ffdshow tryouts project: Discussion & Development

[nevcairiel]

If you switch the VC-1 decoder in ffdshow to "wmv9" (i think it was called), it should work. avcodec just does not support interlaced vc-1.

Other people seem to be using either the MS Decoder (considering its their own format, i would figure they do a good job at decoding it), or the Cyberlink/ArcSoft solutions.

[LigH]

same problem for ntsc mpeg2 when it should read 29.976 the osd shows it as 29.97.

24000/1001 = 23.976023~

30000/1001 = 29.970029~

[jmone]

If you switch the VC-1 decoder in ffdshow to "wmv9" (i think it was called), it should work. avcodec just does not support interlaced vc-1.

Other people seem to be using either the MS Decoder (considering its their own format, i would figure they do a good job at decoding it), or the Cyberlink/ArcSoft solutions.

I've tried both the wmv9 and the MS DMO decoder, both are OK but the deinterlacing is pretty ordinary with visable fields (looks like weave to me)

[fastplayer]

@Devs:
I've fixed a few typos in the ffdshow.rc:
http://www.mediafire.com/?bobhnrul2i1fesq
Please review and commit. Thanks!

[tal.aloni]

fastplayer,
reviewed and committed, Thanks!

[tal.aloni]

Hi all,
there was an oddity that I fixed in rev. 3789:
prior to rev. 3789, all subtitles format except substation alpha formats (SSA / ASS / ASS2) could be turned on / off based on preset.
however, I couldn't see an apparent reason for this discrepancy, of course preset info is not available at early loading stage, but it seemed irrelevant to this case.
so in rev. 3789, substation alpha formats can be turned on / off based on preset as well.

if you have such subtitles and you notice any problems with rev. 3789 or later, please let me know.

b.t.w. it appears as if some other settings on this page can also be preset settings, but I'll leave it for now.

[fastplayer]

fastplayer,
reviewed and committed, Thanks!
Thanks! I see you resized the "Search in" radio button which I was about to report :D
There are some issues with colons in the "Resize & aspect" options, specifically "Specify aspect ratio" and "PAR out of device". Both "reach" into another control element. Not a biggie but maybe you can take a look at those.
Anyway, why isn't DXVA installed by default? It should be, IMO.

[tal.aloni]

Why isn't DXVA installed by default? It should be, IMO.
I believe DXVA can confuse many users who are not familiar with the technology, it's advantages and disadvantages (no post-processing, etc.), or don't even know if their hardware even support DXVA.
I think that for the non-savvy, our recommendation should be to skip DXVA usage completely, and only when there is a good reason to use DXVA, learn about it and install it,
at that stage, the user will figure out for himself to select DXVA during installation.


There are some issues with colons in the "Resize & aspect" options, specifically "Specify aspect ratio" and "PAR out of device"
fixed in rev. 3791

[fastplayer]

I think that for the non-savvy, our recommendation should be to skip DXVA usage completely, and only when there is a good reason to use DXVA, learn about it and install it,
at that stage, the user will figure out for himself to select DXVA during installation.
You're right. It requires quite a bit of knowledge to setup properly and deal with its limitations. Then there are out-of-date, buggy drivers...
fixed in rev. 3791
:thanks:

[fastplayer]

Updated German language file:
http://www.mediafire.com/?v10ngh2wi0mb9i3

[tal.aloni]

Updated German language file:
http://www.mediafire.com/?v10ngh2wi0mb9i3

committed as rev. 3794

[V!roX]

Hey, sorry for asking something that is probably really easy, but...

I just installed MPC-HC and ffdshow tryouts, because I wanted to test the superior subtitle rendering compared to VobSub.
But it looks like I can't get it to work: Apparently, my subtitles are rendered by MPC-HC itself. First without effects, and after turning them on in the MPC options, they started flickering.
It doesnt matter if the subtitles filter in the ffdshow tryouts config is activated or not, MPC-HC seems to ignore ffdshow on subtitle rendering.

How can I connect the ffdshow subtitle rendering to MPC-HC so it works without problems?
by the way, I'd prefer if subtitle effects were enabled.

