ffdshow tryouts project: Discussion & Development

[haruhiko_yamagata]

haruhiko_yamagata:
I have a very weird ffdshow subtitle renderer behavior.
I've written an automatic subtitle stream-selection code based on Language ID, Name, etc.
When I'm enabling a stream matching the user params right after the graph is built, depending on what appears to be a race condition, it seems ffdshow is selecting a different (external file) subtitle stream after I specify the stream I want (one of several embedded streams) using the IAMStreamSelect interface.

If I enable 'give embedded priority over subtitle files', it seems to fix the issue, but I haven't tested it against enough content and other splitter filters.
Can you disable FFDShow from selecting a subtitle stream itself if the IAMStreamSelect interface was issued a call to enable a stream?

Another cosmetic glitch:
If you have both the FFDShow Video decoder and the FFDShow Subtitle filter in the same graph, the FFDShow video decoder's IAMStreamSelect interface contains duplicate streams (The vid decoder grabs the stream list from the sub-filter).
You need to blacklist other ffdshow filters when building the stream list.

I'm not familiar with that part, but I'll have to deal with it to complete my work related to subtitles. Please wait.

[haruhiko_yamagata]

I'm not in a hurry, take all the time you need... :)
It is hopefully fixed at rev 4145. Please try the latest version.

[haruhiko_yamagata]

Haruhiko, regarding the bug I posted earlier,

If you create a preset under ffdshow raw that does not have the same name as a preset under normal ffdshow then there is no default selected (nothing in bold) This is as I would expect.

If you click on create a new profile from selected, ffdshow crashes with an error in rundll32 and exits instead of throwing up an error that no profile is selected.

Also if you click on codecs while no preset is selected the rest of the menu items disappear.
The crash has been fixed at rev 4146. The active preset should be saved separately for video/audio decoder and raw filters. The work is in progress.

[haruhiko_yamagata]

Any idea on that logo away issue? I can't mask out the very edges of a logo that are right up to the edges of the video.
I have never used that filter. How do you use the filter? Settings, etc.
Please send me the sample and the "Logo image", if you have one.

[madshi]

It is hopefully fixed at rev 4145. Please try the latest version.
There's something wrong, I don't get *any* repeatfirstfield flags, anymore, now. See "AM_VIDEO_FLAG_REPEAT_FIELD" here:

http://msdn.microsoft.com/en-us/library/windows/desktop/dd373499%28v=vs.85%29.aspx

[egur]

I have not looked into the issue yet. Did you try a new binary? New version is available from XvidVideo.RU (http://www.xvidvideo.ru/x264-video-codec).
Binary compatibility of libavcodec.dll is not guaranteed between versions.

Looks like the 4052 build had a bad ffmpeg.dll build (x64).
The 4147 build from XvidVideo.RU works great - no memory corruption :) --with resפect to my tests anyway.

[mkanet]

This affect's all ffdshow versions as far as I know. I'm currently using: ffdshow_rev4150_20111215_clsid

I am trying to playback mpeg2 videos which switch between soft and hard telecine using ffdshow's internal mpeg2 decoder + avisynth IVTC/deinterlace.

The problem is that the internal “ffdshow_source” source filter that is used to get frames from ffdshow into AviSynth is not honoring soft pulldowns when fed to the avisynth IVTC plugin (TIVTC or Decomb). It needs to behave the same way as if the video was fed from DGDecode (or any avisynth mpeg2 decoder would).

Could someone PLEASE enhance the existing "ffdshow_source" source filter to honor soft pulldowns in the same fashion when DGDecode does? I'm hopeful this will be fixed in the next beta release.

Normally, I wouldn't make a big deal about this. However, this type of mix soft/hard telecine video is extremely common in America; as, it's the encoding format used by many cableTV content providers like HBO, Showtime, Cinemax, etc; played back by millions of people (mostly on STB cableboxes).

PS: I have "Detect soft telecine and average frame durations" disabled in ffdshow in order for the IVTC plugin in Avisynth to remove 3:2 pulldown correctly to 23.976fps.

