ffdshow tryouts project: Discussion & Development

[sneaker_ger]

From what I've seen libass do, it has become very usable. I would trade MPC-HC's internal renderer for it any day. The former can behave really oddly sometimes and render differently than the standalone vsfilter, plus it's awfully slow and unstable.

[kirakami]

Maybe i'm wrong but i guess ass3/4 means ScriptType: v4.00+ which is nothing new and this is the default format of the aegisub. Just ffdshow's subtitles filter is broken compared to vsfilter

i got new vsfilter here but fail to install it
it didn't work with MPC-HC external filter :(
http://code.google.com/p/threaded-vsfilter/
does anyone know?

[sneaker_ger]

Be more specific as to what does not work.
To register it: start->(run->)"cmd" (On Vista/7 you may need to press ctrl+shift+enter to start the console with admin rights)
Then navigate to vsfilter.dll's folder and enter "regsvr32 vsfilter.dll". You may also need to add "DirectVobSub (Auto-loading version)" as a preferred external filter in MPC-HC.

[Keiyakusha]

i got new vsfilter here but fail to install it
it didn't work with MPC-HC external filter :(
http://code.google.com/p/threaded-vsfilter/
does anyone know?
Also that's not official version. Don't know if its authors are present on this forum, maybe try to ask them directly?
Also they say it won't work on XP, thats not the case for you?

[sneaker_ger]

Here's another vsfilter version, claiming to be faster than the threaded project:
http://www.cccp-project.net/forums/index.php?topic=5976.0

@kirakami
Are you on Win XP?

[kirakami]

yes i m on XP SP3

[sumawo13]

I believe FFDShow Tryouts is doing an improper YUV -> RGB conversion. I think it's using the wrong coefficients (BT601 instead of BT709). I have Rev3978 set to output RGB32.

I created two 8-bit YUV files of the 1080p Crowd Run sequence using sgi2yuv from the original 1080p 16-bit SGI files, one 4:2:0 and one 4:4:4. The 4:2:0 encode produces the correct colors, while the 4:4:4 video produces incorrect colors. Here are examples taken from the videos compared to one of the frames from the sequence (SGI file converted to PNG using ImageMagick).

Original
http://ompldr.org/tYWVoNw (http://ompldr.org/vYWVoNw)

4:2:0
http://ompldr.org/tYWVoOA (http://ompldr.org/vYWVoOA)

4:4:4
http://ompldr.org/tYWVoOQ (http://ompldr.org/vYWVoOQ)

Both were encoded with x264 and used the same encoding parameters except for --input-csp and --output-csp which were set to i420 and i444 respectively.

[Thunderbolt8]

Here's another vsfilter version, claiming to be faster than the threaded project:
http://www.cccp-project.net/forums/index.php?topic=5976.0also: http://xhmikosr.1f0.de/index.php?folder=bXBjLWhjL21wYy1oY19hcHBzL3ZzZmlsdGVy

[kirakami]

also: http://xhmikosr.1f0.de/index.php?folder=bXBjLWhjL21wYy1oY19hcHBzL3ZzZmlsdGVy

its the same original ver.
http://www.videohelp.com/tools/VSFilter_DirectVobSub

[sneaker_ger]

yes i m on XP SP3

In the thread I've linked someone says that the threaded version does not work on XP. Try the one linked in that thread instead. (Best to just read the whole thread, seems to contain some valuable info.)

[kirakami]

In the thread I've linked someone says that the threaded version does not work on XP. Try the one linked in that thread instead. (Best to just read the whole thread, seems to contain some valuable info.)

this vsfilter mod worked on XP
http://code.google.com/p/xy-vsfilter/
but still don't know whether it supports ass sub ver.3/4/+

[mandarinka]

I believe FFDShow Tryouts is doing an improper YUV -> RGB conversion. I think it's using the wrong coefficients (BT601 instead of BT709). I have Rev3978 set to output RGB32.

I created two 8-bit YUV files of the 1080p Crowd Run sequence using sgi2yuv from the original 1080p 16-bit SGI files, one 4:2:0 and one 4:4:4. The 4:2:0 encode produces the correct colors, while the 4:4:4 video produces incorrect colors. Here are examples taken from the videos compared to one of the frames from the sequence (SGI file converted to PNG using ImageMagick).