[khagaroth]

Update for Czech translation:
ticket 3250134 (https://sourceforge.net/tracker/?func=detail&aid=3250134&group_id=173941&atid=867362)

[tal.aloni]

Update for Czech translation:
ticket 3250134 (https://sourceforge.net/tracker/?func=detail&aid=3250134&group_id=173941&atid=867362)

done. please submit a TortoiseSVN patch file next time.

[Gleb Egorych]

Is it known problem that ffdshow has problem with interlaced MPEG2 content?

The situation: Win7, EVR, libavcodec or libmpeg2 (doesn't matter), YV12 or NV12 output (doesn't matter), haali or gabest ts splitter (doesn't matter).
The problem: if "Set interlaced flag in output media type" is set to "Auto" then frame rate flows from ~35 to ~50 fps on my interlaced PAL samples, combing is obvious.

If force bob deinterlacing and proper field order then there is no combing and I have proper 49-50 fps.

[hoborg]

Is it known problem that ffdshow has problem with interlaced MPEG2 content?

The situation: Win7, EVR, libavcodec or libmpeg2 (doesn't matter), YV12 or NV12 output (doesn't matter), haali or gabest ts splitter (doesn't matter).
The problem: if "Set interlaced flag in output media type" is set to "Auto" then frame rate flows from ~35 to ~50 fps on my interlaced PAL samples, combing is obvious.

If force bob deinterlacing and proper field order then there is no combing and I have proper 49-50 fps.

I don't have such problems.
I use MPEG profile for output to NV12 becouse of HW deinterlacing and using it for DVB-t LiveTV.

FFDshow 3775, MSVS 2010 (x86)

[clsid]

I have found a regression related to the recent cleanup of the Output page in ffdshow video decoder. It has broken thumbnailing by Windows Explorer. I have narrowed it down to the setting "Allow output format changes during playback" (which has been removed). Thumbnailing works when that old setting is in indeterminate state, but not when fully checked.

@stargazer69
Can you have a look at this issue?

Here is a reg tweak to enable thumbnailing if needed:
Windows Registry Editor Version 5.00

[HKEY_CLASSES_ROOT\.flv\ShellEx\{BB2E617C-0920-11D1-9A0B-00C04FC2D6C1}]
@="{c5a40261-cd64-4ccf-84cb-c394da41d590}"

Edit: should be fixed in r3800

[Chikuzen]

Hi, devs.
Why did you change Japanese transration ?
r3796 changed some correct transration to wrong.

Please revert r3796, and return them to Yamagata's.

[Ger]

Is it known problem that ffdshow has problem with interlaced MPEG2 content?

The situation: Win7, EVR, libavcodec or libmpeg2 (doesn't matter), YV12 or NV12 output (doesn't matter), haali or gabest ts splitter (doesn't matter).
The problem: if "Set interlaced flag in output media type" is set to "Auto" then frame rate flows from ~35 to ~50 fps on my interlaced PAL samples, combing is obvious.

If force bob deinterlacing and proper field order then there is no combing and I have proper 49-50 fps.

I have the same problem. I always assumed this was due to badly flagged MPEG-2 streams (DVB sourced in my case). IIRC the Microsoft decoder shows the same combing as ffdshow in Auto/Auto mode.

One problem with the current flagging algo (MatMaul's code IIRC) is that if force bob is used and field order can't be detected the auto field order falls back to bottom field first even for PAL content. Since PAL is nearly always top field first it should fall back to TFF for PAL IMHO. The optimal solution would be to improve the auto field order detection and auto flagging in general, but I don't even know if that's possible. TVs and STBs don't have these combing/field order issues though.

The (automatic and not configurable) implementation in MPC-HC's libmpeg2 based MPEG-2 decoder works better for me (with JanWillem32's test build it outputs NV12 with the "set interlaced flag..." option cheked). I see no combing with that decoder with my problem sample (http://www.multiupload.com/R3C27RMK54), but it has other weaknesses compared to ffdshow, less configurable, no support for 4:2:2 sources and needs FTC to be on in MPC-HC + EVR-CP with some other samples.

[ilkertezcan]

My question to developers: There isn't Picture properties on FFDshow DxVA...
Is impossible change to gamma/brightness etc. settings of the video when actived DxVA mode?

[Gleb Egorych]

I always assumed this was due to badly flagged MPEG-2 streams (DVB sourced in my case)
Tested on DVB material too.