Below are some video clips to test against. Each have hard and soft telecine in them at 1080i.

Here they are:
http://www.megaupload.com/?d=8D8Q1Q5G
http://www.megaupload.com/?d=D8X708QK
http://www.megaupload.com/?d=P827WH5F

Thank you for all your hard work guys!
MKANET

[haruhiko_yamagata]

This affect's all ffdshow versions as far as I know. I'm currently using: ffdshow_rev4150_20111215_clsid

I am trying to playback mpeg2 videos which switch between soft and hard telecine using ffdshow's internal mpeg2 decoder + avisynth IVTC/deinterlace.

The problem is that the internal “ffdshow_source” source filter that is used to get frames from ffdshow into AviSynth is not honoring soft pulldowns when fed to the avisynth IVTC plugin (TIVTC or Decomb). It needs to behave the same way as if the video was fed from DGDecode (or any avisynth mpeg2 decoder would).

Could someone PLEASE enhance the existing "ffdshow_source" source filter to honor soft pulldowns in the same fashion when DGDecode does? I'm hopeful this will be fixed in the next beta release.

Normally, I wouldn't make a big deal about this. However, this type of mix soft/hard telecine video is extremely common in America; as, it's the encoding format used by many cableTV content providers like HBO, Showtime, Cinemax, etc; played back by millions of people (mostly on STB cableboxes).

PS: I have "Detect soft telecine and average frame durations" disabled in ffdshow in order for the IVTC plugin in Avisynth to remove 3:2 pulldown correctly to 23.976fps.

Below are some video clips to test against. Each have hard and soft telecine in them at 1080i.

Here they are:
http://www.megaupload.com/?d=8D8Q1Q5G
http://www.megaupload.com/?d=D8X708QK
http://www.megaupload.com/?d=P827WH5F

Thank you for all your hard work guys!
MKANET
Thanks for your report and samples. I'll take a look later.

[haruhiko_yamagata]

There's something wrong, I don't get *any* repeatfirstfield flags, anymore, now. See "AM_VIDEO_FLAG_REPEAT_FIELD" here:

http://msdn.microsoft.com/en-us/library/windows/desktop/dd373499%28v=vs.85%29.aspx
Sorry for that. I reverted rev 4145 at rev 4153.
By the way, is it MPEG-2, VC1 or H.264? Does libmpeg2 or libavcodec matter?

[madshi]

Sorry for that. I reverted rev 4145 at rev 4153.
By the way, is it MPEG-2, VC1 or H.264? Does libmpeg2 or libavcodec matter?
No worries. I had tested with MPEG2 libavcodec. But now I've retested and the same problems occur with libmpeg2 and h264 libavcodec, too. I tested with 4145, though.

Edit: It is most important for MPEG2, but it would be good to have working for h264, too. It's not so important for other codecs because it's mostly only broadcasts which are affected, and those are all either MPEG2 or h264. Ok, some Blu-Ray content may be affected, too, and that in theory could be VC-1, but I've not yet seen such content (affected VC-1 content, I mean).

[clsid]

Everyone please give the latest build some thorough testing. All known regressions have now been fixed I think.

Stuff that needs some extra attention during testing:
- DXVA
- 10-bit colorspace handling. Test in combination with madVR.
- The various processing filters with 10-bit input. Be aware that most are not compatible with those colorspaces, so the expected behavior for those filter is "no effect". Crashes and messed up video are bugs.
- There have been some major audio changes in Libav, so test all audio decoders.
- Intel QuickSync decoder, if you have an Intel Sandy Bridge processor.

Everyone who has recently reported a bug that hasn't been fixed yet, please add it to the bugtracker if you haven't done so already. That way it won't be forgotten.

[haruhiko_yamagata]

- The various processing filters with 10-bit input. Be aware that most are not compatible with those colorspaces, so the expected behavior for those filter is "no effect". Crashes and messed up video are bugs.
Currently only resize (10 to 10) and RGB conversion (10 to 8) supports native 10-bit processing. All other image filters converts 10-bit to 8-bit and apply the filter.
10-bit raw input is not supported. Of course it's useless unless we add more 10-bit native filter.
Yadif is a candidate for 10-bit native processing. If you think it's cool, please provide a 10-bit interlaced sample to motivate me.