Original
http://ompldr.org/tYWVoNw (http://ompldr.org/vYWVoNw)

4:2:0
http://ompldr.org/tYWVoOA (http://ompldr.org/vYWVoOA)

4:4:4
http://ompldr.org/tYWVoOQ (http://ompldr.org/vYWVoOQ)

Both were encoded with x264 and used the same encoding parameters except for --input-csp and --output-csp which were set to i420 and i444 respectively.

This might be because with yv12, ffdshow handles the stuff itself (yv12 is passed through, and if rgb is to be outputed, it will do so using bt601 for SD and bt709 for HD, or based on stream's info iirc).

With 10bit video, swscale needs to be used, which could lead to improper matrix being used (might not matter for yv12 ouptut, but it breaks rgb). Maybe this is the case with 4:4:4 too? Are wrong coeficients being used when ffdshow outputs yv12 (or maybe other yuv formats), or just when outputing rgb?

[nevcairiel]

ffdshows RGB HQ converter only supports 4:2:0 and 4:2:2 input, 4:4:4 is probably going through some other code path.

[Dogway]

I used to use ffdshow_rev3972_20110818_xvidvideo-ru_x86-MSVC2010.exe but Im having some problems I think come from encoding FFV1 so I updated to ffdshow_rev3982_20110915_clsid.exe. But now I can't select ffmpeg-mt in h.264 decoding. Is this OK?

[nevcairiel]

ffmpeg-mt was integrated into the normal decoder quite a while ago. Its all good.

[Dogway]

It should perform better than libav right? I don't know why it doesn't show in the list to select, probably I need to update MPC-HC?
Maybe Id need to update to ffdshow_rev3972_20110818_xvidvideo-ru_x86-MSVC2010.exe instead of clsid builds?. I had a mistake in my previous post, I had before: ffdshow_rev3861_20110527_xvidvideo-ru_x86-msvc2010.zip. But this webpage stopped with its builds one month ago.

[Atak_Snajpera]

FFmpeg-mt is in libavcodec

[Dogway]

ah ok!, sorry for the confusion. I just wondered because before it was selecteable. thanks

[kirakami]

real audio cook codec is still buggy
audio becomes shuttery after playing for a while :(

[kirakami]

ffdshow's real audio cook codec also gets buggy as soon as i seek
using MPC-HC. it doesn't have internal cook codec?

[clsid]

LAV Audio works better for COOK

[Midzuki]

@ kirakami: why not use a "manual install" of the RealAlternative files, instead of insisting on ffdshow? :confused:

http://forum.doom9.org/showthread.php?p=1298640#post1298640 ... explains it all. :)

[Zerofool]

Hello,
I'd like to report an issue I'm having, I don't know if it's a bug, or lack of functionality, but anyways, here it is.

I was experimenting with the 4:4:4 support in x264, and I encoded some gameplay footage. It's all in full range (and so, I used the --fullrange parameter while encoding), but obviously ffdshow doesn't care about that (allways works for 4:2:0 fullrange videos), and treats the video as limited range, and, therefore, it looks darker than it should. I tried to force ffdshow to recognize the clip as full range (Overlay->Output-> and I only enable RGB32 here; Overlay->Output->RGB conversion-> Input levels: Full Range, Output levels: Computer Monitor, also the "HQ YV12 to RGB conversion" is enabled, although that shouldn't matter in this case, I guess), but there's no change at all, no matter what I set in these fields (unlike with 4:2:0 footage, where the difference is clearly visible). So would I be wrong to presume that these settings only apply to 4:2:0->RGB conversion, and not 4:4:4->RGB ? If that's correct, it would be nice if such a feature is implemented (and probably also native YV24 output support).

The only way I was able to correctly play the file was through madVR's internal decoder (and rendered with madVR of course :)).