IIRC the Microsoft decoder shows the same combing as ffdshow in Auto/Auto mode.
Exactly.

[tal.aloni]

Hi, devs.
Why did you change Japanese transration ?
r3976 changed some correct transration to wrong.

Please revert r3796, and return them to Yamagata's.
I've contacted tnetsixenon, which supplied me with the patch.

@developers
apparently there is no constant order to the translation files,
and some people have the tendency to reorder those,
maybe we should force some order to those files?
(alphabetically or based on dialog pages are the two that comes to mind)

Regards,
Tal

[Tom Keller]

Any ideas, why this kind of wmv3 videos:

http://www.megaupload.com/?d=4UQ8SHDP

... plays all blocky/green with ffdshows DirectShow-Decoder inside MPC-HC or other DirectShow-Players, while playing perfectly fine inside VirtualDub with WMV-Plugin (using ffdshows VfW decoder) and looking fine too inside Microsofts old Media Player 6.4 (using ffdshows DirectShow decoder). It's confusing! I assume: ffdshow seems to be the source of the problem but only in combination with Microsofts qasf.dll (v11.0.5721.5145) - and since VirtualDubs WMV-Plugin didn't use it, and the old WMP6 uses the old wmpasf.dll, both work fine. Could someone confirm that?

[Midzuki]

Any ideas, why this kind of wmv3 videos:

http://www.megaupload.com/?d=4UQ8SHDP

... plays all blocky/green with ffdshows DirectShow-Decoder inside MPC-HC or other DirectShow-Players, while playing perfectly fine inside VirtualDub with WMV-Plugin (using ffdshows VfW decoder) and looking fine too inside Microsofts old Media Player 6.4 (using ffdshows DirectShow decoder). It's confusing! I assume: ffdshow seems to be the source of the problem but only in combination with Microsofts qasf.dll (v11.0.5721.5145) - and since VirtualDubs WMV-Plugin didn't use it, and the old WMP6 uses the old wmpasf.dll, both work fine. Could someone confirm that?

I have not looked at your sample videos yet :o , BUT
I suppose the problem is related to hardware acceleration,
please take a look at this thread:

http://forum.doom9.org/showthread.php?t=146346&page=2

[Tom Keller]

No... i don't think so ;) . The Microsoft DMO decoder works pretty well with the file (MPC-HCs internal decoder too) - only the ffdshow decoder won't. And since i did not use hardware accelerated VC-1 decoding and most of the "problem videos" are progressive, it's obviously neither hardware acceleration nor interlacing related...

[space1999]

No... i don't think so ;) . The Microsoft DMO decoder works pretty well with the file (MPC-HCs internal decoder too) - only the ffdshow decoder won't.

OK, I will take the time to download the samples and will "study the case". :)

And since i did not use hardware accelerated VC-1 decoding

Hardware-based video acceleration is not the same as hardware-based video decoding.

[Tom Keller]

I know - but i thought, hw based decoding via DXVA was one of the topics in the named thread (sorry... just skimmed through it... mea culpa :o )...


Just to show, what i mean...

Microsoft DMO Decoder with WM ASF Reader (qasf.dll) => no problems:
http://justpic.info/images2/f647/DMO.jpg

MPC Video Decoder with WM ASF Reader (qasf.dll) => no problems:
http://justpic.info/images2/3750/mpc_dec.jpg

ffdshow Video Decoder (libavcodec) with WM ASF Reader (qasf.dll) => grey screen; keyframes and occasional makroblocks popping through:
http://justpic.info/images2/30cd/ffdshow_libav.jpg

ffdshow Video Decoder (wmv9) with WM ASF Reader (qasf.dll) => green screen; only keyframes popping through:
http://justpic.info/images2/8378/ffdshow_wmv9.jpg

ffdshow Video Decoder (libavcodec/wmv9) with Windows Media source filter (wmpasf.dll) => no problems:
http://justpic.info/images2/4281/ffdshow_wmv_source.jpg

ffdshow Video Decoder (libavcodec/wmv9) with ArcSoft ASF Splitter (ASFSplitter.ax) => no problems:
http://justpic.info/images2/7379/arcsoft.JPG

Oh... and the problem is the same with AviSynth & DirectShowSource - since (in this case) the decoder delivers its uncompressed output directly to AviSynth, neither hardware accelerated video rendering nor video decoding is used. So i assume: it simply HAS to be the fault of the decoder.

