Searching the forum I see that this is/was a known problem, at least with older (2003) builds:

http://forum.doom9.org/showthread.php?p=400588#post400588

Could anyone point me to a recent (FFV1 stable) FFDShow build that does not suffer this problem or advise a remedy (other than use another LC).

Many thanks.



Could you upload a sample file that doesn't play correctly?

I re-encoded a small clip with default settings and it played back ok.

Sure, but as I mentioned the 'problem' appears a few secs (4-5) into playback, so the samples are necessarily around 40-50MB

Here is the HuffYuv-YV12 source:

http://rapidshare.com/files/20713071/Sample_HuffYuv-YV12_source.avi

and here is the FFV1 (AC, Large Context, KF=10) encode:

http://rapidshare.com/files/20716143/Sample_FFV1_AC_Large_Context_KF10.avi

[clsid]

I can confirm the FFV1 bug.

[foxyshadis]

FFV1 bug was filed (http://sourceforge.net/tracker/index.php?func=detail&aid=1679220&group_id=173941&atid=867360) btw. So far it only happens with AC encodes.

[Leak]

generic clsid and xxl builds are installed OK. But seems they do not use VC80.
(do not install msvcr80.dll)
I'm afraid VC80 is all I have, and I'm using the same InnoSetup script everyone else seems to be using. I'll try installing it at work tomorrow and see what happens.

But I'm always open for hints on how to fix this...

(@Haruhiko: didn't you have a similar problem recently?)

[Fizick]

where is the Innosetup script?
I do not see it at
http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/

[Leak]

where is the Innosetup script?
I do not see it at
http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/
Ah yes, that one's a bit non-obvious - it's here. (http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/trunk/bin/distrib/InnoSetup/)

[sillKotscha]

I have noticed that libmad won't decode the sound of this clip (http://www.wikiupload.com/download_page.php?id=101744) (down right you'll find the download link)... mp3lib does work...

as a side note - even though this is a normal *.mpg file Haali's splitter is used as mpg demultiplexer - why??

I have other *.mpg files and as usual my Elecard MPEG Demultiplexer is used.

EDIT:

ffdshow is still (after your various fantastic bug hunting work) unable to decode the sound of that clip (http://www.wikiupload.com/download_page.php?id=101790)...

if I select the Quicktime components to decode the file everything is ok but if I select the DShow framework in MPC to decode the clip the sound is totally fu**ed up...

here you see the sound properties of the file:

http://img389.imageshack.us/img389/2452/001ta4.jpg (http://imageshack.us)

I use ffdshow_rev1021_20070312_xxl.exe and MPC's internal MP4/MOV splitter - with Haali's splitter I hear no sound at all and with Quicktime's movie parser I can hear the sound as it should

[haruhiko_yamagata]

I'm afraid VC80 is all I have, and I'm using the same InnoSetup script everyone else seems to be using. I'll try installing it at work tomorrow and see what happens.

But I'm always open for hints on how to fix this...

(@Haruhiko: didn't you have a similar problem recently?)You use MSVC8 SP1, don't you? In that case, you'll have to replace \trunk\bin\distrib\msvcr80.dll, \trunk\bin\distrib\InnoSetup\msvc80\ffdshow.ax.manifest and maybe microsoft.vc80.crt.manifest.

[Px]

FFDshow still uses only one core for decoding H264. 1080p movies with bitrate at 15mbit are killing my C2D@2.8Ghz
It's strange, because on same CPU I have problems only with streams with average bitrate 24 Mbps....
One another thing - ffdshow can use both cores, for that I use "tweak" with process affinity, I look for core, on which video player running, and set affinity to another core, then switch back, and finally, allows player to use both cores.

[cc979]

Ok, problem solved with 1021.

Now I have the problem, that with anamorphic material, picture seems zoomed and with wrong aspect ratio (flat heads...), but again, only with Overlay Mixer.

overlay anamorphic .avi bug is old, vmr9 or haali will play them correctly

[_xxl]

FFDshow still uses only one core for decoding H264. 1080p movies with bitrate at 15mbit are killing my C2D@2.8Ghz
True.ffmpeg decoder can't handle H264 1080p.
It's strange, because on same CPU I have problems only with streams with average bitrate 24 Mbps....
One another thing - ffdshow can use both cores, for that I use "tweak" with process affinity, I look for core, on which video player running, and set affinity to another core, then switch back, and finally, allows player to use both cores.
You are wrong.Both cores but no multithreaded.No speed improvement.
http://forum.doom9.org/showthread.php?p=922249#post922249
http://forum.doom9.org/showthread.php?p=923944#post923944
http://forum.doom9.org/showthread.php?p=923950#post923950

[_xxl]

Haali's splitter I hear no sound at all and with Quicktime's movie parser I can hear the sound as it should
True.Haali can't handle them.
I have noticed that libmad won't decode the sound of this clip (down right you'll find the download link)... mp3lib does work...

