ffdshow tryouts project: Discussion & Development

[STaRGaZeR]

Originally Posted by nevcairiel

Custom code only sucks when you use subversion trying to manage it .. what you do .. :P
Don't blame code for your version control shortcomings, imho

Custom code sucks the same on git, IMHO

[nevcairiel]

But keeping in sync with the upstream repository is trivial!

[STaRGaZeR]

I don't know git, but it can't be worse than svn

To all, please test r3602. Things to test:

a) Audio channel mapping is fine for 5.1 streams and above. Fine means the same as in previous revisions.
b) Yadif works fine. Please test on CPUs without SSSE3, SSE2 and MMX2, or any combination of them. If your CPU has SSE2 but no SSSE3 for example, test it. Fine means it doesn't crash and the speed is the same as before or within margin of error.

Build: http://www.mediafire.com/?l90nft4ccgam2o4

And to the devs, mplayer now only contains mp3lib, which is IMO not needed at all, with libavcodec and libmad. We can remove it and libmplayer will be gone.

[leeperry]

Originally Posted by STaRGaZeR

unless there is a strong negative from ffmpeg against this, in that case I'll consider it.

Put simply, they don't seem to care

And sorry to beat a dead horse(I know, that's my speciality ), but would you have any idea how to enable PCM24/32 for winamp DSP plugins? the input/output is currently downconverted to 16int, which is a major bummer as there's many DX/VST plugins wrappers...and ffdshow is the perfect place for this kind of job

You can see tons of "int16_t"/"SF_PCM16" in those sections:
http://ffdshow-tryout.svn.sourceforg...p&pathrev=1928
http://ffdshow-tryout.svn.sourceforg...p&pathrev=1928

winamp2 DSP plugins work in PCM24/32 using the foobar wrapper: http://wiki.hydrogenaudio.org/index....oo_dsp_winamp)

16-bit: low quality, failsafe. Choose this setting, if you encounter a problem while playback;
24-bit: hi-quality;
32-bit: highest quality.

I asked its author for help but he told me that the ffdshow code was quite messy and that he had no time to work on it

Christian Budde(a professional VST plugin coder) helped Haruhiko to iron out a few bugs in the winamp DSP plugins host code of ffdshow, and confirmed that PCM24/32 were entirely possible w/ winamp2 DSP plugins.

If anyone's interested, I could make a complete tutorial to VST awesomeness in ffdshow...including upmix/downmix matrixes based on Lexicon's Logic7 routines. The key is to calibrate their coefficients to your gear/brain. Its author, David Griesinger, is the living god of audio DSP for hometheater use.

After a bit of work, you can make very convincing stereo downmixes(very short FLAC samples):
http://www.mediafire.com/download.php?hzgj3fsco1gqsbf
http://www.mediafire.com/download.php?xrfcj9rm2od453k

Originally Posted by STaRGaZeR

b) Yadif works fine. Please test on CPUs without SSSE3, SSE2 and MMX2, or any combination of them. If your CPU has SSE2 but no SSSE3 for example, test it. Fine means it doesn't crash and the speed is the same as before or within margin of error.

I'm a noob at deinterlacing, but YADIF always gives me tons of combing...I was trying it on a 29.97fps NTSC source yesterday, and bingo still a combo feast. It works wonderfully in double frame rate mode, though...but it's not too handy in 89.91Hz.

One thing I've never understood about double frame rate deinterlacing is that it looks like frame interpolation...It really looks as if I were watching a 59.94fps source, when it should actually be 29.97*2

A friend of mine is completely hooked to 1080i*2 on his projector, he finds the picture WAY smoother.

[kieranrk]

Originally Posted by leeperry

One thing I've never understood about double frame rate deinterlacing is that it looks like frame interpolation...It really looks as if I were watching a 59.94fps source, when it should actually be 29.97*2

A friend of mine is completely hooked to 1080i*2 on his projector, he finds the picture WAY smoother.

Double frame rate is the *right* way to do it. That's what true interlaced looks like if you were watching it on a CRT. Just because there's a lot of 24/25/30p misuse out there and that many people are used to watching web video (or pirated video) at a max of 30p doesn't suddenly mean it's right.

