ffdshow tryouts project: Discussion & Development

[max-pain]

I have problem with the audio in this video (with the last couple of SVN builds):

http://www.dcresource.com/reviews/canon/powershot_sx200-review/MVI_0093.MOV

Source: http://www.dcresource.com/reviews/canon/powershot_sx200-review/

[clsid]

That file contains uncompressed PCM audio. It plays OK without ffdshow. Enabling processing of uncompressed audio in ffdshow indeed results in bad sound.

[max-pain]

I don't have audio without ffdshow in MPC-HC:( But that's obviously not a ffdshow problem.

[STaRGaZeR]

I have problem with the audio in this video (with the last couple of SVN builds):

http://www.dcresource.com/reviews/canon/powershot_sx200-review/MVI_0093.MOV

Source: http://www.dcresource.com/reviews/canon/powershot_sx200-review/

I hear that sample without issues, with and without ffdshow.

[ikarad]

I knew it, please add delay (in the dialog 'Delay:' plus 'Keys & remote' numpad + and -) until I have time to fix it.

excuse me but i would like to known if you expect to fix this problem soon or later (some months for example)?

[Snowknight26]

Since I'm not sure how to contact the devs of ffmpeg-mt, I thought I'd post this in the next best place. Would be nice if someone could pass it on to the devs.

Some corruption is noticable in this clip with ffmpeg-mt but not with libavcodec: http://stfcc.org/misc/ffmpeg-mt.mkv

[clsid]

This is the dev you are looking for:
http://forum.doom9.org/member.php?u=72044

[Shakey_Jake33]

Apologies if this is a known issue, it's quite a mission to trawl back through this thread to find out. I use ffdshow to deinterlace content decoded by other codecs (by enabling 'Raw Video->All Supported' in the codec options), because YADIF does an excellent job. So I have CoreAVC decoding my H.264 videos, and ffdshow deinterlaces.

This doesn't work for me in Windows 7 however, even though ffdshow is set up correctly, it is unable to place itself after CoreAVC in the filter chain. This means I either have to deinterlace using CoreAVC (giving inferior results) or decode using ffdshow. This is probably a Windows 7 problem, but I figure it's worth mentioning.

Using Windows 7 Release Candidate (7100) 64-bit, MPC-HD Beliyaal build 32-bit (SVN 1044, Beliyaal fixes ver. 25), ffdshow 32-bit 2896, EVR Custom renderer.

[clsid]

That's a Windows 7 problem. Setting ffdshow as a preferred external filter in MPC might help.

[Shakey_Jake33]

^Setting 'ffdshow raw video filter' as preferred in MPC did solve the problem, thanks!

[ikarad]

edit error

[Monarc]

one question: how can i use the crf Mode of the x264 encoder inside ffdshow? is it supported?

[haruhiko_yamagata]

It's likely that the Subtitle filter in ffdshow was never intended to allow external subs to be used with DVD's, and I'm not sure if it's possible to do so technically, but I'll ask anyway.

I am able to point the subtitle filter to an external .srt file while a DVD is playing, and if I play it from the beginning without jumping around in the file, then the timecodes in the .srt match up with the proper time the subs are expected in the video, but I guess the subtitle filter isn't able to read the actual timestamp from either the video stream or the player like it can for a normal media files, so when I chapter slip or even just jump ahead a minute or two, the subs get out of sync with the video.

Is it possible to modify the behavior so that external .srt files could be loaded and read the proper timecode and display at the correct time when playing DVD's, the same way they do for other media files? This would save a lot of hassle when watching foreign films that don't have english subs on the disc itself.

Many thanks!
You are right, it's not supported. As far as my skill goes, it's technically impossible.

[STaRGaZeR]

one question: how can i use the crf Mode of the x264 encoder inside ffdshow? is it supported?

Also, it's possible to add x264 to the VfW part of ffdshow64 now that it compiles for 64-bit Windows?

[STaRGaZeR]

Bug with the lastest clsid x64 build: when an interlaced DVD is played in MPC and yadif is used for deinterlacing, jumping letters can be seen in the menus for example. This does not happen with rev2895.

DVD menu sample: here (http://www.megaupload.com/?d=R3JHNSQH) (22MB). You only have to open it as DVD in MPC and use yadif to deinterlace the menu.