Like i said: ffdshow in combination with qasf.dll seems to be the source of the problem. Since both, the DMO decoder and the MPC video decoder, work fine, ffdshow seems to have some problems in handling this kind of video stream delivered by the qasf.dll...

[cmbe]

Hi,

The ffdshow audio codec bitstreaming the dts core of a dts-hd track over spdif, isn't working in the latest releases (from almost a year by now).

This works ok with the old ffdshow beta 7.

Anyone knows if the support for this feature have been dropped?

thanks

[clsid]

apparently there is no constant order to the translation files,
and some people have the tendency to reorder those,
maybe we should force some order to those files?
(alphabetically or based on dialog pages are the two that comes to mind)It may be quite a lot of work, but if you want to do it, it would be great.

I would suggest sorting the dialog pages based on their numerical value, from low to high. Sort the contents of each page alphabetically.

An English template with all translatable strings would be useful as well. That would allow quicker translations compared to using the internal translation functionality. It would also allow seeing what is missing in other translations by using a text compare program (like ultracompare).

[clsid]

Hi,

The ffdshow audio codec bitstreaming the dts core of a dts-hd track over spdif, isn't working in the latest releases (from almost a year by now).

This works ok with the old ffdshow beta 7.

Anyone knows if the support for this feature have been dropped?

thanks
I don't have the hardware to confirm. Maybe anyone else here can confirm whether it works or not.

You may need to adjust the compatibility settings on the Output page, like the "connect as pcm first" setting.

[khagaroth]

It may be quite a lot of work, but if you want to do it, it would be great.

I would suggest sorting the dialog pages based on their numerical value, from low to high. Sort the contents of each page alphabetically.

An English template with all translatable strings would be useful as well. That would allow quicker translations compared to using the internal translation functionality. It would also allow seeing what is missing in other translations by using a text compare program (like ultracompare).
Yes, a template is badly needed. But I wonder how that would be created. A switch to gettext based translations would be even better.
The internal translation system works pretty similar to gettext, all the strings are marked by a keyword - _l - and are processed by a translation function. There is a catch though, the keyword is used even for strings that shouldn't be translated (so no easy migration to gettext unfortunately). Another problem with the internal translation system are the limitations of the translation file 'ini' format - one example is the inability to translate strings that contain '=' (there are a few there). And don't get me even started about the translation mode annoyances/limitations.

[cmbe]

I don't have the hardware to confirm. Maybe anyone else here can confirm whether it works or not.

You may need to adjust the compatibility settings on the Output page, like the "connect as pcm first" setting.

Thanks for your help clsid.

Nope, it won't work! no matter what settings I have.

It works perfectly with beta 7.

[tal.aloni]

Maybe anyone else here can confirm whether it works or not.

I have the hardware and will test within the next few days.

[Gew]

I feel obligated posting this here as well, since I'm not sure that any of the developers has actually read the thread I wrote ~10 months ago at SrcFrg/ffdshow board (http://sourceforge.net/projects/ffdshow-tryout/forums/forum/597593/topic/3784486).

Like this. Old builds of ffdshow are fabulous. Here, I pick an .IDX (VobSub) file from the right click menu in the system tray, you know, the ffdshow applet. Then, after choosing your .IDX VobSub file, you get to chose the desired language in the same menu; all languages found/available are shown in a list below.

This excellent feature was some how dropped(!) almost a year ago. Now it just defaults on the languages chosen in the preferred section of ffdshow configuration, which makes actual "quick change" much more a pain. Why this change? It's really only downside to it, AFAICS.

[cmbe]

I have the hardware and will test within the next few days.

:thanks:for your help.

[tal.aloni]

The ffdshow audio codec bitstreaming the dts core of a dts-hd track over spdif, isn't working in the latest releases (from almost a year by now).

EDIT:
I've nailed it down to a change in rev. 3160,
when detecting that the stream is DTS-HD, we had erroneously set the sample rate to 96000 instead of 48000 (DTS over SPDIF requires 48000),
fixed for rev. 3802, thanks for reporting.