Details:
ffdshow version: rev3984_20110922_clsid (32bit)
Player: MPC-HC (BE mod) v1.5.3.3734 (32bit)
renderer: I tested almost all of them - no difference; madVR is v0.74
OS: Windows 7 Ultimate x64
Video Card: ATI (AMD) Radeon HD5870 (non-reference PCB, XFX), currently with Catalyst 11.9 RC, but two days ago I had 11.8 WHQL - the same thing.
Display: 22" Mitsubishi Diamond Pro 2070SB (CRT)

Thanks and keep it up guys!
Cheers

[clsid]

The HQ RGB converter is not used in that case. The low quality swscale one is used. Any problems with that will NOT be fixed unless someone volunteers to do it.

[Zerofool]

I see. Thank you for clearing that up.

Best regards

[magic144]

just trying "The Killing" BD on my PC - seems like ffdshow Audio is choking on the LPCM 24-bit mono (48000Hz, uncompressed) soundtrack for some reason - plays back ~1/2 speed???
if I switch over to LAV Audio, all is A-OK
no other problems with ffdshow Audio to date...
...am using latest r3984, but reverted back as far as 3919 but the same issue persists

I'm using ZoomPlayer 8 MAX (latest release candidate).
Win7 32-bit, Analogue output via Speakers, DELL XPS420

[kirakami]

i compared xy-vsfilter http://code.google.com/p/xy-vsfilter with ffdshow 3984 sub render on MPC-HC
SSA sub plays fine but fonts size, bold broken vsfilter works fine check here http://www.mediafire.com/?icoztozy2nh
ASS subs some 3d effects seems to be broken try playing Dragon Crisis BD Opening once with ffdshow sub render then with xy-vsfilter & you will know.
http://www.mediafire.com/?0m7j6u2butp4cq5
i couldn't post screenshots 'cuz print screen doesn't work with video.

[kirakami]

Dragon Crisis BD Opening
http://www.mediafire.com/?0m7j6u2butp4cq5

[kirakami]

ffdshow 3984 weirdly fails to decode this H.264 codec
http://www.mediafire.com/file/zmmgky2gnny/Gen-ttt
potplayer plays it fine

[TheShadowRunner]

LAV Audio works better for COOK

Hmm in LAV Audio dec 0.37, I don't see COOK anywhere in the format list, nor does the decoder want to accept subtype {4B4F4F43-0000-0010-8000-00AA00389B71}

Strangely, ffdshow doesn't want to accept this subtype either!?

When it comes to COOK, only thing I found working is the Gabest splitter + cook.dll

Is COOK supposed to be handled by current ffdshow/LAV?

[roytam1]

Hmm in LAV Audio dec 0.37, I don't see COOK anywhere in the format list, nor does the decoder want to accept subtype {4B4F4F43-0000-0010-8000-00AA00389B71}

Strangely, ffdshow doesn't want to accept this subtype either!?

When it comes to COOK, only thing I found working is the Gabest splitter + cook.dll

Is COOK supposed to be handled by current ffdshow/LAV?

LAV Splitter will split COOK audio with 4cc "cook"(YES lowercased) which LAV Audio will accept.

[TheShadowRunner]

roytam, I see.
I guess ffdshow only accepts the lowercased "cook" 4cc as well. Strange (and annoying) that these decoders won't accept COOK uppercase as presented by Gabest's FLV Splitter.

[ikarad]

When I see a hi10p video I have these information with ffdshow+madvr+ mpc-hc

http://img11.hostingpics.net/pics/575998osdincorrect.png (http://www.hostingpics.net/viewer.php?id=575998osdincorrect.png)

In this site http://haruhichan.com/wpblog/?p=205#confirm, it said it's incorrect


The correct osd is here
http://img11.hostingpics.net/pics/184932madvrosdcorrect.png (http://www.hostingpics.net/viewer.php?id=184932madvrosdcorrect.png)

is-it normal or not?

If I use h264 internal decoder of madvr instead of ffdshow I have
http://img11.hostingpics.net/pics/809221Sanstitre.jpg (http://www.hostingpics.net/viewer.php?id=809221Sanstitre.jpg)

Anybody could explain me the differences and the good method to see hi10p movie with the best quality?

[nevcairiel]

ffdshow does not support 10bit output, it'll always be converted to 8-bit.

[TheRyuu]

With the addition of this (http://git.libav.org/?p=libav.git;a=commit;h=27237d524e56210992b18486924894bb4f3fdbb8) to libav external pthreads should no longer be required and everything should work all the same.

