ffdshow development

[oddball]

Ah. I didn't realise it had any effect on resize. I always wondered what that option was for thanks ;)

Still would love a better less CPU intensive way to do black borders at some point in the future ;)

Oh and the crop and zoom tool never seems to work the way I think it's gonna work. Lets say for instance you have a movie where some bozo encoded the black boarders. For widescreen TV output it's a pain the butt as it looks wrong on the TV. In DiVXG400 it has an option 'Cut sides to horizontally fill at least 100% of the screen'. Which works great without pulling any CPU load. It crops the black bars automagically top and bottom and spits it out on the overlay correctly.

In ffdshow if you crop it seems to pull the image up and down but not crop the bars in the way I expect. It's very fiddly indeed.

[Chibi Jasmin]

Originally posted by oddball
Still would love a better less CPU intensive way to do black borders at some point in the future ;)

Like DivXG400 does it :-) My words...

[kastro68]

I noticed that the postprocessing for the 24-4-2003 ffdshow is not as strong as that of the 3-1-2003 ffdshow.

Using the 24-4-2003 ffdshow with postprocessing all the way up and with a processing strenght of 512 I hardly notice any PP taking place unless Nic's PP is ticked.

I'm not complaining, I'm just wondering if it may be a bug, or if PP has been deliberately watered down.

Have a nice day.

[simon.mittag]

Originally posted by kilg0r3
How is it supposed to work?


Ok, just tried it. You can just enter the filters you want, and levae the clip position unspecified).

This could be a nice weapon agains the black locking issue, without having to modify the the stream during the encode. It was suspected that the phenomenon is caused by luma values which are invalid for TV display. If this is true, a simple 'limiter()' inserted in ffdshow should do the trick.

Very nice! THANK YOU MILAN!


avisynth in ffdshow works fine for built in avisynth commands for me, however it behaves erratic when I try to load a plugin.

scripts that otherwise work as avs files fail with "unexpected error"

I observed that once entering only the load plugin statement into the script, commencing and pausing playback to allow it to load, then entering a second line running the plugin command seems to work.

Are there some problems with dynamic binding of the dll at runtime?

Anybody else tried that so far? I had problems like that with AutoCrop and WarpSharp plugins for 2.5

[simon.mittag]

I can make the description a little more precise after further testing.

I actually CAN run the AutoCrop plugin when applying the filter to an already running movie clip, meaning entering the AutoCrop(1) statement into the ffdshow dialog at playback time and hitting apply produces a working crop.

What doesn't work is starting playback when that value is already configured in the ffdshow tab - it then throws exception when I hit play in zoomplayer. seems something goes wrong with the ffdshow avisynth source filter at startup time?

Milan, Glenn, any ideas? Since it runs afterwards it can't be that grave an issue?

regards,

Simon

[athos]

This might have something to do with the dynamic loading of dlls, as this was the cause of the libavcodec problems and the ffvfw trouble.

[simon.mittag]

Milan told me he is currently working on the avisynth plugin loader for ffdshow and I hope he will post us his test results when he's finished.

[int 21h]

Any chance we'll see the new Sorenson additions to ffmpeg in ffdshow?

[simon.mittag]

I am not yet sure when any possible next release for ffdshow will be, we'll have to wait for Milan.

[Blight]

VP31 decoding from ffmpeg could be useful too, ther TrueMotion one kinda sucks. Also, there's quirks in the huffyuv decoder which merit a look-over.

[rjamorim]

Originally posted by int 21h
Any chance we'll see the new Sorenson additions to ffmpeg in ffdshow?

Isn't Sorenson limited to the MOV container - that doesn't use dshow for playback (uses QT instead)?

Or is there something I don't know? :confused:

[Blight]

Parhaps someone should write an MOV parser... there is one for QuickTime v2 that comes with windows, but I doubt it applies anymore.

[LigH]

I always thought "a codec is a codec - no matter which kind of container format you wrap around". I remember early MOV <=> AVI converters which quite blindly exchanged the surrounding containers; so it was indeed possible to have a video format in an AVI where no VfW codec or DirectShow filter existed for.

