LAV Filters - DirectShow Media Splitter and Decoders

[roytam1]

Actually, no -- -tryouts uses a weird mix of ffmpeg and libav as far as I can remember -- with random modifications all over because of legacy reasons ("All hail Milan!").
but from the svn log, it should count as libav-based as it updates from libav but not ffmpeg.

for difference between ffmpeg and libav, see this blog post:
http://codecs.multimedia.cx/?p=370

[BloodySword]

For me it doesn't matter if it is ffmpeg or libav, it should just work.

[JEEB]

but from the svn log, it should count as libav-based as it updates from libav but not ffmpeg.

for difference between ffmpeg and libav, see this blog post:
http://codecs.multimedia.cx/?p=370
It's just been called 'libav' in the changelogs :P

I recommend you do
grep "FFmpeg" -R *
grep "Libav" -R *in -tryouts' src/ffmpeg

And I know the difference of libav and ffmpeg, I've been posting bugs and having stuff fixed as well as just conversed on both sides.

Anyways, this is off-topic here so I'll stop here.

For me it doesn't matter if it is ffmpeg or libav, it should just work.
Indeed. Also, it shouldn't exactly be hard to compile ffmpeg, just grab a proper'ish msys/mingw environment, f.ex. one from here (http://sourceforge.net/apps/trac/mpc-hc/wiki/How_to_compile_the_MPC) (although I still use komisar's 4.5.3 since some stuff still has problems with 4.6.x).

Then just compile the thing a la the shell script in LAV Filters' root and you'll get yourself a library (of course you might also need MS's compiler's tools in your path in case you need the dot-lib).

[Thunderbolt8]

I cant get this file http://www.mediafire.com/?wtzcu8guck5cvur to play with LAVsplitter and WMVDecoderDMO. plays fine with ffdshow instead, though.

edit: works though when remuxing the video to .mkv. is it a problem of the dmodecoder with the m2ts structure or of other files inside?

[pankov]

Any progress on SW based deinterlacing?
or a CUVID deinterlacing with software decoding?
Nev,
what happened to your tests of this? what were the results?

[nevcairiel]

I cant get this file http://www.mediafire.com/?wtzcu8guck5cvur to play with LAVsplitter and WMVDecoderDMO. plays fine with ffdshow instead, though.

edit: works though when remuxing the video to .mkv. is it a problem of the dmodecoder with the m2ts structure or of other files inside?

The file plays just fine for me with the WMVideo Decoder DMO on Windows 7.

[Thunderbolt8]

oh well, playback doesnt start here, just stay at 0:00:00 with black screen -.-

[SamuriHL]

There may soon be no need for that... ;)

Wouldn't that be nice. ;)

Sent from my Xoom using Tapatalk

[cyberbeing]

P016 output seems broken (not used) in LAV Video going way back, is that a known issue?

[nevcairiel]

P016 output seems broken (not used) in LAV Video going way back, is that a known issue?

It would be used if a file were actually 16-bit. I don't offer it for 10 or 8 bit files.

[cyberbeing]

Why not offer it for 10-bit files? Is there something wrong with swscale 10-bit to 16-bit conversion?

While there isn't much of a point in outputting 16-bit, it'd be better if you just set it as the lowest priority format for each sub-sampling (4:2:0/4:2:2/4:4:4) level, while disallowing it only for 8-bit. That way it could be forced if desired for testing purposes.

[madshi]

I see no sense in 16bit output if the files are only 10bit (or lower). swscale conversion from 10bit to 16bit will not improve anything. The officially recommended 10bit to 16bit conversion method for 16-235 content just shifts the pixel values, so there will be exactly zero change in output, at least when using madVR.

P.S: The only sense in allowing P016 output for 10bit files would be to support renderers which accept P016 but not P010.

[Keiyakusha]

16bit output? no way. People would use it and would claim that quality is so much better so it even way beyond compare. Who needs that? One should make single custom build if a some point he'll need to test something like that.

[nevcairiel]

P016 and P010 are exactly the same memory layout (with P010 defined to have the 6 extra bits set to 0), so if anything supports P016 but not P010, it seriously fails.
I don't see the point allowing it.

