LAV Filters - DirectShow Media Splitter and Decoders

[wanezhiling]

http://www.sendspace.com/file/2y7z2h
LAV detects its duration 3min49s, MediaInfo says 3min43s.

[filler56789]

http://www.sendspace.com/file/2y7z2h
LAV detects its duration 3min49s, MediaInfo says 3min43s.

It's because LAV Splitter thinks the cover picture is part of the audio stream :)

[wanezhiling]

Make sense.

[nevcairiel]

@Nev
I've sent you (via pm) a TS sample which doesn't play smoothly. Bad timestamps?

EDIT:
Jerky playback using the latest madVR (MPC-HC 1.7.4).
Slow playback (~20fps) using custom EVR.
Weird (but best) playback using EVR.
I wonder what causes it.

Yes, it has weird timestamps, but I'm not 100% sure why.
I have some plans to try to come up with a sane way to try to automatically improve such problems, but we'll just have to wait and see if that works out.

[clsid]

Does the WMV9 MFT decoder have any advantages compared to the DMO one?

[nevcairiel]

If not being stupid broken on Windows 8.1 counts, sure.

[Finnish Flash]

Hey nev, awesome job with the decoder, mate!

I've been using LAV + MPC-HC for streaming MPEG-TS from my ISP's recording service. Aside from the playback taking some extra seconds to start comparing to VLC, it's working very well.

In order to be able to rewind and fast-forward and get the full duration of a file instead of just a couple of seconds from the beginning, I have to insert the following to registry:

[Windows Registry Editor Version 5.00

[HKEY_CLASSES_ROOT\http]
"Source Filter"="{B98D13E7-55DB-4385-A33D-09FD1BA26338}"

This enables splitting video and audio from HTTP protocol with LAV Splitter, right?

Could you make this a default setting, or is there some downside to this? I mean, this could help someone else, too.

[sneaker_ger]

Should LAV signal bt601 colors for VP8 video? At the moment it leads to the usual resolution based guessing by the renderer but the spec seems to only supports "YUV color space similar to the YCrCb color space defined in [ITU-R_BT.601]".
http://datatracker.ietf.org/doc/rfc6386/?include_text=1 (9.2)

Opinions?

[mandarinka]

That's a problem, because it is quite clear that when an HD video gets encoded, it is originally BT709. And I bet the chances are pretty high that the transcoding tool used won't convert the colormatrix...

Yay On2/Google...

[sneaker_ger]

That's a problem, because it is quite clear that when an HD video gets encoded, it is originally BT709. And I bet the chances are pretty high that the transcoding tool used won't convert the colormatrix...

That may be, but VP8 is pretty much only used in the context of websites so it makes sense if the browsers use BT.601 as well (which they allegedly do).

[e-t172]

Uh? I would rather expect the browsers to conform to sRGB (since it's the web standard), which has the same primaries as BT.709.

[foxyshadis]

Should LAV signal bt601 colors for VP8 video? At the moment it leads to the usual resolution based guessing by the renderer but the spec seems to only supports "YUV color space similar to the YCrCb color space defined in [ITU-R_BT.601]".
http://datatracker.ietf.org/doc/rfc6386/?include_text=1 (9.2)

Opinions?

The VP8 spec is so ass-backwards that it's likely that On2 simply didn't know BT.709 existed and would do all YUV<->RGB conversions with BT.601, and they put that in because they'd never considered signaling it, but in nearly all cases VP8 shouldn't be involved in any RGB conversion at all. It won't know how the original source was converted, and shouldn't care, like any other video codec (unless manually signaled).

Resolution-dependent switching is a de-facto standard, and in the absence of signaling there's never a good reason to ignore it. It's sad, but it's not the only place you have to defy the spec to work with VP8....

[sneaker_ger]

Resolution-dependent switching is a de-facto standard
This is what I'd like to test. As I said: when browsers do confirm to the BT.601 part it's hardly the defacto-standard for VP8. It would make sense for LAV to use BT.601 as well so people see the same colors in their player as they see in their browser. Most use cases would be (1) test files before uploading to your own HTML5 website, (2) watch files downloaded from the web. Nobody uses VP8 to backup their Blu-Ray collection.