From this point of view, it is no problem to have Sorenson video in an AVI. No matter how hard it could become to play it back... :D Probably, I wouldn't mind if Sorenson would create some software for Windows PCs, just as cross-platformed as 3ivx. But I doubt that this would definitely be my first choice then: On a 1800 MHz P4 it was almost impossible to play back the "Matrix Reloaded Final Trailer", created in Sorenson 3 Pro at 1000x540 pixels.

I transcoded it to MPEG2 to watch it on TV instead... And this did not work using the QuickTime 2 DirectShow import filter in AviSynth: I had to try the QuickTime VFAPI plugin and the VFAPI import snippets for AviSynth, and even had to pull down the frame rate by 25 (ever seen a movie at 600 fps?).

[gabest]

Originally posted by Blight
Parhaps someone should write an MOV parser... there is one for QuickTime v2 that comes with windows, but I doubt it applies anymore. Actually, the parser should be still fine for any mov. A few weeks ago I started to write a qt video decoder using the sdk, but the decompressor calls always failed for the incoming "svq3" data. With other encoding types I had success (mainly with "all-keyframe" types), but not with this sorenson codec.

[mf]

Originally posted by gabest
Actually, the parser should be still fine for any mov. A few weeks ago I started to write a qt video decoder using the sdk, but the decompressor calls always failed for the incoming "svq3" data. With other encoding types I had success (mainly with "all-keyframe" types), but not with this sorenson codec.
Well that's where a new ffdshow would come in handy :D.

[Blight]

I'm not sure that the QT parser from QTv2 is still working as Apple inserted a lot of other video stream support and may have restructured the format beyond backward compatibility, can't really tell for sure.

But if FFMPEG can decode the contnet, perhaps it's possible to do. As far as writing your own QT Decoder, I know that Sorenson have some sort of licensing deal with Apple that their codec won't be used outside of their own player. Perhaps that is why it doesn't decode for you, it detects that your program instance is not valid.

[raistlin2k]

Actually, you can use MPC for playing QT WITHOUT installing Quicktime player :)

And I added those QT - filters to Media Center 9, so you can play QT there as well without Quicktime installation!

Isn't this enough quicktime-support?

Raist

[HughMagoo]

I need a little help. I've searched around, but found no hints. I'm using the latest alpha.

1) when using the Resize function in conjunction with VMR9 and quick pans or scene changes, I am seeing lots of tearing, like the picture is broken in a line from left to right.

I've tried setting D3D vsync on to see if that has any effect, and I have also tried different Nvidia Dets, as well as Omega drivers. Nothing seems to work. my refresh rate is 60Hz. I don't know what has changed recently, but I am sure I wasn't seeing this tearing before.

2) when using Raw Video processing, the Info screen in the ffdshow panel says output 25.0fps... for every file with different fps. is this usual?

3) I'm not exactly sure about what colorspaces I should accept for Raw Video Processing, and what colorspaces I should output to, in the ffdshow codecs panel. gf4ti4600 --> YV12 forced for both?

thanks for any help.

[Stux]

the MS QT parser only support QT v1 and v2

since Apple is up to QT6, that's fairly out-of-date

QT3 introduced compressed movie header
QT4 introduced vbr audio tracks

etc

the 3ivx Media Splitter handles more modern (simple) mov files, but only for mpeg-4 variants (ie not sorenson)

[athos]

New build is up on SF. This time I skipped ICL7 and used the regular MSVC++6 compiler, as the builds since 2002-12-13 has caused Explorer to crash when building thumbnails for video files in thumbnail view. This one doesnt seem to do that, at least not yet. Please post your findings regarding stability and possible performance loss (or gain?).

[kastro68]

@athos
Thanks. I'll give it a try

@gabest
are you the same person that wrote vobsub and Media player classic?
Do you know where I can go to post suggestions and bugs for MPC?