[haruhiko_yamagata]

Bug with the lastest clsid x64 build: when an interlaced DVD is played in MPC and yadif is used for deinterlacing, jumping letters can be seen in the menus for example. This does not happen with rev2895.

DVD menu sample: here (http://www.megaupload.com/?d=R3JHNSQH) (22MB). You only have to open it as DVD in MPC and use yadif to deinterlace the menu.
Thanks, I have fixed at rev 2914.

[clsid]

Are all known DVD related problems fixed now?

Recently I have seen a lot of reports from people that get a crash in Vista Media Center when they browse the video library. Can anyone here confirm if this problem is still present in current ffdshow builds?

[tetsuo55]

Are all known DVD related problems fixed now?

Recently I have seen a lot of reports from people that get a crash in Vista Media Center when they browse the video library. Can anyone here confirm if this problem is still present in current ffdshow builds?

Not everything iirc, and definitely not with all combination's of filters.

[movmasty]

Thanks, I have fixed at rev 2914.
In 2895 huffyuv encoding is broken, in the output vid appears a black bar of 16 pixels on the right


-
downloaded here http://www.free-codecs.com/download/FFDshow.htm
on source forge there is another version, 2527 ,where to dl the right then?
-
same prob also with 2913, dont find the 2914
-
back to the 2527, it is ok

[mark0077]

Are all known DVD related problems fixed now?

Recently I have seen a lot of reports from people that get a crash in Vista Media Center when they browse the video library. Can anyone here confirm if this problem is still present in current ffdshow builds?

I will do a full test this evening with some DVD menus that often crash / freeze / act extremely slowly with some recent ffdshow builds.

I also don't remember haruhiko adding the code he talked about here, to enforce 16:9 or 4:3 aspect ratios during resizing of dvd's (like some renderers seem to do) as this often leads to bad aspect ratios. This isn't an ffdshow problem but a workaround that I feel needs to be in there to make resizing of dvd's useful. Will test this later also unless haruhiko has input on it.

[clsid]

I have just tested the latest revision on Vista x64 and ffdshow still crashes Media Center. It happens while browsing the video library when MC is generating thumbnails for multiple files. It seems to be some kind of race condition problem.

@Albain & Haruhiko, any idea on a solution for this nasty problem?

@all, anyone here with 32-bit Vista that can confirm if ffdshow also in 32-bit Media Center?

[Mercury_22]

I CAN'T play (some) files with LPCM sound stream(s) although MPC-HC's internal LPCM filter it's playing fine the same file(s). (FFDShow svn 2913 default settings)

Is this a known problem (bug) ?

Can this be fixed ?

Sample (http://rapidshare.com/files/226414424/LPCM_TEST.m2ts.html) or Mirror (http://www.megaupload.com/?d=AO7343UP)

[albain]

I have just tested the latest revision on Vista x64 and ffdshow still crashes Media Center. It happens while browsing the video library when MC is generating thumbnails for multiple files. It seems to be some kind of race condition problem.

@Albain & Haruhiko, any idea on a solution for this nasty problem?

@all, anyone here with 32-bit Vista that can confirm if ffdshow also in 32-bit Media Center?

I have reproduced this problem in recent revisions but randomly and only in release builds.
However I remember that the boost library was in cause something like a missing lock in a thread (and I think that the problem matches with a bug report on sourceforge tracker).
Is this possible to set in boost the number of threads to 1 for testings ?

[clsid]

I did some testing with a couple old builds and I think the problem starts at revision 2876.

[therealjoeblow]

The "set interlace flag in output media type" (formerly called HW deinterlacing) doesn't appear to be working for me. I tried it in several recent builds from 2880 to the current 2918, and it doesn't seem to do anything (XP32-SP3, VMR7 or VMR9, using ffdshow libmpeg2 to decode DVD's). When enabled, the video is still showing combing just the same as when it's disabled, it doesn't seem to trigger the hardware deinterlacing routines. PDVD and TheaterTek do, when their "use hardware accelerator" tickbox is enabled DVD's get deinterlaced by the video card.

I tried on 2 different machines, one has a gigabyte mb with an AMD 780G chipset (Radeon HD3200) and the other one has an asus mb with an AMD 690 chipset (Radeon X1250). I tried on several different versions of Catalyst, including the latest v-9.4 and it doesn't seem to make any difference, hardware deinterlacing never actually gets invoked with this ffdshow setting.