Please provide a decent link.

[sillKotscha]

Please provide a decent link.

all there - see my post above...

just open the provided link and open your eyes for "click here for download" and as I said you will find that invitation down right ;)

EDIT:

True.Haali can't handle them.

ok, but even MPC's internal MP4/MOV splitter is unable to connect to ffdshow to let ffdshow decode the audio of that clip... thus isn't ffdshow the culprit as it should (at least it would be nice) handle that sort of audio...

http://img389.imageshack.us/img389/2452/001ta4.jpg (http://imageshack.us)

[Leak]

You use MSVC8 SP1, don't you? In that case, you'll have to replace \trunk\bin\distrib\msvcr80.dll, \trunk\bin\distrib\InnoSetup\msvc80\ffdshow.ax.manifest and maybe microsoft.vc80.crt.manifest.
Ugh... now that you mention it... >_<

Okay, I'll fix that when I get home...

But, ummm... is there anyone here not using SP1 for VC80?

EDIT: Come to think of it, there's something not right with the manifests in SVN anyway - the msvcr80.dll at trunk/bin/distrib/ (http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/trunk/bin/distrib/) that gets included in the installer is version 8.00.50727.42, but the manifest (http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/trunk/bin/distrib/microsoft.vc80.crt.manifest?revision=677&view=markup) for it specifies version 8.0.50608.0...

I guess I'll have to replace the runtime DLL in the distrib folder with the one I'm using and update the manifest, right?

[igor1st]

I have noticed that libmad won't decode the sound of this clip (http://www.wikiupload.com/download_page.php?id=101744) (down right you'll find the download link)... mp3lib does work...
No problem here. libmad works.

[sillKotscha]

ok, sorry... someone PM'ed me that the webhosting service I have uploaded that clip to is out of service right now - so here is a second attempt providing you with the clip in question...

Steve Balmer won't talk to me (http://www.savefile.com/files/550371) if I select libmad to decode the audio of that clip...

EDIT:

No problem here. libmad works.

which stream splitter do you use?? ... In my case Haali's splitter gets used as stream splitter - I've un-registered Haali's splitter concerning MPEG-PS support and now MPEG-I stream splitter (windows native MPEG-I splitter) is used and now the clip does play using libmad...

so, I nailed it down :) must be Haali's MPEG implementation which does caused my trouble... sorry for bothering the ffdshow devs

/ Sill

[Peuj]

No problem here. libmad works.

Yes libmad works for me too.

[Peuj]

ffdshow is still (after your various fantastic bug hunting work) unable to decode the sound of that clip (http://www.wikiupload.com/download_page.php?id=101790)...

if I select the Quicktime components to decode the file everything is ok but if I select the DShow framework in MPC to decode the clip the sound is totally fu**ed up...

I confirm that one.

[LotharZ]

I'm having problems with subtitles, If I select yellow as a colors, appears some defects.

Running ffdshow 1023 + MPC latest

http://img133.imagevenue.com/loc61/th_78568_snapshot20070313103415_122_61lo.jpg (http://img133.imagevenue.com/img.php?image=78568_snapshot20070313103415_122_61lo.jpg) http://img174.imagevenue.com/loc196/th_78885_Clip_122_196lo.jpg (http://img174.imagevenue.com/img.php?image=78885_Clip_122_196lo.jpg)

[haruhiko_yamagata]

I'm having problems with subtitles, If I select yellow as a colors, appears some defects.

Running ffdshow 1023 + MPC latest

Yes, I have been trying to fix this issue. I have many other issues related to subtitles.
But now, pehaps I should prioritize some other bugs that should be fixed before beta2. Please wait.

[CiNcH]

overlay anamorphic .avi bug is old, vmr9 or haali will play them correctly

With tryouts beta 1 and Overlay, anamorphic source material is displayed correctly.

I am not replaying from an AVI container, but from DVB stream.