[Reino]

- There have been some major audio changes in Libav, so test all audio decoders.Crash when trying to open:
- http://samples.mplayerhq.hu/A-codecs/amr/sample.amr (http://samples.mplayerhq.hu/A-codecs/amr/sample.amr)
- http://samples.mplayerhq.hu/A-codecs/lossless/luckynight.tta (http://samples.mplayerhq.hu/A-codecs/lossless/luckynight.tta)

Decoding failure:
- DVD-Audio-testfile (ARAW;24bit,96000Hz,4608Kbs).mka (http://www.mediafire.com/?km42s999u8rgfcc)
Demuxing the WavPack Audio from the Matroska container and trying to play that fails too.

I believe it's been reported before, but anyway:
- http://media.nero.com/Standard720x576_1MBit.mp4 (http://media.nero.com/Standard720x576_1MBit.mp4)
AAC Audio part decoding failure with libavcodec. No problem though with libfaad2.

[mkanet]

Thank you so much. I've waited a long time for someone to be able to have a realtime hard/soft pulldown IVTC solution in ffdshow.

I know this has been asked many times, but I will ask anyway... adding full IVTC support (hard and soft telecine) natively in ffdshow would be ideal.

However, I'm guessing it would be less hassle to get soft/hard IVTC switching to work correctly in ffdshow -->avisynth.

Thanks for your report and samples. I'll take a look later.

[clsid]

I believe it's been reported before, but anyway:
- http://media.nero.com/Standard720x576_1MBit.mp4 (http://media.nero.com/Standard720x576_1MBit.mp4)
AAC Audio part decoding failure with libavcodec. No problem though with libfaad2.Fixed.

[clsid]

Crash when trying to open:
- http://samples.mplayerhq.hu/A-codecs/lossless/luckynight.tta (http://samples.mplayerhq.hu/A-codecs/lossless/luckynight.tta)Fixed.

[clsid]

Decoding failure:
- DVD-Audio-testfile (ARAW;24bit,96000Hz,4608Kbs).mka (http://www.mediafire.com/?km42s999u8rgfcc)
Demuxing the WavPack Audio from the Matroska container and trying to play that fails too.Fixed.

[clsid]

Crash when trying to open:
- http://samples.mplayerhq.hu/A-codecs/amr/sample.amr (http://samples.mplayerhq.hu/A-codecs/amr/sample.amr)Fixed.

[TheShadowRunner]

Clsid, maybe you remember the FLV4/VP6 cropping issue.
Nev has found a solution (http://code.google.com/p/lavfilters/source/detail?r=859e3bcc3aa6) for LAV Video Dec, could it be implemented in FFDshow too?

[ryrynz]

The crash has been fixed at rev 4146. The active preset should be saved separately for video/audio decoder and raw filters. The work is in progress.

I tested with 4158 and the crash still occurs, also if I just highlight any preset and exit whilst it's highlighted (when a raw preset is set to default)
it crashes as well, same error in rundll32, thanks for looking into it though, I know it's not terribly important.
Did you happen to look into the problem with Avisynth scripts carrying over in the export registry file by chance?

[haruhiko_yamagata]

I tested with 4158 and the crash still occurs, also if I just highlight any preset and exit whilst it's highlighted (when a raw preset is set to default)
it crashes as well, same error in rundll32, thanks for looking into it though, I know it's not terribly important.
Did you happen to look into the problem with Avisynth scripts carrying over in the export registry file by chance?I cannot reproduce. If I enter random value to "activePreset" in registry, the highlight is lost. ffdshow still does not crash.
But I think loss of highlight won't happen with 4155.
Avisynth script exporting: Not yet. I'll do it later.

