ffdshow tryouts project: Discussion & Development

[wyrd]

Does anyone have a Snow sample file for me? (encoded with a recent version of mencoder)

sample here (http://tirnanog.fate.jp/tmp/sample/snow/)

Encode by MEncoder 1.0pre8-3.4.2 (2006/06/12)

Decode results:
vlc0.8.6a : ok
mplayer pre8,rc1,20061220(celtic_druid) : ok
mplayer 20070126 : ng
ffmpeg(ffplay)rev7333,7375 : ok
ffmpeg(ffplay)rev7430,7579 : ng
latest ffdshow : ng

related post by akupenguin
http://forum.doom9.org/showthread.php?p=918839#post918839

Best regards

[Jeremy Duncan]

>>Some of them don't work
I can't reproduce this. Please upload a sample.

bypass: Doesn't work
Linear interpolation: Works, but does a very poor job, showing comb artifacts, and a rippling line artifact.
Linear blending: Works, but removes a lot of detail
cubic interpolation: Works, but same problem with jagged comb artifacts, and rippling line artifact.
cubic blending: Doesn't work
median: Works, but suffers from a rippling ling artifact
tomsmocomp: Works, Search effort 1, vertical filter on, but suffers from rippling ling artifact
dgbob: Works if Artifact protection is checked and the second and third modes are used
framerate doubler: doesn't work
ffmpeg deinterlacer: Works but has jagged comb artifacts and rippling line artifact.
5 tap lowpass: works
kernal deinterlacer: Doesn't work
kernal bob: doesn't work

In FFDshow. Deinterlacing has to be before resize & aspect to function.

Rippling Line (http://www.slibe.com/fullimage/0428471a-Rippling_Line.png)

No Deinterlacing (http://www.slibe.com/fullimage/83f3de8c-No_deinterlacing.png)

This is using rev 814, Jan 22 2007 17:17:32 msvc 2003, x86 unicode.

The image is a pattern on the hqv test disk, and the whole dvd is interlaced.

[Yong]

sample here (http://tirnanog.fate.jp/tmp/sample/snow/)

Encode by MEncoder 1.0pre8-3.4.2 (2006/06/12)

Decode results:
vlc0.8.6a : ok
mplayer pre8,rc1,20061220(celtic_druid) : ok
mplayer 20070126 : ng
ffmpeg(ffplay)rev7333,7375 : ok
ffmpeg(ffplay)rev7430,7579 : ng
latest ffdshow : ng

related post by akupenguin
http://forum.doom9.org/showthread.php?p=918839#post918839

Best regards

your sample crashing here, mplayer svn-r22069.

snow encoded by ffdshow(clsid) r853 is both play fine in mplayer svn-r22069 and ffdshow too;)

[foxyshadis]

Jeremy, are you sure about the artifacts being ffdshow's fault? I'd try the avisynth versions, because I think the artifacts (stray combing, rippling, etc) are simply common artifacts of certain classes of deinterlacing algorithms. In fact, the HQV disc is specifically made to show those shortcomings in glaring detail. That's why Didée is working on MCBob, one deinterlacer for all things. ;)

Can you be more specific on "doesn't work"? Errors, major arifacts (like discoloration or blocking), simply does nothing at all?

[SeeMoreDigital]

NDIG is only missing in the notes (fixed in SVN). Playback should be working.It does not seem to be: -

http://img222.imageshack.us/img222/7229/ndigdj9.png

Admittedly, except for testing, it going to be "very" rare for users to de-mux NeroDigtal (Part-2) video streams out of the .MP4 container to the .AVI container (and correct the 4CC code).... But, that's me ;)


Cheers

[wyrd]

@yong
thanks for reply.
I've confirmed it. (It works fine with both mplayer and r853)
test sample uploaded, but i do not tuning, so it's bad quality:D
also i could not encode with recent mencoder(rc1 or later)...
off topic:can anyone do it?