[kieranrk]

Originally Posted by leeperry

There's no point in decoding lossy audio in anything else than 32fp.

Why not use 64fp (doubles); in fact why not use 128fp (long double); don't forget you're losing quality at each stage...

[leeperry]

Originally Posted by kieranrk

Double frame rate is the *right* way to do it. That's what true interlaced looks like if you were watching it on a CRT. Just because there's a lot of 24/25/30p misuse out there and that many people are used to watching web video (or pirated video) at a max of 30p doesn't suddenly mean it's right.

I know what 24fps looks like, same goes for 25 and 29.97....29.97 double frame rate deinterlace looks like 29.97 interpolated to 59.94 to my eyes. Trimension DNM all the way.

Originally Posted by kieranrk

Why not use 64fp (doubles); in fact why not use 128fp (long double); don't forget you're losing quality at each stage...

my favorite audio player(Lilith) decodes all lossy audio in 64fp, and so are its volume attenuator and VST plugins pipeline. SQ is amazing.



But right now, libavcodec DTS and VST plugins are stuck to 16int in ffdshow

[kieranrk]

Originally Posted by leeperry

I know what 24fps looks like, same goes for 25 and 29.97....29.97 double frame rate deinterlace looks like 29.97 interpolated to 59.94 to my eyes. Trimension DNM all the way.

my favorite audio player(Lilith) decodes all lossy audio in 64fp, and so are its volume attenuator and VST plugins pipeline. SQ is amazing.



But right now, libavcodec DTS and VST plugins are stuck to 16int in ffdshow

Clearly you don't understand the difference between interlace and progressive then....

I'm glad you enjoy your 64-bit double placebo. The rest of us live in reality.

[leeperry]

they took a 29.97fps progressive source, once properly deinterlaced it should be reconstructed to 29.97 progressive...no need for frame interpolation. anyway, I don't care for interlaced stuff...I have a bunch of poorly authored DVD's using it, that's all.

oh yes, placebo...good thing I didn't tell you about software induced jitter, I don't think you'd be ready for that just yet.

and in case you were wondering, you can measure it too...so much for placebo.

even a 11yo girl can hear the difference, but she has better hearing than you and I.

foobar's volume attenuator also works in 64fp, just like all the pro-audio software....anyway, my original point was that 16int for lossy audio decoding and post-processing butchers the SQ.

[_xxl]

And to the devs, mplayer now only contains mp3lib, which is IMO not needed at all, with libavcodec and libmad. We can remove it and libmplayer will be gone.

Remove it. It is hard to maintain.

[STaRGaZeR]

Originally Posted by leeperry

I'm a noob at deinterlacing, but YADIF always gives me tons of combing...I was trying it on a 29.97fps NTSC source yesterday, and bingo still a combo feast. It works wonderfully in double frame rate mode, though...but it's not too handy in 89.91Hz.

One thing I've never understood about double frame rate deinterlacing is that it looks like frame interpolation...It really looks as if I were watching a 59.94fps source, when it should actually be 29.97*2

A friend of mine is completely hooked to 1080i*2 on his projector, he finds the picture WAY smoother.

First, please do not parrot your feature request/bug fixes/whatever each time you have the chance. It's annoying, and this is not the first time I tell you this.

Second, you should really learn what interlaced video is. If you have 29,97fps pure interlaced content you have 59,94fps of temporal information there. Outputting 29,97fps is an abortion, but needed in some cases. That's why deinterlacers have that option.

Originally Posted by _xxl

Remove it. It is hard to maintain.

OK. What about theora? It hasn't been included in the installer for quite some time, why is still there?

[Astrophizz]

That XXHighEnd developer and the people that buy into his overpriced software sound a bit loony to me. You can still measure something with a sensitive device and yet not be able to sense a difference and so you would be subject to a placebo effect. I could point out the differences in spectral patterns of two stars to determine that one is of slightly different composition, but I could never see the difference with my eyes.

[leeperry]

Originally Posted by Astrophizz

You can still measure something with a sensitive device and yet not be able to sense a difference and so you would be subject to a placebo effect. I could point out the differences in spectral patterns of two stars to determine that one is of slightly different composition, but I could never see the difference with my eyes.