[ryrynz]

I think a reboot might have helped because now when I select a default in raw the default in normal ffdshow is not changed and it remains in bold, thanks!
I've always thought it would be nice to be able to access the raw config by shortcut, is this possible?

[haruhiko_yamagata]

I've always thought it would be nice to be able to access the raw config by shortcut, is this possible?
Copy the short cut for video decoder and replace "configure" with "configureRaw".

[Midzuki]

Copy the short cut for video decoder and replace "configure" with "configureRaw".

It would be nice if the installer itself created those shortcuts that "nobody uses" :rolleyes: :)

[ryrynz]

LOL, thanks for the backup there Midzuki, I don't want to seem too demanding, he's already fixed a bug I've put up with for years I think.
If he fixes the other one I be a happy chappy. It's great having a developer getting stuck in and moving ffdshow forward.

[haruhiko_yamagata]

It would be nice if the installer itself created those shortcuts that "nobody uses" :rolleyes: :)
It would be one line, but it depends on the needs. More items in the start menu may confuse the beginners.
clsid, what do you think?

[clsid]

There is no and will be no shortcut for the raw filter because it is not a filter that is used automatically. It is a filter that must be manually inserted by an application, and that app is responsible for access to its config. Apps that use it for the purpose it was designed for do typically provide a way to access its settings. Be it by a config button or otherwise. In MPC the config can be accessed by double-clicking on the filter name in the external filters dialog window.

It is a filter for expert users. It is not for the average joe. Adding it to the start menu will make people mess with its settings, without them understanding the purpose of the filter, making them complain about it not working as they might expect.

I suspect some here are also misusing the filter. In most cases the regular ffdshow video filter should be used with raw video enabled.

[clsid]

Clsid, maybe you remember the FLV4/VP6 cropping issue.
Nev has found a solution (http://code.google.com/p/lavfilters/source/detail?r=859e3bcc3aa6) for LAV Video Dec, could it be implemented in FFDshow too?A similar solution was used for a short while in past, but it is not a proper solution for ffdshow because ffdshow has filters that can change the dimensions of the video (resize,crop,letterbox). Such changes must be taking into account when sending cropping info downstream. Fixing it in ffdshow is thus less trivial.

[TheShadowRunner]

A similar solution was used for a short while in past, but it is not a proper solution for ffdshow because ffdshow has filters that can change the dimensions of the video (resize,crop,letterbox). Such changes must be taking into account when sending cropping info downstream. Fixing it in ffdshow is thus less trivial.
Arg, that's what I feared ^^;
Thanks for your reply.

[Midzuki]

@ clsid: Thanks for making things clear :thanks:

[ryrynz]

It is a filter for expert users. It is not for the average joe. Adding it to the start menu will make people mess with its settings, without them understanding the purpose of the filter, making them complain about it not working as they might expect.

I suspect some here are also misusing the filter. In most cases the regular ffdshow video filter should be used with raw video enabled.

I was thinking the reason behind this was along these lines, perhaps it could be an option selected under components when installing? Anyway it's rather trivial and Haruhiko has already provided a solution. I have always accessed it in when it was functioning through the player.

I am curious as to how one could misuse the filter.

[cyberbeing]

Libavcodec still decodes AC3 4ch (Front Left, Front Right, Front Center, Back Center) with the incorrect channel mapping. Only Liba52 and LAV Audio get that uncommon mapping correct.

See the sample JEEB posted a couple weeks ago:
http://forum.doom9.org/showpost.php?p=1543336&postcount=14270

[clsid]

LAV Audio is also not doing it correctly I think. It indicates a channel mask of 0x3f as input, which is the standard 5.1 mapping. It is just a coincidence that the three channels that contain sound in jeebs sample are at the same positions in both mappings.

[nevcairiel]

LAV Audio is also not doing it correctly I think. It indicates a channel mask of 0x3f as input, which is the standard 5.1 mapping. It is just a coincidence that the three channels that contain sound in jeebs sample are at the same positions in both mappings.