[mandarinka]

So in case HD videos are bt709, LAV should do it incorrectly just because other software is also dumb?
Browser video playback implementations are rather crude in any case, so they shouldn't be viewed as a reference for anything (besides sucking).

/OTOH - in many cases, e.g. youtube, the HD streams will actually be upscales of SD sources. But LAV Video isn't a youtube player, it is a directshow decoder./

[sneaker_ger]

So in case HD videos are bt709, LAV should do it incorrectly just because other software is also dumb?
If the spec says videos are BT.601 than LAV would not be doing it incorrectly, it would be the files that have been created incorrectly. And who says other software is dumb? They are following the specs. If you encode video for the web without regard to specs and browsers on the other hand...

[wanezhiling]

http://www.sendspace.com/file/5uoefl
Can't play the ape file.

[mandarinka]

They are following the specs. If you encode video for the web without regard to specs and browsers on the other hand...

That would be if they didn't do the same for H.264 too. IIRC (I might be wrong...) it is not about spec, it is because they don't know any better and VP8 happens to be the same.

[RealSnoopyDog]

Hi, i just recongized that "newer" versions of the LAV filters (versions which contain the "avfilter-lav-4.dll") are not able to playback mp4 videos with a low frame rate (e.g. 16.6667 Hz) synchronous anymore. Video is played back way too fast, video and audio are totally out of sync. When i use older versions of the LAV filters where the "avfilter-lav-3.dll" is still included, playback is ok.

[filler56789]

Hi, i just recongized that "newer" versions of the LAV filters (versions which contain the "avfilter-lav-4.dll") are not able to playback mp4 videos with a low frame rate (e.g. 16.6667 Hz) synchronous anymore. Video is played back way too fast, video and audio are totally out of sync. When i use older versions of the LAV filters where the "avfilter-lav-3.dll" is still included, playback is ok.

Over here, version 0.61.2 plays a 12fps MP4 file at 12fps.

Perhaps your low framerate MP4s are borked? :confused:

[qtwebkit]

Problem decoding this file (http://pan.baidu.com/s/1c0nFC6c) with black screen, both software and hardware decoding.
Only software decoding with potplayer's internal filter works fine, broken with pot dxva.

[P.J]

When will we get HEVC H.265 support through QuickSync? :)

[LigH]

Not before intel would produce a GPU which is able to accelerate H.265 decoding at all.

[zerowalker]

Got a question,

Those who can use , QuickSync, DXVA2 (Native) and/or NVIDIA CUVID.

Do you get a delay when seeking?
For example, if i jump forward 5 sec using the software decoder, it will almost instantly show that image (Depending on keyframes and CPU usage of course).
While DXVA2 (Copy-Back) does not, it has a bit of a delay which is kinda annoying as i seek a lot (I can only use Copy-Back, hence why i ask about the others).

I am not sure if it's supposed to be that way, or is a Copy-Back limitation etc, or something that can be fixed/improved.

It's not really an issue, but would be interested to hear about it.

[wanezhiling]

Problem decoding this file (http://pan.baidu.com/s/1c0nFC6c) with black screen, both software and hardware decoding.
Only software decoding with potplayer's internal filter works fine, broken with pot dxva.
Baidu is extremely slow for nevcairiel, I uploaded it to another host (http://yunpan.cn/QiuVDPerjd4RF), nev please give a try.

password: 2240

[sneaker_ger]

Got a question,

Those who can use , QuickSync, DXVA2 (Native) and/or NVIDIA CUVID.

Do you get a delay when seeking?
For example, if i jump forward 5 sec using the software decoder, it will almost instantly show that image (Depending on keyframes and CPU usage of course).
While DXVA2 (Copy-Back) does not, it has a bit of a delay which is kinda annoying as i seek a lot (I can only use Copy-Back, hence why i ask about the others).

I am not sure if it's supposed to be that way, or is a Copy-Back limitation etc, or something that can be fixed/improved.

It's not really an issue, but would be interested to hear about it.

Many hardware decoders (especially AMD) are much slower than modern CPUs so it's normal depending on your hardware.

[zerowalker]

Oh, so it's simply a delay caused by "CPU Usage" but for the hardware decoder?

I was in the thought that Hardware decoders were extremely low-level and optimized, resulting in much faster decoding compared to a CPU of the same level.
How does it work for you when seeking, if you have any other of the hardware decoders?