[cyberbeing]

I never said it would improve anything, I'm more interested in it for benchmarking things which accept P016. I'm already aware that speed and quality between P010 (10-bit padded to 16-bit) and P016 output (10-bit shifted to 16-bit) should be identical if handled properly. xy-VSFilter now supports P010 and P016 input/output, but for whatever reason P016 was slower than P010 when using CoreAVC. I wanted to cross-check with LAV Video, but I'm obviously unable to in its current form. I'm thinking xy-VSFilter may not be handling P016 optimally, but without cross-checking with another decoder I don't know if it's a CoreAVC issue...

Well if someone could PM me a build with working P016 when LAV Video 0.38 is released I'd be grateful.

[clsid]

Thanks so much clsid and ney2x, you guys were both extremely helpful. I noticed when using the default windows shell extension; even after enabling the extensions I wanted via Codec Tweak Tool, I noticed a few flvs or mkv would not get their thumbnails.

When I used Icaros, every single file received a thumbnail. However, what worries me about Icaros is if it is a memory resident app or not. I'm hoping it simply replaces the default shell extension with a slightly more enhanced version without taking any extra resources.The MS one respects the preferred filter settings of Windows 7. Use Win7DSFilterTweaker to set LAV as preferred and it should work. It will use the first frame of the video though, limitation of the shell extension.

Icaros is buggy, so it should be used with caution.

[nevcairiel]

P010 (10-bit padded to 16-bit) and P016 output (10-bit shifted to 16-bit)

Actually, both are shifted to 16-bit. In P010, the renderer just knows to ignore the 6 lowest bits.

PS:
Painting subtitles on a 4:2:0 image is still a bad idea, you should convert it to RGB before - 10-bit support doesn't change that.

[cyberbeing]

Painting subtitles on a 4:2:0 image is still a bad idea, you should convert it to RGB before - 10-bit support doesn't change that.
Explain.

Why should I care that my rasterized TrueType/OpenType subtitles are painted on YUV if it's faster?

[nevcairiel]

The problem is not YUV, the problem is that the chroma is still subsampled. If you make the whole thing YUV 4:4:4, its also good.
Chroma subsampling means that you will probably get blurred borders and slight chroma errors on the borders.

[madshi]

The problem is that you need to do chroma upsampling in order to get YUV 4:4:4 or RGB, which costs CPU performance, especially if you want more than just Bilinear upsampling. IMHO the best subtitle logic would be to have the subtitle renderer render to an RGB texture with an alpha channel (totally separate from the video image) and ask the renderer to blend it over the video image. That's somewhat similar to what the MPC-HC internal subtitle renderer does. VsFilter could do a similar thing, but custom communication with the renderer would be needed for that.

[cyberbeing]

Chroma subsampling means that you will probably get blurred borders and slight chroma errors on the borders.
I've never seen that as a significant problem. The majority of subtitles are just black and white anyway, and the rest you couldn't care if they had pristine chroma as it's not something you'd notice during playback with font-based subtitle (madVR does a good job upsampling anyway). Now if you only care about image-based DVD/Blu-ray subtitles, then yes, rendering in RGB can make a big difference.

The problem is that you need to do chroma upsampling in order to get YUV 4:4:4 or RGB, which costs CPU performance, especially if you want more than just Bilinear upsampling.
That is the problem, which is why it's nice that madVR does the chroma upsampling to 4:4:4 on the GPU.

IMHO the best subtitle logic would be to have the subtitle renderer render to an RGB texture with an alpha channel (totally separate from the video image) and ask the renderer to blend it over the video image. That's somewhat similar to what the MPC-HC internal subtitle renderer does. VsFilter could do a similar thing, but custom communication with the renderer would be needed for that.
If your interested in such a interface/interaction between VSFilter and madVR, submit a Feature Request to the xy-VSFilter (http://code.google.com/p/xy-vsfilter/) project. The developer working on it has already re-written large portions of VSFilter and added many new features and speed enhancements (at a very quick pace I might add, since he seems comfortable with the code), so I doubt he'd mind adding a few more useful things. He's been a bit stumped on how to go about adding DXVA support, so he may be open to other options like you're suggesting. Back when he first made the project xy-VSFilter public, he also had expressed interest in eventually working on porting libass, so if you decide to go that route and need another dev it's possible he'd be interested. Speeding up and enhancing VSFilter was just an easier target for the time being.