[Leak]

The "set interlace flag in output media type" (formerly called HW deinterlacing) doesn't appear to be working for me.
Make sure you force ffdshow to output NV12 video by unchecking all other colorspaces.

[clsid]

A tip useful for reproducing the MC thumbnails issue:
1) Enable view of hidden system files in the Explorer folder options
2) Delete the files "ehthumbs_vista.db" to clear the thumbnail cache. This file is created in each folder of the video library of MC.

[therealjoeblow]

Make sure you force ffdshow to output NV12 video by unchecking all other colorspaces.

I tried that, and using the OSD to verify, it shows it's outputting "NV21,VU" (not NV12), even though that's the only colorspace enabled; and it doesn't make any difference any way, the hardware deinterlacing never kicks in.

[alter4]

In all my mpeg1 25 fps 320x240 video files seeking does not work (video freeze but sound carries on) when decoding with libMpeg2 (libavcodec is ok)!
Tested on WINXp sp3 x86, MPC and WMP11, ffdshow beta6 and latest build rev2913
CPU core 2 duo 6550, nvidia 630i/7100 motherboard
Can anybody confirm the bug?

[albain]

A tip useful for reproducing the MC thumbnails issue:
1) Enable view of hidden system files in the Explorer folder options
2) Delete the files "ehthumbs_vista.db" to clear the thumbnail cache. This file is created in each folder of the video library of MC.

I have reproduced the problem but I need some help :

FFDShow crashes in TimgFilterSubtitles.cpp:572 :
boost::unique_lock<boost::mutex> lock(mutex_prefs);

An assertion is raised there :
basic_timed_mutex.hpp:64
void lock()
{ BOOST_VERIFY(timed_lock(::boost::detail::get_system_time_sentinel()));
}


http://damienbt.free.fr/Capture.PNG

EDIT : apparently there is a bug that has been fixed in boost 1.38 concerning this line.
Which version of boost are we using ?

[haruhiko_yamagata]

I tried that, and using the OSD to verify, it shows it's outputting "NV21,VU" (not NV12), even though that's the only colorspace enabled; and it doesn't make any difference any way, the hardware deinterlacing never kicks in.
I think the DVD is flagged progressive. Please try "Force bob" in "Set interlace flag in output media type"-"Method:".

[haruhiko_yamagata]

I have reproduced the problem but I need some help :

Is this issue limited to x64 version?

[albain]

Is this issue limited to x64 version?

No, I am using 32 bits version (Vista, also Windows 7 has the issue).

EDIT :

Also there is a bug with this function which has been fixed in boost 1.38 (released in february 2009) and as far as I know our version is older so this may be it


Also I have seen a similar issue in the following thread :
http://www.nabble.com/-thread--Assertion-failed-get_system_time_sentinel()-td22779251.html

Something like :
"This is undefined behaviour --- your threads are outliving the variables
they are accessing. You need to join with the thread before you destroy
the object. Since you start the thread in the constructor, it would make
sense to call join in the destructor. "

[therealjoeblow]

I think the DVD is flagged progressive. Please try "Force bob" in "Set interlace flag in output media type"-"Method:".

I tried auto, force bob and force weave on both of the different machines with exactly the same result, it still shows an interlaced picture with combing with every setting, the picture is never changed.

It's the R1 NTSC WS version of Good Night and Good Luck (2005) [original factory dvd, not a backup], but I also see the same behavior with all of the SDTV DVD's that I record on my Pioneer DVR-640 standalone dvd recorder, they are interlaced regular definition television content, and ffdshow fails to invoke the hardware deinterlacing on all of these too, while the commercial decoders somehow are be able to.

[albain]

Meanwhile, it seems that defining the following preprocessor macro solves the problem : BOOST_DISABLE_ASSERTS

This is not a definitive solution but it will help before figuring out the real problem.

[Amour]

I installed ffdshow on Windows 7 (Release Candidate), but it is not working!
Instead of that, my videos are being played using the default Windows Media Foundation system.
As I have issues with this, is there a way for me to use ffdshow filters on Windows 7 for any video?