//EDIT:
oops, I'm embarrassed:P.
I could encode in rc1 or later.(sorry,my setup was too old...)
This file (http://tirnanog.fate.jp/tmp/sample/snow/sakuratan%5bsnow(mencoder20070126)+mp3%5d.mkv) works fine in r853.
(with mencoder svn-r22069(20070126))
Best Regards

[Leak]

That's why Didée is working on MCBob, one deinterlacer for all things. ;)
Are you sure that's going to be feasible to use in realtime during playback? :D

On another note - would it be feasible to integrate TIVTC into ffdshow, as an IVTC step with optional deinterlacing for still-combed frames? (i.e. as another sub-page of the deinterlacing filter)

That sure would give a smoother result than simply deinterlacing telecined sources - but I'm not sure if ffdshow can handle reducing the number of frames per second and looking/decoding ahead a few images. Any ideas?

And while I'm asking - has anyone ever had a chance to look at DVD decoding? Or rather, what exactly is making it non-functional currently?

(Incidentally, those two things are the only two things I'm still missing to get my anime DVDs to look really smooth on my projector... :p)

[HeadBangeR77]

And while I'm asking - has anyone ever had a chance to look at DVD decoding? Or rather, what exactly is making it non-functional currently?
Have you tried? ;) Then plz do ;) For me it works perfectly, apart from seeking, but that's a splitter problem (MPC) I think. There are artefacts (large 8x8 squares) while seeking, but the playback itself is already ok. :)

Have a look at some screencaps here:
http://forum.doom9.org/showpost.php?p=945832&postcount=37

cheers,
HDBR77

PS. libmpeg2 of course.

[foxyshadis]

Avisynth framerate change works fine as long as you don't do anything with VFR. (Trivia: It interpolates the new timecodes between the old ones, using the difference between input and output framerates, since DShow has no real framerate.) Of course, you wouldn't normally want to IVTC a VFR file anyway, so it's sort of academic.

The only problem with IVTC is that decimation takes most of the processing power at the beginning of the cycle, so there are sharp periodic bursts of CPU usage. A good way to make it look jerky after all the effort not to. ;_; Queuing can help, of course. The weave/deinterlace usage is pretty regular though.

TIVTC could probably be integrated, but all the framerate stuff would have to be modified (and big chunks would be useless for plain viewing); since tritical's always coming out with new fixes it would mean more maintenance effort to stay on top of.

I think the only thing wrong with DVD decoding is 2-thread decoding at the moment, but I might be wrong.

[SeeMoreDigital]

Thanks clsid,

MPEG-4 with NDIG 4CC code works a treat now, with ffdshow_rev858_20070130_clsid.exe


Cheers

[Jeremy Duncan]

bypass: Does not Deinterlace.
Linear interpolation: Works but has combing artifact, and Rippling line Effect.
Linear blending: Works but removes too much detail.
cubic interpolation: Works but has combing artifact, and Rippling line Effect.
cubic blending: Works
median: Works but has Rippling line Effect.
tomsmocomp: Works but has Rippling line Effect. Latest Avisynth version does not have rippling line artifact.
dgbob: Only first mode works. Has combing artifact, and Rippling line Effect.
framerate doubler: Does not Deinterlace.
ffmpeg deinterlacer: Works but has slight combing artifact, and Rippling line Effect.
5 tap lowpass: Works.
kernal deinterlacer: Does not Deinterlace.
kernal bob: Does not Deinterlace.

Here's a picture of Tomsmocomp version 0.0.1.8 used in FFDshow Avisynth, tomsmocomp(1,5,1).
No Rippling (http://www.slibe.com/fullimage/af9352f3-Tomsmocomp.png)

[MacAddict]

libmpeg2 seems to decode DVD's just fine for me but the problem comes in when seeking. I get random crashes/lockups on DVD's when seeking or navigating the DVD menu.

[Leak]

libmpeg2 seems to decode DVD's just fine for me but the problem comes in when seeking. I get random crashes/lockups on DVD's when seeking or navigating the DVD menu.
That's what I meant, basically. Like the first US volume of Ergo Proxy crashing MPC as soon as you click on "Play all"... :(

[lexor]

there is something very wrong with Audio Decoder.

If you open it up (the gui I mean) and enable only the Mixer, set Speaker Configuration to 2/0 (this far everything works fine), but now enable the LFE checkbox and your system will reboot once you try to play back a movie. Setting Speakers to 2/1 fixes it. But still even if 2/0 is a wrong setting, the computer shouldn't reboot. (gray out LFE checkbox in unsupported settings maybe?)

ffdshow 844, xp sp2 + haali splitter (xvid + mp3 input)

[wozio]

Hi,

Is it possible to include option to disable adjusting color space in ffdshow? I would like to have it in native scale as it was from decoder without any recalculations. It shouldn't be much work.