[DarkSpace]

Is there a way to make LAV Splitter auto-select its audio based on languages only when no default flag was found?
I read the ReadMe for the Splitter auto-selection rules, searched this thread, and even tried some configurations myself (for example "*,eng" for "Use Default, when available, otherwise default to English audio), but none of them worked.
I also searched this thread, but came up with no results.

[michkrol]

Oh, so it's simply a delay caused by "CPU Usage" but for the hardware decoder?
Yes, the decoder needs to process all the frames from a keyframe to the frame you want to see.
You might want to check out fast seek (on keyframe) if you're using MPC-HC (find it under tweaks in options).
I was in the thought that Hardware decoders were extremely low-level and optimized, resulting in much faster decoding compared to a CPU of the same level.
They are optimized for power consumption, not speed. You rarely want to play a movie faster than realtime, although the extra speed helps while seeking.
How does it work for you when seeking, if you have any other of the hardware decoders?
For me the seeking is almost instant (on same level as software decoding) with Intel HD4000 using QuickSync and Geforce GTX 750Ti using CUVID/DXVA-CB/DXVA-N.
Is there a way to make LAV Splitter auto-select its audio based on languages only when no default flag was found?
I read the ReadMe for the Splitter auto-selection rules, searched this thread, and even tried some configurations myself (for example "*,eng" for "Use Default, when available, otherwise default to English audio), but none of them worked.
I also searched this thread, but came up with no results.
I don't think you can do that. Such level of flexibility is only offered for subtitle streams selection.

[DarkSpace]

I don't think you can do that. Such level of flexibility is only offered for subtitle streams selection.
Thank you for confirming that. I'm going to open an issue for that, but I wanted to make sure it's not possible already.

[zerowalker]

Yes, the decoder needs to process all the frames from a keyframe to the frame you want to see.
You might want to check out fast seek (on keyframe) if you're using MPC-HC (find it under tweaks in options).

Okay that explains the Delay as it's processed faster than realtime, or rather is forced to.
I am using ZoomPlayer however so not an option, though can check it on MPC-HC as i use it in some occasions.

They are optimized for power consumption, not speed. You rarely want to play a movie faster than realtime, although the extra speed helps while seeking.

But they are better than Software decoding, in pure efficiency right?
As if not, the power consumption idea falls down the floor.

For me the seeking is almost instant (on same level as software decoding) with Intel HD4000 using QuickSync and Geforce GTX 750Ti using CUVID/DXVA-CB/DXVA-N

That's great to hear!

A quick one to that, is all those limited to H264 standard settings like DXVA2, or are they more flexible?

[vivan]

But they are better than Software decoding, in pure efficiency right?
As if not, the power consumption idea falls down the floor.That depends... for example on my laptop (intel i5-2410M + nVidia 540M), in terms of power consuption: QS (intel) < software < CUVID (nVidia). Which means that nVidia's hw decoder is useless there.

A quick one to that, is all those limited to H264 standard settings like DXVA2, or are they more flexible?Check this thread http://forum.doom9.org/showthread.php?t=163110

[nevcairiel]

That depends... for example on my laptop (intel i5-2410M + nVidia 540M), in terms of power consuption: QS (intel) < software < CUVID (nVidia). Which means that nVidia's hw decoder is useless there.

Thats not the hardware decoders fault, using CUVID puts the GPU into high-performance mode, which obviously eats more battery power.

If you use DXVA2-CB or Native on the same GPU, it'll use much less.
If you system has both Intel and NVIDIA in a mobile product, the Intel is probably going to be using less though.

[vivan]

If you use DXVA2-CB or Native on the same GPU, it'll use much less.Yeap. But the only way to do this - is to force media player to use dGPU (otherwise it'll use intel hw decoder). And using it for rendering will result in higher power consuption again.

Yes, it's nVidia fault, not decoder's. What I wanted to say is that while h/w decoder could be more efficient than s/w decoder - using it will not always result in lower power consumption.

[u28151]

Hi,

I wonder if there are any chance making LAV Filter settings systemwide opposed to user based? Reason is that the settings seem to be hardware related. Also it will be easier to auto deploy LAV since you do not have to consider configuring each logged on users registry hive.

Thanks for a great piece of software!