[Keiyakusha]

Oh, ah, if madvr can cooperate with xy-vafilter to get this done, this would be so awesome... that's sounds too good to be true, I shouldn't be exited too much, should calm myself down... =)

[madshi]

If your interested in such a interface/interaction between VSFilter and madVR, submit a Feature Request to the xy-VSFilter (http://code.google.com/p/xy-vsfilter/) project.
Maybe later. Busy with other stuff right now.

[Thunderbolt8]

xy-VSFilter (http://code.google.com/p/xy-vsfilter/) project.is there a thread here on doom9?

[pankov]

Did anyone face a similar problem with the latest nvidia whql 285.62: viewing particular h.264 1080i channel in DVBViewer with LAV CUVID or LAV Video (+CUVID acceleration) causes slow motion effect. After a few seconds DVBViewer resets video playback and its source filter says there was a buffer overflow error. No problem with 285.27 beta or while playing recorded stream from HDD. Cyberlink decoder is OK on both video driver versions. Seems like Nvidia broke CUVID in the release.

I've just noticed the same problem.
I'll try the older drivers and report back later

[cyberbeing]

is there a thread here on doom9?
No there isn't. Feel free to make one, but I don't think the dev is currently a member of doom9. GoogleCode seems to work just fine for reporting issues and requesting features, but other than that, there is his email and a small (half-dead) thread on the CCCP Forums where he first made the project public.

Either way, this is getting side-tracked for this thread when all I wanted was working P016 output from LAV Video for testing purposes... Oh well.

[Pat357]

Kia Ora nevcairiel,
I have encountered an issue when using any FLV splitter and files from www.tvnz.co.nz/video - obviously as it is an NZ site you may not be able to view media streams.

I access this site thru Media Portal and the OnlineVideos plug-in. I am using hoborg's SAF6 (latest pkg using LAV v0.37), unlocked, and the only files I have issues playing are .flv from above mentioned site.

A sample file is here - http://homepages.slingshot.co.nz/~perrin_g/TVNZ_Sample.flv (24MB)

I just played this sample using LAV-splitter, LAV-video and LAV-audio v0.38RC (link was here posted by nev) and it played just fine !
Try updating your LAV to 0.38RC.

[Sebastiii]

I just played this sample using LAV-splitter, LAV-video and LAV-audio v0.38RC (link was here posted by nev) and it played just fine !
Try updating your LAV to 0.38RC.

I can confirm too :) it's working here.

[azaze1]

First let me say this issue may not belong in this thread but truth is I do not know where the problem lies. It's either with MPC-HC, LAV, or the decoding.

For a few months now I've noticed that most of my movies (usually 1080p material) have a periodic issue where a 5 second portion of the movie will suddenly become ultra choppy, and then smooth itself out. It'll happen once, maybe twice during a 2.5 hour movie. This is not a case of a dropped frame every so often, it's a full 5 seconds of nearly every other frame being dropped so it almost looks like a cinematic effect.

This didn't used to happen 6 months ago. I do stay on top of the latest nightlies for MPC-HC, LAV Filters, MadVR. I want to say it never happened before I moved to LAV so that's why I'm posting here. I don't think it ever happened when using DXVA either, but I don't believe it's the software decoding because it happens when using MadVR's internal decoders AND LAV Video decoding.

My setup:

Windows 7 32bit
AMD Radeon HD 6450
Catalyst 11.9 drivers
Latest nightly of:
MPC-HC
LAV Filters 0.37
MadVR 0.77
ReClock 1.8.7.7

Has anyone else seen this behavior even once? If not, is there anything I should check that may cause this?

[SamuriHL]

Try the catalyst 11.4 driver and see if the problem goes away. Just a hunch.

[Plutotype]

First let me say this issue may not belong in this thread but truth is I do not know where the problem lies. It's either with MPC-HC, LAV, or the decoding.