Regards
Piotr

[haruhiko_yamagata]

Revision 864
H.264 MBAFF hardware deinterlacing support.
It was a request from stream47.

Samples originally (http://ffdshow-tryout.sourceforge.net/phpBB2/viewtopic.php?t=144&start=41) posted by stream47 (In Japanese).

English translation
ASP top field first
AVI (http://asnxload.dip.jp/files/ppfx00340.avi)
MP4 (http://asnxload.dip.jp/files/ppfx00341.mp4)

ASP bottom field first
AVI (http://asnxload.dip.jp/files/ppfx00346.avi)
MP4 (http://asnxload.dip.jp/files/ppfx00347.mp4)

AVC
MP4 (http://asnxload.dip.jp/files/ppfx00344.mp4)
MKV (http://asnxload.dip.jp/files/ppfx00345.mkv)

MPEG4-ASP hw-deinterlacing is supported only by libavcodec, not xvid.

[fastplayer]

If you open it up (the gui I mean) and enable only the Mixer, set Speaker Configuration to 2/0 (this far everything works fine), but now enable the LFE checkbox and your system will reboot once you try to play back a movie.
Doesn't happen for me. No matter how many times I check/uncheck the LFE box. Maybe the problem is your sound card driver?

[foxyshadis]

Doesn't happen for me. No matter how many times I check/uncheck the LFE box. Maybe the problem is your sound card driver?

It couldn't really be anything but the soundcard driver, since an entirely user-mode thing like ffdshow could never cause a reboot.

lexor, either disable reboot on bug check, or use Event Viewer (eventvwr.msc) to determine the probable cause of the fault. It'll give you a something.sys, you might have to google around to find out just which driver that is.

Revision 864
H.264 MBAFF hardware deinterlacing support.

Awesome, getting AVIVO and Purevideo to work should be much simpler now.

[SeeMoreDigital]

MPEG4-ASP hw-deinterlacing is supported only by libavcodec, not xvid.Finally.... This is excellent news!

EDIT: I've only been banging on about the need to provide proper support for decoding interlaced MPEG-4 Part-2 sources for years (plain old "Bob" would have done in the meantime). I wonder if this move has been promted by DivX's recent announcement regarding their 6.5 direct-show decoder offering support too?!

[fastplayer]

Finally.... This is excellent news!
It gets even better: the "SVQ3 freeze at the end of file"-bug has been fixed, too.
:thanks:

[Seb.26]

Hi world ...

First : thanks to the new dev team for the work !

FFDShow reborn is really a great news !!!

Second, where can I find a list of each FFDShow's filter whith the color space allowed by the filter please ? ...

( This to avoid the YV12 conversion done by some filters ... )

Thanks !
Seb

[_xxl]

I can't play .evo files on my computer properly.
Tested with AMD x2 + 2gb DDR2 + 7900 + ffdshow + MPC.
No Open-Source solution can handle it.:(
Maybe someday libavcodec h264 decoder is going to be multithreaded.
ftp://mplayerhq.hu/MPlayer/samples/evob/Departed%20Trailer.EVO

[fastplayer]

ftp://mplayerhq.hu/MPlayer/samples/evob/Departed%20Trailer.EVO
I hear no sound. GSpot tells me it's AC3... Is a special splitter/decoder needed?

[clsid]

Check out this thread for info about .evo:
http://forum.doom9.org/showthread.php?t=120652

[SeeMoreDigital]

I hear no sound. GSpot tells me it's AC3... Is a special splitter/decoder needed?Yes... HD-DVD and Blu-ray use E-AC3 (http://en.wikipedia.org/wiki/E-AC3)....

[fastplayer]

Thanks for the info guys. Certainly a nice read for a rainy afternoon :D

[Scoty]

which settings needet for best video quality ? i use the latest Build with MPC.

[foxyshadis]

Scoty, see this thread (http://forum.doom9.org/showthread.php?t=115727) for starters. There is no one setting that will magically be best for everything, though, and it's pretty cpu intensive.