/U

[mecedo]

Can LavVideo decode 4k video on Intel Haswell processors?
On my Core i3 4130T when decoding with QuickSync there are frame drops. madVR (even with nearest neighbour scaling algorithms) shows only 1-2 frames in decoder queue so I suppose that LavVideo is weak point. And it's strange that GPU and CPU utilization is below 70%.

[nevcairiel]

Don't use QuickSync, it has a CPU overhead. Use DXVA2-Native instead to avoid this overhead.
You should also check if 4K/UHD hardware decoding is actually enabled in LAV Video, its not by default.

[andyvt]

Hi,

I wonder if there are any chance making LAV Filter settings systemwide opposed to user based? Reason is that the settings seem to be hardware related. Also it will be easier to auto deploy LAV since you do not have to consider configuring each logged on users registry hive.

/U

System wide settings would require admin rights. It's possible to do that without users experiencing UAC prompts (or needing admin rights at all), but it would be significantly more complicated to build and support and provide little ROI.

Depending on the exact scenario, this is something you could address quite easily with a login script or something similar.

[nevcairiel]

You can write default settings to HKLM manually, LAV will read them if HKCU doesn't contain settings yet. That way your deployment could define defaults for all users, but users could still customize individually - and reading from HKLM doesn't require admin.

[mecedo]

Don't use QuickSync, it has a CPU overhead. Use DXVA2-Native instead to avoid this overhead.
You should also check if 4K/UHD hardware decoding is actually enabled in LAV Video, its not by default.

But with "DXVA Native" CPU usage is 100% and "Active decoder" points to avcodec. Does avcodec mean that decoding is by software?
UHD (4K) checkbox is selected of course.

[nevcairiel]

Then your hardware refuses to decode those clips.

[LigH]

May depend on the video format. I doubt that the decoding of all possible formats can already be accelerated by hardware.

[mecedo]

Theoretically Intel Haswell can decode 4k. With QuickSync decoder CPU usage is noticeably lower, but as I described frames are dropped.
With DXVA2 even FullHD 24p isn't hardware decoded on Intel processors, because CPU usage is as high as with software decoding. It was in my previous CPU SandyBridge and with current Haswell processor.

I've tried 4k clips from this site:
http://box.houkouonchi.jp/4k/

[andyvt]

Theoretically Intel Haswell can decode 4k. With QuickSync decoder CPU usage is noticeably lower, but as I described frames are dropped.
With DXVA2 even FullHD 24p isn't hardware decoded on Intel processors, because CPU usage is as high as with software decoding. It was in my previous CPU SandyBridge and with current Haswell processor.

I've tried 4k clips from this site:
http://box.houkouonchi.jp/4k/

Something is wrong with your system. HSW can decode 4K AVC in hardware..

http://babgvant.com/images/hsw4k.jpg

[u28151]

You can write default settings to HKLM manually, LAV will read them if HKCU doesn't contain settings yet. That way your deployment could define defaults for all users, but users could still customize individually - and reading from HKLM doesn't require admin.

Thank you, this option looks like the most suitable for me!

[mecedo]

Something is wrong with your system. HSW can decode 4K AVC in hardware..


What operating system and videoplayer are you using?

[andyvt]

What operating system and videoplayer are you using?

In the screenshot I was using GraphStudioNext (a DirectShow prototyping/testing tool) on Windows 8.1 (the OS, as long as it's modern shouldn't matter). Any application using the same filters shown (i.e. LAV + EVR) should enjoy the same experience.

[mecedo]

I'll try with 64 bit OS because I'm using 32bit Win 7.
I don't have idea what causes DXVA2 not working.

[andyvt]

I'll try with 64 bit OS because I'm using 32bit Win 7.
I don't have idea what causes DXVA2 not working.

Probably drivers.

[mecedo]

I've tried with Intel® Iris™ and HD Graphics Driver 15.33.18.3496 and beta 15.33.0.3574 from Intel's download page.

[andyvt]

I've tried with Intel® Iris™ and HD Graphics Driver 15.33.18.3496 and beta 15.33.0.3574 from Intel's download page.

What app are you using? Have you tried doing it explicitly like I did?

Do you have anything like VSFilter (DirectVobSub) installed? If so, it will disable DXVA in many cases.