For a few months now I've noticed that most of my movies (usually 1080p material) have a periodic issue where a 5 second portion of the movie will suddenly become ultra choppy, and then smooth itself out. It'll happen once, maybe twice during a 2.5 hour movie. This is not a case of a dropped frame every so often, it's a full 5 seconds of nearly every other frame being dropped so it almost looks like a cinematic effect.

This didn't used to happen 6 months ago. I do stay on top of the latest nightlies for MPC-HC, LAV Filters, MadVR. I want to say it never happened before I moved to LAV so that's why I'm posting here. I don't think it ever happened when using DXVA either, but I don't believe it's the software decoding because it happens when using MadVR's internal decoders AND LAV Video decoding.

My setup:

Windows 7 32bit
AMD Radeon HD 6450
Catalyst 11.9 drivers
Latest nightly of:
MPC-HC
LAV Filters 0.37
MadVR 0.77
ReClock 1.8.7.7

Has anyone else seen this behavior even once? If not, is there anything I should check that may cause this?

My assumption:
This looks like your GPU downclocks during the movie playback and when there is a fast/panned/high bitrate sequence, the downclocked GPU can not process all frames as required. In about 5 seconds the GPU reacts and comes back to full core/memory clock frequencies. Then the movie is smooth again. Altough 11.10 WHQL (http://sites.amd.com/us/game/downloads/Pages/radeon_win7-64.aspx) have ben released couple of minutes ago, I have tested 11.10 RC3 ( should be the same as WHQL basically or very close ) a couple of days ago by increasing the minimum core/memory clock frequencies as I have mentioned in this post (http://forum.doom9.org/showthread.php?p=1534744#post1534744).

This post (http://www.overclock.net/amd-ati/701797-how-fix-flickering-overclocking-2-monitors-5.html#post9279708) describes how to create such gently overclocked profile for AMD Catalyst.

Madshi has also adviced in one of his posts at madVR thread to keep the GPU clocks as stable as it is possible.

[dukey]

Would it make sense to integrate the LAV filters into FFdshow (at least the video ones anyway)? That way subtitle rendering, and all the other great filters it has would work .. just nicely.

[SamuelMaki]

Would it make sense to integrate the LAV filters into FFdshow (at least the video ones anyway)? That way subtitle rendering, and all the other great filters it has would work .. just nicely.

Better idea would be to write own methods and don´t use the old-dated(?) ones that ffdshow uses... At least deinterlacing, deband, avisynth support (if possible) and subtitles. And for the audio filter new DRC, night mode and mixer... Then it would be fantastic:)

[Keiyakusha]

IMO it will be nice to see avisynth support in LAV so you can execute any possible script. And other possible things like deband, sharpen, smoothing, whatever is not needed. yes, this will be feature for advanced users... so what? better to have something for advanced users than nothing for everyone (+ normal users will be able to use it too by reading some guide ) Also looking on avisynth support in ffdshow, I don't think its something that hard to use...

[dukey]

Better idea would be to write own methods and don´t use the old-dated(?) ones that ffdshow uses... At least deinterlacing, deband, avisynth support (if possible) and subtitles. And for the audio filter new DRC, night mode and mixer... Then it would be fantastic:)

why re-invent the wheel ? most off the ffdshow is stable and well optimised :p

[azaze1]

My assumption:
This looks like your GPU downclocks during the movie playback and when there is a fast/panned/high bitrate sequence, the downclocked GPU can not process all frames as required. In about 5 seconds the GPU reacts and comes back to full core/memory clock frequencies. Then the movie is smooth again. Altough 11.10 WHQL (http://sites.amd.com/us/game/downloads/Pages/radeon_win7-64.aspx) have ben released couple of minutes ago, I have tested 11.10 RC3 ( should be the same as WHQL basically or very close ) a couple of days ago by increasing the minimum core/memory clock frequencies as I have mentioned in this post (http://forum.doom9.org/showthread.php?p=1534744#post1534744).

This post (http://www.overclock.net/amd-ati/701797-how-fix-flickering-overclocking-2-monitors-5.html#post9279708) describes how to create such gently overclocked profile for AMD Catalyst.