Thats a bug on the status page, the actual layout of the file is 0x107, which you can see if you turn off "Convert Output to Standard Channel Layouts" (also, fixed the bug)

[JEEB]

Libavcodec still decodes AC3 4ch (Front Left, Front Right, Front Center, Back Center) with the incorrect channel mapping. Only Liba52 and LAV Audio get that uncommon mapping correct.

See the sample JEEB posted a couple weeks ago:
http://forum.doom9.org/showpost.php?p=1543336&postcount=14270
This also reminds me that the sample I originally posted had no audio on the BC channel, which would make check-ups harder >_>

This (http://www.cccp-project.net/beta/test_files/4ch_ac3sample_with_bc.mka) one actually has that as well.

[clsid]

Fixed at 4172.

[Reino]

I believe it's been reported before, but anyway:
- http://media.nero.com/Standard720x576_1MBit.mp4 (http://media.nero.com/Standard720x576_1MBit.mp4)
AAC Audio part decoding failure with libavcodec. No problem though with libfaad2.Fixed.Although I can confirm the other 3 issues have indeed been fixed, I have to say this one still persists.

Btw, with your latest rev, clsid, the About-tab still reports rev4127.

[hoborg]

BTW, what about "removing FFDshow DXVA decoder from graph (http://www.youtube.com/watch?v=KZreMRTsuXk)" crash? It is still there. It make it nearly useless in LiveTV applications :/
Anyone interested to fix it?

Maybe this crash is connected with FFDShow DXVA video decoder icon which remaining in systray after playback, until hower cursor over it.

This issue is still active, so i added it to bug tracker (https://sourceforge.net/tracker/?func=detail&aid=3462373&group_id=173941&atid=867360).

[hoborg]

- this h.264 sample (http://hobring.esero.net/saf/samples/ffdshow_dxva_freeze.zip) have still problem with FFDShow DXVA decoder - picture freeze, but sound continue to play. This doesnot happend with MPC-HC DXVA decoder. Strange.

Still here. Added to bugtracker (https://sourceforge.net/tracker/?func=detail&aid=3462376&group_id=173941&atid=867360).

[clsid]

@CoRoNe
It works fine here with Haali splitter and LAV Splitter. Those are the only splitters I care about. Also, I have already done everything I can to fix it.

@hoborg
I get no crash. I can't fix it.

Has the DXVA seek issue existed forever, or did it work in an older revision?

[Dstruct]

This file gives incorrect display aspect ratio (1:1 Pixel) in WMP11 and MPC-HC via ffdshow rev4170 DXVA:

General
Format : AVI
Format/Info : Audio Video Interleave
File size : 1.29 GiB
Duration : 1h 59mn
Overall bit rate : 1 551 Kbps
Writing library : VirtualDub build 32839/release

Video
ID : 0
Format : AVC
Format/Info : Advanced Video Codec
Format profile : Main@L3.0
Format settings, CABAC : Yes
Format settings, ReFrames : 3 frames
Codec ID : H264
Duration : 1h 59mn
Bit rate : 1 346 Kbps
Width : 720 pixels
Height : 576 pixels
Display aspect ratio : 16:9
Frame rate : 25.000 fps
Standard : PAL
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 8 bits
Scan type : Progressive
Bits/(Pixel*Frame) : 0.130
Stream size : 1.12 GiB (87%)

Audio
ID : 1
Format : MPEG Audio
Format version : Version 1
Format profile : Layer 3
Mode : Joint stereo
Codec ID : 55
Codec ID/Hint : MP3
Duration : 1h 59mn
Bit rate mode : Constant
Bit rate : 192 Kbps
Channel(s) : 2 channels
Sampling rate : 48.0 KHz
Compression mode : Lossy
Stream size : 164 MiB (12%)
Alignment : Aligned on interleaves
Interleave, duration : 24 ms (0.60 video frame)
Interleave, preload duration : 504 ms


Proper display aspect ratio in Media Player Classic HC 1.5.3.3899 via internal MPC DVXA decoder!

[Dstruct]

"Info & CPU" tab also doesn't display any values from the video stream. "Surface Overlay" post processing option in ffdshow's DXVA decoder makes the videos go black or non-working here.