Is ffdshow going to use Windows Media Foundation in the near future in order to override some of Windows 7 default settings?

[clsid]

The Microsoft players will use their own decoders, regardless of what framework ffdshow would use. If you want to use ffdshow then either use a different player or rename the Windows Media Foundation files (Google for more info).

[alter4]

Hey guys! Looks like nobody have read my post (http://forum.doom9.org/showthread.php?p=1279113#post1279113)
Please, help. Could you confirm this?

[therealjoeblow]

Hey guys! Looks like nobody have read my post (http://forum.doom9.org/showthread.php?p=1279113#post1279113)
Please, help. Could you confirm this?

You only posted the message 2 hours ago! What are you expecting - personal on demand service?

Most of the dev's for ffdshow are scattered around the world. I'd hazard a guess that some of them are sleeping right now, and may or may not respond sometime later... Often, replies come days later after someone has time to do some testing.

[_xxl]

Can Boost be updated to 1_38_0?

[rack04]

Hey guys! Looks like nobody have read my post (http://forum.doom9.org/showthread.php?p=1279113#post1279113)
Please, help. Could you confirm this?

It would also be nice if you could post a sample.

[clsid]

ffdshow compiles fine with Boost revision 51018 (= v1.38) and 52652 (latest), but the crashing problem is still present. So some changes to the ffdshow code might also be needed as Albain pointed out. Nevertheless, updating Boost will be a good idea.

I am now building with Albain's temporary workaround.
Edit: the workaround didn't work for me.

[tetsuo55]

It might be a good idea to create a troubleshooting/bug reporting guideline similar to the one i made for MPC-HC.
Often problems reported here are vague and dev's waste a lot of time trying to reproduce

see example here
http://forum.doom9.org/showthread.php?p=1278636#post1278636

(I am willing to write it)

[Amour]

I'm not using any Microsoft players. I'm using MPC.

[clsid]

Set ffdshow as a preferred external filter in MPC. If that doesn't work, go complain to Microsoft.

[albain]

ffdshow compiles fine with Boost revision 51018 (= v1.38) and 52652 (latest), but the crashing problem is still present. So some changes to the ffdshow code might also be needed as Albain pointed out. Nevertheless, updating Boost will be a good idea.

I am now building with Albain's temporary workaround.
Edit: the workaround didn't work for me.


Okay, I will digg this around. I think that this is a design problem as pointed out in the post:
http://www.nabble.com/-thread--Assertion-failed-get_system_time_sentinel()-td22779251.html

Haruhiko solved some similar bugs maybe he has some leads.

EDIT : @clsid, have you made a "touch" (file modification) on the file where the macro is defined (to make it recompile) ?

[haruhiko_yamagata]

Okay, I will digg this around. I think that this is a design problem as pointed out in the post:
http://www.nabble.com/-thread--Assertion-failed-get_system_time_sentinel()-td22779251.html

Haruhiko solved some similar bugs maybe he has some leads.
The thread is joined (terminated) before it is destructed.

I think it will crash if the thread that locks the mutex and is being waited for by another thread crashes, the mutex may be unlocked illeaglly and the thread that is waiting for the mutex would crash with that message.

Do the settings matter? Subtitle filter, external subtitle file?

[movmasty]

You only posted the message 2 hours ago! What are you expecting - personal on demand service?

mine http://forum.doom9.org/showthread.php?p=1278711#post1278711
30 hurs ago :helpful:

[lych_necross]

Hey guys! Looks like nobody have read my post (http://forum.doom9.org/showthread.php?p=1279113#post1279113)
Please, help. Could you confirm this?
Patience alter4, patience.

[albain]

The thread is joined (terminated) before it is destructed.

I think it will crash if the thread that locks the mutex and is being waited for by another thread crashes, the mutex may be unlocked illeaglly and the thread that is waiting for the mutex would crash with that message.

Do the settings matter? Subtitle filter, external subtitle file?

Why would the thread crash ?
The lock is necessary if the variable may be modifed while another thread is reading it. Is it the case here ?

This problem seems to occur when there are several instances of FFDShow running at the same time to extract a frame and build the thumbnails.
I think that the crash occurs because the threads termination is close to the thread creation (extracting one frame is very fast) and maybe before.
I am trying to output the error code that is generated by the assertion, it will explain the exact cause