Madshi has also adviced in one of his posts at madVR thread to keep the GPU clocks as stable as it is possible.

is the GPU core and memory clock even relevant when decoding in software mode? I'm not using DXVA, and the issue occurs whether I'm using MadVR or EVR-CP.

thx for the tip though, I'll give it a shot either way.

[Keiyakusha]

But we don't want to make ffdshow's clone isn't it? Also asking for deband or something... I guess its asking to much from Nev! Avisynth is the middle ground here. And, look at deband in ffdshow, its a crap, and broken in 64bits. The best debanding solutions available in avisynth only (my English is poor enough, I don't know synonyms to the word "best"). What else ffdshow can do that avisynth can not at the same or better quality level?
Also look here, lately there is pretty nice openCL avisynth plugins that works blazingly fast. Very good for realtime playback. This is something Nev probably never integrate in LAV itself...

[cegy]

how can i force lav splitter to always be the splitter i want to use for mpeg.ts (h.264) as each time i get it sorted a few days later it changes back and then am stuck again :( (even happens of fresh install of windows 7 x86/x64)
i mainly use lav splitter to help with encoding speed from record tv shows via dss as i get a faster fps then using dgnv :/

p.s. doesn't matter what version of lav filter i use

[mandarinka]

PS:
Painting subtitles on a 4:2:0 image is still a bad idea, you should convert it to RGB before - 10-bit support doesn't change that.

(My thoughts on this topic)

Only if you care about the subtitle rendering quality *a lot*. It could be argued that the subs aren't an original part of the video and thus aren't really needed to be superb with regards to chroma resolution... Personally, I don't care about the subtitle rendering, the current quality of vsfilter (when rendering at video resolution & on yv12) is okay imho. And anyway - people that do care about sharpness of the font rendering either 1) use mpc renderer to draw in fullscreen resolution 2) use software upscaling prior to vsfilter to achieve the same.

[Plutotype]

is the GPU core and memory clock even relevant when decoding in software mode? I'm not using DXVA, and the issue occurs whether I'm using MadVR or EVR-CP.

thx for the tip though, I'll give it a shot either way.

I also recommend you to create a custom power plan in Windows:

- hard disk - turn off hard disk after: never
- usb selective settings: disabled
- multimedia settings: when playing video: optimize video quality

External drives also use to have proprietary software where standby is defined in minutes or "never".

[clsid]

how can i force lav splitter to always be the splitter i want to use for mpeg.ts (h.264) as each time i get it sorted a few days later it changes back and then am stuck again :( (even happens of fresh install of windows 7 x86/x64)
i mainly use lav splitter to help with encoding speed from record tv shows via dss as i get a faster fps then using dgnv :/

p.s. doesn't matter what version of lav filter i use
You can configure preferred source filters with Codec Tweak Tool (http://majorgeeks.com/K-Lite_Codec_Tweak_Tool_d5738.html).

[golagoda]

how can i force lav splitter to always be the splitter i want to use for mpeg.ts (h.264) as each time i get it sorted a few days later it changes back and then am stuck again :( (even happens of fresh install of windows 7 x86/x64)
i mainly use lav splitter to help with encoding speed from record tv shows via dss as i get a faster fps then using dgnv :/

p.s. doesn't matter what version of lav filter i use

In MPC-HC add the splitters you don't want to use to 'External Filters' and set them to block. When you want them just un-block them. That's what I do at least, there's probably easier ways to do it though.

[cegy]

clsid: i have used codec tweak tool and the other one that they do but it keeps changing back to "system default" and even if lav splitter is set and golagoda: the mpc-hc idea won't work as i can't block something i want know what its using in the first place

this is the outcome of a encoded avi of when the splitter isn't set http://rghost.net/28028481

i even get the same problem when using dgnv (nvidia 450 gts @ 100 fps max!) this is why i use dss as i can get the cpu to do the decoding give me over 60-120% boost faster then dgnv can give me when using my gfx card

[Pat357]

i even get the same problem when using dgnv (nvidia 450 gts @ 100 fps max!) this is why i use dss as i can get the cpu to do the decoding give me over 60-120% boost faster then dgnv can give me when using my gfx card
For h264 you could also try DGAVCdecdi : it also uses the CPU to decode, hence could be much faster if your CPU can handle it.