[Px]

You are wrong.Both cores but no multithreaded.No speed improvement.
When I last time tested it (
http://forum.doom9.org/showthread.php?p=920783#post920783) I've got speed improvement ;)

[_xxl]

When I last time tested it (
http://forum.doom9.org/showthread.php?p=920783#post920783) I've got speed improvement ;)
Use timecodec.
No speed improvement for ffmpeg's h264 decoder on AMD X2.

[haruhiko_yamagata]

With tryouts beta 1 and Overlay, anamorphic source material is displayed correctly.

I am not replaying from an AVI container, but from DVB stream.
I don't have DVB. Is there any way to reproduce your problem?

[sillKotscha]

I don't have DVB.

here you go (http://www.savefile.com/files/550664)... DVB-S stream captured as PVA

EDIT: PVA DS Filter...

- RadLight PVA DirectShow filter v1.0.0.0 (http://www.free-codecs.com/download/RadLight_PVA_DirectShow_filter.htm)

- or use MPC's internal PVA filter

[Leak]

I don't have DVB. Is there any way to reproduce your problem?
Errrr... I had the same problem with the DVD I used for testing when it tried to switch from 4:3 to 16:9 (and failed), so I guess it's the same problem - but it only happens when using the Overlay Mixer.

The image gets stretched horizontally and is cut off at the right side.

[clsid]

I vaguely remember Overlay Mixer not supporting on-the-fly size changes.


@sillKotscha, submit your file to Haali, so that he can fix his splitter. He only just recently added MPEG PS support.

[Leak]

I vaguely remember Overlay Mixer not supporting on-the-fly size changes.
Well, the tooltip on the "Allow output format changes during playback > Connect to compatible filters only" checkbox lists Overlay Mixer as the very first compatible filter... :confused:

[ring0]

The following encoders only work with 25fps input: MPEG4, MPEG 1 and MPEG 2.

They seem to work with any frame rate which can be represented as 25*2^n, where n is an integer (even negative). So, it works for 6.25, 12.5, 25, 50, 100 etc.

[haruhiko_yamagata]

Sure, but as I mentioned the 'problem' appears a few secs (4-5) into playback, so the samples are necessarily around 40-50MB

Here is the HuffYuv-YV12 source:

http://rapidshare.com/files/20713071/Sample_HuffYuv-YV12_source.avi

and here is the FFV1 (AC, Large Context, KF=10) encode:

http://rapidshare.com/files/20716143/Sample_FFV1_AC_Large_Context_KF10.avi
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.

[haruhiko_yamagata]

With tryouts beta 1 and Overlay, anamorphic source material is displayed correctly.

I am not replaying from an AVI container, but from DVB stream.
Confirmed, on testing DVD playback.
Thank you, sillKotscha and Leak.

Well, the tooltip on the "Allow output format changes during playback > Connect to compatible filters only" checkbox lists Overlay Mixer as the very first compatible filter...
I'm wondering if I should remove overlaymixer from that list.

[haruhiko_yamagata]

Ugh... now that you mention it... >_<

Okay, I'll fix that when I get home...

But, ummm... is there anyone here not using SP1 for VC80?

I dislike the complicated run time problem, so I don't use SP1.
If you want to distribute ffdshow using SP1, you'll need a proper testing environment that msvcr80 (or .NET frame work) is not installed as shared assembly. I installed WindowsXp on a new partition for that purpose. Windows 2000 and Xp differs completely in that regard.

[clsid]

Can someone give a clear and precise description of the Overlay bug? Then I can update the opening post.

Haruhiko, can you list some of the bugs you know of that are in the subtitle code? Then I can add those to the list as well.

@all, are there any known bugs that are not yet in my opening post?

[CiNcH]

The image gets stretched horizontally and is cut off at the right side.

Sounds pretty much like the same problem.


I vaguely remember Overlay Mixer not supporting on-the-fly size changes.

It is working with tryouts beta 1 and Overlay Mixer.

But the problem with beta 1 is (as I posted earlier) that on-the-fly resolution changes do not work properly with Overlay Mixer now. I solved this problem by performing a graph rebuild at resolution change.
Another way to solve this problem is to use a later daily build. There you do not need to send down the information any longer but again have the problem with wrong anamorphic displaying.

[CiNcH]

Can someone give a clear and precise description of the Overlay bug?


Anamorphic material is just not displayed correctly when using the Overlay Mixer.
Like Leak said "The image gets stretched horizontally and is cut off at the right side."
I am just using ffdshow as a raw video post processor, so no decoder is involved and the problem also occurs when no filter (deinterlacing, sharpening or whatever) or anything else is used.