Of course, many FREE audio engines provide the same sort of low jitter(cMP/Reclock/Lilith)...I couldn't care less about XXHighEnd myself. Did you compare all the media players, do they sound the same to you? Jitter is real, and far from placebo.

Originally Posted by STaRGaZeR

this is not the first time I tell you this.

If you had a bugtracker anyone cared about, this wouldn't be happening...bumping bugfixes requests is the only chance to see anything happening(just like MPC).

I've asked three very skilled audio plugins coders for help and they've all told me that the ffdshow audio code was a bug feast(full of syntax errors and beginners mistakes) and a huge mess...so I guess this is not going to change anytime soon. They were even wondering how it could be working at all

ffdshow is a lost cause for studio grade audio, atm it's just a toy to watch movies in WMP...sorry for rubbing it in

[fastplayer]

Originally Posted by STaRGaZeR

OK. What about theora? It hasn't been included in the installer for quite some time, why is still there?

clsid disabled it some time ago in the hope that someone would update it. IIRC the current version in the trunk is hopelessly outdated.

[nm]

Originally Posted by leeperry

I'm a noob at deinterlacing, but YADIF always gives me tons of combing...I was trying it on a 29.97fps NTSC source yesterday, and bingo still a combo feast. It works wonderfully in double frame rate mode, though...

I haven't seen yadif leaving residual combing in neither single nor double-rate mode if the spatial deinterlacing check is enabled (not skipped, modes 0 and 1 in the MPlayer implementation). Do you have a sample clip?

[leeperry]

Originally Posted by nm

I haven't seen yadif leaving residual combing in neither single nor double-rate mode if the spatial deinterlacing check is enabled (not skipped, modes 0 and 1 in the MPlayer implementation). Do you have a sample clip?

I didn't have it disabled, I'll try to come up w/ a sample.

so videos shot interlaced actually have 59.94fps, great! that would explain why my friend is so hooked to 1080i concerts on his projector...but I believe he also likes 1080i movies, that can't possibly have more than 24fps AFAIK(IVTC put aside).

[_xxl]

OK. What about theora? It hasn't been included in the installer for quite some time, why is still there?

If you can update it ok. If not remove it.

x264 should be updated to latest, I can t update it anymore....

[Reimar]

Originally Posted by madshi

An eternity ago I had asked both Benjamin Larsson (DTS) and Justin Ruggles (AC3 + E-AC3) about that and IIRC both said that native bitdepth output was planned for a future libav version. But that was ages ago. I had sent my eac3to patches to both of them, but neither was willing to apply them. So I keep on patching libav, year after year...

I think some were changed. The biggest issue is that the format conversion inside the codec uses some tricks to be a lot faster, particularly on systems with slow FPU like ARM. Just removing them would mean that FFmpeg would get a lot slower on these systems.
And for FFmpeg speed (even on non-mainstream platforms) is most important than output resolution.
But if someone sends a patch to fix this issue, there'd probably no objections to changing the output format.

[madshi]

Originally Posted by Reimar

I think some were changed. The biggest issue is that the format conversion inside the codec uses some tricks to be a lot faster, particularly on systems with slow FPU like ARM. Just removing them would mean that FFmpeg would get a lot slower on these systems.
And for FFmpeg speed (even on non-mainstream platforms) is most important than output resolution.

I understand. I'm just wondering why there isn't a simple (compile time) configuration switch for activating "native" output. I did try to get such a switch added to the (E-)AC3 and DTS decoders, but my patches were declined.

[Reimar]

Originally Posted by madshi

Incorrect. Dithering does not *replace* the lowest bit. You clearly don't understand how dithering works.

Bad choice of words maybe. But to clarify: If you take a 16 bit input, add random bits to expand it to 24 bit and then dither that to 16 bit the dithering will have added noise. If the encoder never had more than 16 bit to start with, anything more than 16 bit in the decoded output can't be much more than noise (I guess this is not 100% true since the encoder can "expand" the input data to more than 16 bit by knowing how audio data is "supposed" to look).