I like doom9. I actually get to meet and chat to the people that develop and maintain the software that I use. It's a bit like meeting a celebrity :)

[athos]

Originally posted by kastro68
@gabest
are you the same person that wrote vobsub and Media player classic?
Do you know where I can go to post suggestions and bugs for MPC?

yes it is the same person, check out the MPC thread:
http://forum.doom9.org/showthread.php?s=&threadid=29798
Originally posted by kastro68
I like doom9. I actually get to meet and chat to the people that develop and maintain the software that I use. It's a bit like meeting a celebrity :)
indeed ;)

[kastro68]

Hi,

I just tried the latest ffdshow, and it works fine with WMP 6.4 but it crashes Media player classic. I just gave it a reboot.

I'll try reinstall former ffdshow now and see if MPC still crashes.

I just reinstalled ffdshow-20030424.exe and both wmp and MPC work fine with this. I uninstalled it and installed ffdshow-20030523.exe again and MPC crashes. WMP works fine though

Can anyone confirm same problem? If not, I'll post more details.

[gabest]

Looks fine to me. Running MPC v6450.

[kastro68]

Oh. I was running MPC 6.4.4.2

Sorry gabest, I just checked your site yesterday and it still had the old version up. You must've updated MPC very very very recently :)

@Gabest: Are you pulling my leg? :) I just came back from your site. I guess you didn't update your MPC files yet. Any chance I can get a copy of MPC 6.4.5.0 early?

[kastro68]

MPC 6.4.5.0 is up now

Thanks.

Feedback: I don't experience any problems with latest ffdshow and MPC6450

[Defiler]

Does the new (kinda) double-precision IDCT support SSE2 extensions?

[Defiler]

Also.. I (still) see negative numbers in the "CPU Usage" OSD sometimes.

[kastro68]

Hi,

I think there may be a possible bug with ffdshow-20030523 while 'Luma Smoother' feature of 'Blur & NR' it causes both MPC6450 and WMP to crash. I haven't found anything wrong with any of the other 'Blur&NR' features...just the luma smoother.

Thanks

Edit: Chroma smoother causes MPC and WmP to crash too.

[HarryM]

23052003 dont works on win98SE (for me).
24042003 works fine.

[athos]

Originally posted by HarryM
23052003 dont works on win98SE (for me).

What error messages do you get?

[HarryM]

Originally posted by athos
What error messages do you get?

Freezes without error messages. ZoomPlayer and MPC dont work with ffdshow 23052003!
Interest- WMP6.4 works(?)

Maybe cause in VMR7 (VMR7 is default renderer for win9x, best choice for win9x) using. If I switch for VMR9 (in ZoomPlayer), all is O.K.
But VMR9 works forced only in YUY2 (for me), degrades slightly quality and consumes much more CPU power...
:(

[HughMagoo]

I noticed that the tearing that I was noticing with resize in ffdshow and VMR9 is only when using YV12 output. I have not noticed the tearing effect with YUY2 output. If I enable almost all the output colorspaces, ffdshow uses YUY2 in all cases I have seen so far. So perhaps that is the best for my system? GF4ti4600, running nView

limited testing so far, but 23052003 seems to be a big improvement over 24042003 on my system. I use ZoomPlayer, and have not experienced any problems so far with Overlay or VMR9 modes and the new ffdshow.

[HarryM]

Originally posted by HughMagoo
I noticed that the tearing that I was noticing with resize in ffdshow and VMR9 is only when using YV12 output. I have not noticed the tearing effect with YUY2 output. If I enable almost all the output colorspaces, ffdshow uses YUY2 in all cases I have seen so far. So perhaps that is the best for my system? GF4ti4600, running nView

limited testing so far, but 23052003 seems to be a big improvement over 24042003 on my system. I use ZoomPlayer, and have not experienced any problems so far with Overlay or VMR9 modes and the new ffdshow.

Yes, VMR9 is more 'compatible' (it seems to me), but often use YUY2 for output.
I prefer VMR7 personally.