[haruhiko_yamagata]

It is working with tryouts beta 1 and Overlay Mixer.

But the problem with beta 1 is (as I posted earlier) that on-the-fly resolution changes do not work properly with Overlay Mixer now. I solved this problem by sending the connection format down the filter graph.
Another way to solve this problem is to use a later daily build. There you do not need to send down the information any longer but again have the problem with wrong anamorphic displaying.
On-the-fly resolution changes do not work with overlay mixer. It's the spec of the overlaymixer. Are there any ffdshow that can do that?
On-the-fly AR change should work with overlay mixer, while current ffdshow does not work correctly.

[CiNcH]

On-the-fly resolution changes do not work with overlay mixer. It's the spec of the overlaymixer. Are there any ffdshow that can do that?

As I said, it works with latest daily build if I am not completely mistaken. There does not seem to be a graph rebuilt, so it is really on-the-fly.

[haruhiko_yamagata]

Haruhiko, can you list some of the bugs you know of that are in the subtitle code? Then I can add those to the list as well.

As for srt subtitle and OSD, the color problem should be fixed.

Font->Charset->Default should use the native language of the country (version of the Windows language). For Japanese version of Windows, it should use shift-Jis, but it uses ANSI. This issue is true with UNICODE build as long as the subtitle file is ANSI (or shift-Jis).

As for SSA/ASS/ASS2, it's in the very early stage of the development, not implemented rather than very buggy.

[Delerue]

Anamorphic material is just not displayed correctly when using the Overlay Mixer.
Like Leak said "The image gets stretched horizontally and is cut off at the right side."
I am just using ffdshow as a raw video post processor, so no decoder is involved and the problem also occurs when no filter (deinterlacing, sharpening or whatever) or anything else is used.

Yeah. It seems to be a MPC (or maybe non-WMP) bug, not a FFDShow bug, because even using Microsoft codecs, some WMV anamorphic videos (like Step Into Liquid (http://download.microsoft.com/download/e/a/d/eadb9b42-728b-42b0-bfdf-b472fa2a2464/Step_into_Liquid_1080.exe)) can't be resized correctly. But if you open it with WMP, everything is fine even with overlay.

[chros]

I'm using ffdshow to convert dts to aac through avisynth:
graph: mkv (Haali's splitter) -> ffdshow (on output only 16 bit allowed)
script: DirectShowSource("audio.grf",video=false,Audio=true)

When I try to encode this in megui or behappy it gives me an error (see below). Strange, that I can playback the avs script in MPC ...
If I switch ffdshow to AC3filter in the graph, the encoding works well ...

The megui log:
Log for job job32

Error:
MeGUI.AviSynthException: DirectShowSource: Couldn't open GRF file.audio.grf:
This objectsurface is not supported
(audio.avs, line 5)
at MeGUI.AviSynthClip..ctor(String func, String arg, AviSynthColorspace forceColorspace, AviSynthScriptEnvironment env)
at MeGUI.AviSynthAudioEncoder.encode()
----------------------------------------------------------------------------------------------------------
The current job contains errors. Skipping chained jobs

[_xxl]

As for SSA/ASS/ASS2, it's in the very early stage of the development, not implemented rather than very buggy.

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

[haruhiko_yamagata]

As I said, it works with latest daily build if I am not completely mistaken. There does not seem to be a graph rebuilt, so it is really on-the-fly.
It also works with beta 1, but only when sending the connection format downstream.
Only when new resolusion is smaller than before.
Milan's 20051129 does not try to resize if new resolusion is bigger.

but only when sending the connection format downstream.What is this?

[Leak]

I dislike the complicated run time problem, so I don't use SP1.
If you want to distribute ffdshow using SP1, you'll need a proper testing environment that msvcr80 (or .NET frame work) is not installed as shared assembly. I installed WindowsXp on a new partition for that purpose. Windows 2000 and Xp differs completely in that regard.
Yeah, I was thinking of installing XP in a VirtualBox instance for testing...

Oh well, the good thing about VM images is that they're highly reuseable... :)

[CiNcH]

What is this?

Actually a mistake. Because what solved the problem was automatic graph rebuilding at resolution change... sorry.
Sending connection format downstream solved another problem, not ffdshow related...


Yeah. It seems to be a MPC (or maybe non-WMP) bug not a FFDShow bug

Using DVBViewer. Fact is, with beta 1 it works, with latest daily build it doesn't ;) .