NVIDIA GeForce GTS 450

[hoborg]

@hoborg
I get no crash. I can't fix it.


If you drag and drop h.264 video in Graphstudio - does you have FFDshow DXVA video decoder + EVR (VMR9) chain by default?
The merit of EVR (VMR9) needs to be changed first.

[szabi]

Hi

Some avi video were recorded with ffdshow. Encoder set HuffYUV (FFmpeg variant).
When i insert them to an avisynth script as avisource this is the result: couldn't locate a decompressor for fourcc ffds
How could i solve this?

bye
szabi

[Reino]

szabi, you could use FFVideoSource() (http://avisynth.org/mediawiki/FFmpegSource) (preferred), or alternatively DirectShowSource().

@CoRoNe
It works fine here with Haali splitter and LAV Splitter. Those are the only splitters I care about. Also, I have already done everything I can to fix it.That's really odd. I'm using the latest LAV Splitter and I've tried your rev4174 and xvidvideo.ru's, but with libavcodec, unlike libfaad2, the issue still persists. I'll keep an eye on it.

[Eliminateur]

since rev >4142 i'm experiencing crashes on MPC-HC whenever i try to open big hi10p 1080p MKV files.

i have 5 of them in one directory and they all crash on open directly, if i force LAV decoder in the mpc-hc preferences, it works ok. Same if i use the internal filters.

i've updated to clsid r4174 and it's the same.

i'm using all x64 versions with haali splitter.
no DXVA -all soft decode-, RGB32 output, HQ YV12 conversion.
Latest mpc-hc r3903, was usin r3887 before(it crashed the same)

Builds i also tested:
4150 crash
4158 crash

[Snowknight26]

r4174 crashes when trying to play a lossless H.264 stream encoded with x264, but only when --output-csp (output colorspace) is specified as 'rgb'. 'i420', 'i422', and 'i444' all work correctly.

[ageback]

When compiling ICL12 version it always fail on IntelQuickSyncDecoder project, since rev.4150.
Here's the error log:

1>src\QuickSyncExports.cpp(63): error : expression must have (pointer-to-) function type
strcpy(ver, QS_DEC_VERSION " by Eric Gur. Compiled using " COMPILER " " COMPILER_VER " (" __DATE__ " " __TIME__ ")");
^

1>src\QuickSyncExports.cpp(63): error : expected a ";"
strcpy(ver, QS_DEC_VERSION " by Eric Gur. Compiled using " COMPILER " " COMPILER_VER " (" __DATE__ " " __TIME__ ")");

1>icl : error #10298: problem during post processing of parallel object compilation
TimeManager.cpp
1>C:\Program Files\MSBuild\Microsoft.Cpp\v4.0\Platforms\Win32\PlatformToolsets\Intel Parallel Composer 2011\Microsoft.Cpp.Win32.Intel Parallel Composer 2011.targets(196,5): error MSB6001: “icl.exe”的命令行开关无效(translation by me:Command line swich invalid)。参数“path”的长度不能为零(translation by me:The length of parameter "path" cannot be 0.)。

[clsid]

r4174 crashes when trying to play a lossless H.264 stream encoded with x264, but only when --output-csp (output colorspace) is specified as 'rgb'. 'i420', 'i422', and 'i444' all work correctly.
Sample file needed.

[clsid]

since rev >4142 i'm experiencing crashes on MPC-HC whenever i try to open big hi10p 1080p MKV files.

i have 5 of them in one directory and they all crash on open directly, if i force LAV decoder in the mpc-hc preferences, it works ok. Same if i use the internal filters.

i've updated to clsid r4174 and it's the same.

i'm using all x64 versions with haali splitter.
no DXVA -all soft decode-, RGB32 output, HQ YV12 conversion.
Latest mpc-hc r3903, was usin r3887 before(it crashed the same)

Builds i also tested:
4150 crash
4158 crash
Hi10p plays fine here in GraphStudio x64.
What is the crashing module?
Does x86 version crash as well?