[fastplayer]

Would it be possible to make the priority over which type of subtitle gets loaded by default configurable? For example, like a drop-down box where you can choose between the subtypes like SRT, VOB, SUB etc.

[SeeMoreDigital]

Are there any "Revision 864" builds (with interlaced support) available yet?

The most up-to date builds on SourceForge stop at 30 Jan 07.


Cheers

[clsid]

I'll upload latest build asap.

VP5/VP6 decoding seems to be broken in current SVN revision. Picture freezes at the beginning. Sample file (http://www.zshare.net/video/vp62-avi.html).

Edit: revision 865 broke it

[cc979]

got error building rev866

make -C ffmpeg
make[1]: Entering directory `/home/user/svn/ffdshow-tryout/trunk/src/ffmpeg'
make[1]: *** No rule to make target `libavcodec/lzo.h', needed by `libavcodec/cscd.o'. Stop.
make[1]: Leaving directory `/home/user/svn/ffdshow-tryout/trunk/src/ffmpeg'
make: *** [FFMPEG] Error 2

[_xxl]

Can you compile latest GCC for MinGW?

[cc979]

Can you compile latest GCC for MinGW?

i tried other day, still not compile ffdshow with gcc-4.2 or gcc-4.3 still patches needed i think

[_xxl]

Maybe GCC prerelease-4.1.2-20070128?

[clsid]

@cc979, make clean and remove cscd.d

[JarrettH]

Revision 867

-added libavcodec vc-1

Will this open my DRMd WMV-HD files?:sly:

[Scoty]

Revision 867

-added libavcodec vc-1

Will this open my DRMd WMV-HD files?:sly:

link ?

[haruhiko_yamagata]

Would it be possible to make the priority over which type of subtitle gets loaded by default configurable? For example, like a drop-down box where you can choose between the subtypes like SRT, VOB, SUB etc.
A case ffdshow is inputing multiple formats of subtitles?
What kind of case do you assume?

[fastplayer]

A case ffdshow is inputing multiple formats of subtitles?
What kind of case do you assume?
For example: Consider the following 4 files in one directory:

video.avi
video.idx
video.sub
video.srt

Now when you start playing the AVI file, ffdshow opens the Vobsub subtitle (IDX/SUB) by default. Is it possible to make this default behavior configurable? Like making SRT files to be opened by default.

[haruhiko_yamagata]

For example: Consider the following 4 files in one directory:

video.avi
video.idx
video.sub
video.srt

Now when you start playing the AVI file, ffdshow opens the Vobsub subtitle (IDX/SUB) by default. Is it possible to make this default behavior configurable? Like making SRT files to be opened by default.
OK, it should not be too difficult.

// EDIT
added at rev 870.
Ideally the extensions in the dialog should be dragable so that user can drag and re-order.
Considering the users who use this feature would be experts, the simplified dialog may be enough.

[fastplayer]

OK, it should not be too difficult.

// EDIT
added at rev 870.
Ideally the extensions in the dialog should be dragable so that user can drag and re-order.

(:thanks:)^10
Wow, this feature required more changes than I expected... :o
I hope this feature is not just useful for me but for others as well!
Considering the users who use this feature would be experts, the simplified dialog may be enough.
I would've been happy even if this feature was exposed via the registry :D

[Leak]

Avisynth framerate change works fine as long as you don't do anything with VFR. (Trivia: It interpolates the new timecodes between the old ones, using the difference between input and output framerates, since DShow has no real framerate.)
Yeah, just changing the framerate works, but all hell (well, not really... ;)) breaks loose once you try to actually return less frames than the original source has:

Info()
SelectEvery(30,0)
AssumeFPS(1)should really give you a 1 FPS slideshow but plays normally - except for the frame numbers printed by Info which increment by 30... :(

Obviously AviSynth's seeking ahead does nothing - shouldn't the ffdshow source decode ahead 30 frames in that case? Referencing past frames (within limits) is handled by AviSynth's cache usually, but of course it can't look into the future... :D

And as far TIVTC is concerned, TFM works of course, but TDecimate doesn't, for similar reasons.