Regards,
Tal

[hoborg]

EDIT:
I've nailed it down to a change in rev. 3160,
when detecting that the stream is DTS-HD, we had erroneously set the sample rate to 96000 instead of 48000 (DTS over SPDIF requires 48000),
fixed for rev. 3802, thanks for reporting.

Regards,
Tal

:thanks:

[cmbe]

EDIT:
I've nailed it down to a change in rev. 3160,
when detecting that the stream is DTS-HD, we had erroneously set the sample rate to 96000 instead of 48000 (DTS over SPDIF requires 48000),
fixed for rev. 3802, thanks for reporting.

Regards,
Tal

:thanks::thanks::thanks::thanks::thanks:

[madshi]

@developers, Michael Niedermayer (ffmpeg) is currently preparing for a new official ffmpeg release. He's merged some of the ffdshow changes into the ffmpeg tree. He's asking if there's anything missing that we'd like to have committed to ffmpeg. Here's your chance to checkout the latest ffmpeg sources, double check it with the ffdshow ffmpeg patches and try to get as many internal patches committed to official ffmpeg as possible.

[clsid]

I noticed that as well :)

I don't think we have any custom code left that is suitable for committing upstream. They are mostly are ffdshow specific tweaks/hacks.

[madshi]

Ok, that's fine.

BTW, he's also ffmpeg_mt fully merged in now. Do you plan to make use of it in a future ffdshow version? As far as I understand, in order to activate it you need to ask for SLICE based multi threading.

[nevcairiel]

ffdshow already uses ffmpeg-mt, but until now they used to ship two copys of ffmpeg, one "normal" and one build of the ffmpeg-mt fork. That can be disabled now, and possibly an option added instead if multithreading is supposed to be used.

[madshi]

Makes sense.

[Jeremy Duncan]

This clip: link, (http://www.mediafire.com/?3jvt2k2mu25fydo) was made by petitdragon. It is 23.976fps. in ffdshow avisynth when I type: info() the osd shows the clip as 23.97 fps, using mpc-hc 32 bit build newest build.



Could somebody dl the clip, it's not mpeg2 it's vc1 I think, or mkv. Run the code: info() in megui and then in ffdshow avisynth tab to see the frame rate info the osd shows is not the same?

[clsid]

Most -mt stuff has also been merged in Libav, but not yet the H.264 decoder because there are a few issues they want to fix first. Ronald and Alexander are working on that afaik.

@madshi
One thing Michael forgot is to add a configure option for the float stuff. Maybe he can also SIMDify the interleaving function that was added. The noscale one used by AC3 can maybe be dropped if scaling is adjusted (currently *1.0f). Dunno if that matters for quality, you know more about it.

[nevcairiel]

One thing Michael forgot is to add a configure option for the float stuff.
He didn't forget, he commented on the mailing list that its lacking configure support, so he knows.

Overall the patch he merged seems to be pretty low-quality in ffmpeg standards. I should do a patch to clean it up.

For the scaling thing, the noscale version could be dropped completly if you just don't touch the scale value in ac3dec at all.
The float-to-int conversion expects all values in the range of 32767 to -32768, so converting those to float 1.0 to -1.0 could always be done in that function, and you don't need to have different scale values based on the configure option.

Unless someone beats me to it, i will create a patch that adds the option to the configure script later.

[madshi]

Overall the patch he merged seems to be pretty low-quality in ffmpeg standards. I should do a patch to clean it up.
That would be welcome - thanks!

[Reimar]

Unless someone beats me to it, i will create a patch that adds the option to the configure script later.

I don't know if/when I will have time, but I actually intended to change it to make it possible to select at runtime.
I just can't imagine that that should be hard to do, and it would be a lot more useful than just a configure option.

[madshi]

I don't know if/when I will have time, but I actually intended to change it to make it possible to select at runtime.
I just can't imagine that that should be hard to do, and it would be a lot more useful than just a configure option.
That would be even better. BTW, if you look a bit through the older posts, a couple of days/weeks ago Justin already proposed a fully featured patch to allow runtime selection for AC3 decoding. The patch looked good to me, it just was never reviewed/applied, as far as I could see.

[nevcairiel]

Of course runtime would be preferable, no question.