If you use Win2k/XP, your default overlay renderer is VMR9.
If you use Win9x/Me, your default overlay renderer is (old, good) VMR7.

[faxmactor]

Originally posted by HarryM
23052003 dont works on win98SE (for me).
24042003 works fine.

I have Win98 SE Hungarian.

The last one working eas 20021223
I haven't been able to use ffdshow sine 20030103...
I was unable to use 20030423 as well. :(

(On WinXP, it worked well)

At last, ffdshow-20030523 works on my Win98!! :p

But now I see from your post, that it still has problems with Win9x... :(

Will this issue ever be solved? The developers must have aware of this problem for almost half a year now. When a new ffdshow build released, it is still like gambling that it works or not on Win9x. Since it is kind of "random" I can imagine that it is quite difficult to track this bug down, but it would be just about time :scared:

[faxmactor]

Ah, sorry everyone, now I see HarryM, you had crashes on Win98, not that general problem, that even prevented ffdshow to get registered in Windoze.

[Blight]

HarryM:
you're wrong there matey...

VMR7 only exists under WinXP, there is no way you can get it under any other OS. What you get is Overlay and not VMR7. On my GF4, VMR9 has better scaling than overlay (using the 30.82 drivers, the 4x.xx drivers are buggy and give either aliasing or a slightly blurred image).


BTW, While "Theora" is listed as a codec, VP31 AVI files won't decode using ffdshow.

[HarryM]

Originally posted by Blight
HarryM:
you're wrong there matey...

VMR7 only exists under WinXP, there is no way you can get it under any other OS. What you get is Overlay and not VMR7. On my GF4, VMR9 has better scaling than overlay (using the 30.82 drivers, the 4x.xx drivers are buggy and give either aliasing or a slightly blurred image).


BTW, While "Theora" is listed as a codec, VP31 AVI files won't decode using ffdshow.

Sorry, my mistake. I think, that

'Video Renderer' = 'VMR7' (direct YV12 decoding, I prefer this)

'Video Mixing Renderer 9' = 'VMR9'. (I don't like this, YUY2 decoding, much CPU power needed)

[Sigmatador]

@Milan
since the beginning there's a color bug with awarpsharp

[Animaniac]

Two things I noticed with the latest build:

-Turing Warpsharp on while the stream is play causes WMP to crash.
-Perspective correction now makes the colors all funky.

As far as performance in this build, I can't really tell, but by virtue of looking at the Info section of the filter properties, I would have to say it's slower on my (P4) system, since the frame rate seems to dip lower more frequently and the data in the Info pane refreshes at a lower frequency. With the older builds, explorer rarely crashed when generating thumbnails.

[sysKin]

Hi,

Just to mention that I have some stability problems with this latest ffdshow. It crashed once when I enabled postprocessing, once when I seeked, several times when I opened a file.

I can only say much about the last example: I could open some files, couldn't open others. All were xvids with bframes, with dx50 fourcc. It crashed because of access violation at some big unrolled mmx function.
Changing decoder from libavcodec or xvid allowed to open the file.

My cpu is athlon xp, I'm using w2k (with service pack) and DX 8.0. Video is geforce2mx400 running at detonators 44.03.

Thanks for excellent work!

Radek

PS. ah and also, this version is still unable to say "qpel" in OSD.

[Defiler]

If I enable the highest level of post-processing, then Nic's (default settings), with "Nic's first" selected, I get repeatable crashes when opening DivX / XviD files.

Also, the Chrominance filter under "Blur & NR" seems to crash quite rapidly.

Finally, I've been unable to get this new version to post-process raw video (from the Sonic Cineplayer DVD decoder). It claims that ffdshow denied the attempt to attach a pin. (Yes, I have raw video enabled.)


The third item could easily be something I am doing wrong, though.