As for TIVTC producing load spikes for each cycle, like you said earlier - I rather doubt it. After all, it bases it's decisions on metrics calculated by comparing each frame with it's previous and next frame, for each frame in the cycle, which can be done as soon as each frame is coming in, which distributes the load nicely. At least I didn't get any weird CPU spikes when decoding a ripped VOB file and using the AviSynth filter with just TFM() followed by TDecimate() - CPU usage stayed somewhere between 20 and 30 % according to Task Manager.

When figuring out which frame to drop, it just looks at the metrics that have been calculated for the past frame, which shouldn't be noticeable performance-wise. Of course, what you do need to do is decoding ahead 5 (or 10) frames before returning the first frame, then always buffering those frames and returning frames from that buffer - but that'd only give you a slight hiccup/load spike at the very beginning or after seeking (unless you ignore seeking and let TDecimate figure out what to do - shouldn't be treated much different than any other scene change), which shouldn't be too much of a problem.

So here's my question - is it possible to request some frames ahead via DirectShow when ffdshow is not doing the decoding (dunno how the graph thing works exactly)? I guess it should at least be possible as soon as ffdshow itself is used as the decoder - but it's not done currently for AviSynth, as far as I can see...

EDIT: Yeah, looking at the source for Tffdshow_source, the passed in frame number (n) isn't used, ever.

np: Mira Calix - Belonging (No Longer Mix) (Eyes Set Against The Sun)

[fastplayer]

about rev870:
When setting the vertical subtitle position from 90% (default) to 95%, then only the first row of subtitles with 2+ lines is shown.
In case of vobsubs, nothing is displayed.
The only way to bypass this, is to reduce font size resp. the scale factor when dealing with vobsubs but this makes the subtitles harder to read... :o

[JarrettH]

I'm just waiting on a new ICL from clsid :D

Will these VC-1 codecs be able to playback DRMd WMV-HD stuff?

[Kado]

I've been waiting for the wmv9 codec to be added as a decoding alternative to libavcodec. Thanks.
However after doing some tests I have registered the following cpu usage (dual-core cpu):
Internal wmv9 decoder => 10 to 25% cpu usage (no artifacts).
ffdshow libavcodec => 25 to 45% cpu usage (some artifacts like trails or something)
ffdshow wmv9 => 20 to 35% cpu usage (no artifacts).
Internal wmv9 + ffdshow raw processing => 20 to 35% cpu usage (no artifacts).

Is there any special reason to ffdshow use more cpu?

Details:
Video: Windows Media Video 9 1280x720 29.97fps 3.8Mbit (VBR)
Media Player Classic v6.4.9.0 v3.2+ (same settings for all codecs)
CPU: Intel Pentium D 930@3.6ghz
ffdshow build 875 with no extras like post-processing active.

Kado

[haruhiko_yamagata]

If this is not the right place to bring forth bugs, beg pardon. But in the mean time, there's a tiny one that's been there as long as I remember

http://luuranko.freeshell.org/ffdshow_subs.jpg

The lines get cut off by a few pixels on right side if the font is angled very much. The font in the picture is in italic style.
Thank you for report and sory to be late.
fixed at rev 883.
Some other bugs related to italic characters are fixed at the same time.

[_xxl]

http://img165.imagevenue.com/loc72/th_79303_Capture_122_72lo.jpg (http://img165.imagevenue.com/img.php?image=79303_Capture_122_72lo.jpg)
Anybody knows what's wrong with mingw gcc-build-4.1.2-121479?

[Yong]

http://img165.imagevenue.com/loc72/th_79303_Capture_122_72lo.jpg (http://img165.imagevenue.com/img.php?image=79303_Capture_122_72lo.jpg)
Anybody knows what's wrong with mingw gcc-build-4.1.2-121479?

gcc4x cant find the standard include dir.
i have the same problem too, when the ffdshow source code was in my drive D:, mingw gcc was in C:,
then i moved the ffdshow source code back to drive C: and the problem is gone :p

[cweb]

gcc4x cant find the standard include dir.
i have the same problem too, when the ffdshow source code was in my drive D:, mingw gcc was in C:,
then i moved the ffdshow source code back to drive C: and the problem is gone :p

gcc4 isn't stable under mingw, that's why there are no builds using gcc4 in mingw.