Only really matters if you want to reduce a dependency, otherwise it really makes no difference.

[seo125]

Am I wrong or 'Cubic Blending' is broken? I see a lot of artifact in build 1090. I'm using this post as a base of the best deinterlacing method. Any suggestion?

__________________
´ (http://www.travelling2peru.com)´´´My Profile´´´ (http://forum.doom9.org/member.php?u=202175)´ (http://www.travelling2peru.com)

[kirakami]

ffdshow rev. 3996 ???
http://www.afterdawn.com/software/audio_video/codecs/ffdshow.cfm
its not uploaded here
www.videohelp.com/tools/ffdshow/version-history

[roytam1]

ffdshow rev. 3996 ???
http://www.afterdawn.com/software/audio_video/codecs/ffdshow.cfm
its not uploaded here
www.videohelp.com/tools/ffdshow/version-history

http://ffdshow-tryout.svn.sourceforge.net/viewvc/ffdshow-tryout/

[Moti172]

Hi

In order to decode dts I'm using lav-audio+dtsdecoderdll |(mpc-hc), and ffdshow to process it (downmix...).
The audio decoded out put is 24bit while ffdshow process it with 32-bit integer floating point.
Does it has any impact on the sound quality?

Thanks

[LigH]

At least it would not make it worse: Single Precision IEEE float has an effective mantissa of 24 bit too, but that is the precision over the whole volume range, not only for the loudest parts.

[Thunderbolt8]

when activating OSD and frame type, then it looks as if the video consits only of I frames. there is no other change of frame types, every video I open just has I frames written all over (ffdshow for raw video & LAVsplitter & coreavc & WMVdecoderDMO)

[zmaster]

I saw this message:For those who are wondering why Hi10P isn't working with the x64 build. It has been disabled because it crashed.

The crash happens in the golomb code used in the SPS parser. It isn't the first time that this code has been causing problems in x64 builds. It really needs to be replaced with something better.

I tried replacing it with the CByteParser class from LAV Filters, but a simple function call replace didn't work properly (assertion failures in both x86 and x64 build). Need to be analyzed deeper. If anyone else want to give it a go, please do. but decided to ask because the inability to decode Hi10p becoming a growing problem and would like to know takes place whether the development in this direction.

[clsid]

when activating OSD and frame type, then it looks as if the video consits only of I frames. there is no other change of frame types, every video I open just has I frames written all over (ffdshow for raw video & LAVsplitter & coreavc & WMVdecoderDMO)Only the decoder knows the frame type. You are not using ffdshow for decoding.

[clsid]

I saw this message: but decided to ask because the inability to decode Hi10p becoming a growing problem and would like to know takes place whether the development in this direction.LAV Video can decode it. ffdshow development is pretty much frozen at this moment.

[zmaster]

clsid,
LAV Video can decode it...Yes, but I'm a fan of ffdshow ;)

Thanks for reply!

[haruhiko_yamagata]

Hi, everybody. Sorry for my long absence and thank you for the good maintenance work.

I've been playing with the subtitles for a month. By importing a lot of code from MPC-HC,


Implement rotation (\frx,\fry,\frz,\org).
Implement \clip.
Implement \h.
Implement \r<style>
Implement \blur<strength>
Implement \be<strength>
Implement polygon drawings (\p)

Improved rendering (boundary of the body and the outlines)
Improved gradation of the outlines.
Improved, rotatable opaque box.

Subpixel accurate positioning.

Some compatibility fixes.
a few regression fixes

Other bug fix

ffdshow color space converter:YV12->RGB32:If the output horizontal size is not a multiple of 4, the heap gets broken.

TODO

Implement \t
Implement \clip polygon bitmap mask version
Improve YV12 rendering

I've created a branch in the svn. A test build. (http://ffdshow-tryout.sourceforge.net/doom9/ffdshow_rev3998_20111025_branch.exe)
Please give it a try, if you are interested in ffdshow's subtitles filter.

[fastplayer]

Welcome back, Haruhiko! :)

[haruhiko_yamagata]

Hello! Nice to see you are still here.

[_xxl]

@haruhiko_yamagata

Nice to see you again !

[_xxl]

Do you have new idea's for ffdshow?:)