[sysKin]

Originally posted by Defiler
If I enable the highest level of post-processing, then Nic's (default settings), with "Nic's first" selected, I get repeatable crashes when opening DivX / XviD files. You were faster, Defiler :))
Yes, I confirm: the crashes I was talking about are strictly related to chroma deringing using "Nic's" method (for me: first or not first, just enabled). As I said, it's access violation in the middle of mmx code.

Regards,
Radek

[drebel]

@Defiler
I had the same problem with April's version, thing which kept me away from using Cinemaster's video filter for a while.Good compatibility with Intervideo's filter (Platinum) serves me well ,for the time being .But due to my hardware limitations i can't enable more than one pp filter now.In a very strange way , TheaterTek was "lighter" than ZoomPlayer3 with ffdshow

Chrominance filter under "Blur & NR" seems to crash quite rapidly.
confirmed

[Defiler]

Is there any preliminary documentation for the Avisynth features of ffdshow? What I'm really interested in is being able to refer to the input ffdshow passes to Avisynth by name, so that I can use it with functions that require clip names.
For example:

blah=ffdshowSource()
AwesomeStuff(blah,bugs=0,good_acting=true)

Is this, or something like it, possible?
So far, I've only been able to use functions that do implicit last.

Edit: Also, I've been unable to make any functions that refer to the previous frame in the video. I take it this isn't supported?
Try this, for an example:
ScriptClip(last,"Subtitle(String(YDifferenceFromPrevious()))")
Doesn't work. It just reports one value, and never changes. (for me)

However, this example:
ScriptClip(last,"Subtitle(String(YDifferenceToNext()))")
Works properly, and continuously updates.

I suppose this makes sense. We can't keep a lot of raw video lying around in memory. Perhaps it could keep the last few frames when in Avisynth mode? That would make the Avisynth feature significantly more powerful, since many filters need to refer to the previous frame in order to make correct decisions about the next one.

[mf]

Originally posted by Defiler
Is there any preliminary documentation for the Avisynth features of ffdshow? What I'm really interested in is being able to refer to the input ffdshow passes to Avisynth by name, so that I can use it with functions that require clip names.
For example:

blah=ffdshowSource()
AwesomeStuff(blah,bugs=0,good_acting=true)

Is this, or something like it, possible?
So far, I've only been able to use functions that do implicit last.
I think you can do that with something like:
input = nop()
Or anything else that doesn't do anything to your input. I'd have to check though.

[Wilbert]

Is there any preliminary documentation for the Avisynth features of ffdshow? What I'm really interested in is being able to refer to the input ffdshow passes to Avisynth by name, so that I can use it with functions that require clip names.
For example:

blah=ffdshowSource()
AwesomeStuff(blah,bugs=0,good_acting=true)
I'm probably misunderstanding you, but can you use ffdshowSource in an AviSynth script? How does AviSynth recognize the command ffdshowSource. Again, I'm probably just misunderstanding you.

[Defiler]

Originally posted by Wilbert
I'm probably misunderstanding you, but can you use ffdshowSource in an AviSynth script? How does AviSynth recognize the command ffdshowSource. Again, I'm probably just misunderstanding you. "ffdshowSource()" is a command I just made up, because I don't know the actual name for it. Heh.
You can't use ffdshow as the source for an external Avisynth script (..yet.. heh..) but you can execute Avisynth scripts on an ffdshow output stream, as post-processing.
Like this:
http://hellninjacommando.com/temp/ffdshow-avisynth.png

My question earlier is about whether you can refer to the input that Avisynth (inside ffdshow) gets from the ffdshow decoder by a clip name, or as an input function.

[pandv]

blah=ffdshowSource()
AwesomeStuff(blah,bugs=0,good_acting=true)

Is this, or something like it, possible?
So far, I've only been able to use functions that do implicit last.

I tested it with the last ffdshow version and this works:

myclip=last
myclip2=SwapUV(myclip)
myclip=GreyScale(myclip)
return myclip2

returns the clip swapped (and not greyscaled).

So, I supose the key is the first line:

myclip=last

pandv

[Defiler]

Cool. Good work.