[azaze1]

is the GPU core and memory clock even relevant when decoding in software mode? I'm not using DXVA, and the issue occurs whether I'm using MadVR or EVR-CP.

Thx for the tip though, I'll give it a shot either way.

Not finding the XML tags for these overdrive sections in my custom profile under catalyst 11.10. A little hesitant to add them.

BTW: Something I should have said is that GPU-Z has always shown the GPU Core and Mem to be at it's rated spec of 625 MHz core, 500 MHz mem when a movie is playing.

As I close the movie it moves down to the modest 100 MHz core, 150 MHz mem within 3 seconds. But again every time a movie is playing in MPC-HC it correctly bumps up the speeds to the "normal" values. Not sure this minimum speed tweak is going to do anything for me if it's always at 625/500 when playing movies.

[azaze1]

I also recommend you to create a custom power plan in Windows:

- hard disk - turn off hard disk after: never
- usb selective settings: disabled
- multimedia settings: when playing video: optimize video quality

External drives also use to have proprietary software where standby is defined in minutes or "never".

Thanks,

I've got all those settings in my power plan already. My media is being accessed via gigabit network. HTPC is just a 64GB SSD with windows and media software.

[ruinevil]

First let me say this issue may not belong in this thread but truth is I do not know where the problem lies. It's either with MPC-HC, LAV, or the decoding.

For a few months now I've noticed that most of my movies (usually 1080p material) have a periodic issue where a 5 second portion of the movie will suddenly become ultra choppy, and then smooth itself out. It'll happen once, maybe twice during a 2.5 hour movie. This is not a case of a dropped frame every so often, it's a full 5 seconds of nearly every other frame being dropped so it almost looks like a cinematic effect.

This didn't used to happen 6 months ago. I do stay on top of the latest nightlies for MPC-HC, LAV Filters, MadVR. I want to say it never happened before I moved to LAV so that's why I'm posting here. I don't think it ever happened when using DXVA either, but I don't believe it's the software decoding because it happens when using MadVR's internal decoders AND LAV Video decoding.

My setup:

Windows 7 32bit
AMD Radeon HD 6450
Catalyst 11.9 drivers
Latest nightly of:
MPC-HC
LAV Filters 0.37
MadVR 0.77
ReClock 1.8.7.7

Has anyone else seen this behavior even once? If not, is there anything I should check that may cause this?

It's probably Reclock, it's an extremely tempermental piece of software. I had issues like that with the Auto playback speed setting, though I mostly noticed them in the beginning of the file. I force Reclock to play at original speed now. Though before you do that, try uninstalling and reinstalling MPC-HC, keeping the settings intact.

[Andy o]

It's only "temperamental" if you don't know exactly what it's doing. It can clash with other settings of players or renderers, so you must know what to enable or disable. That's why it was recommended to disable VSync when using with madVR, for instance.

[THX-UltraII]

I own a MSI GTX 460 1gb OC edition and use CUVID with it. Will it make any sense (will I benefit) to overclock my card?

[cruse]

First thing I just noticed is that madVR reports a display refrash rate of 23,971xx (NVIDIA 23Hz setting) while madVR reported a display rate of 23,975xx which is far more close to EXACTLY 23,976 which it should be ideally.

Do you guys also get 23,971xx reported by madVR and if so, will this result in far more dropped frames during a movie? With my ATI 5450 I got about 1-3 dropped frames during a 2-3 hour movie which was pretty satisfying to me.

I recently switched from an ATI 5750 to an NVIDIA GTS450, and my ATI came closer to 23.976 as well (though, still not close enough). The easiest solution is to just use Reclock, and go with 24hz, which both cards come much closer to.

gee had to wait 5 days to post this...

For NVIDIA cards you have to create a custom resolution to get exactly 23.976Hz/FPS.

My NVIDIA system settings are in german, so I don't know what the options are called in English but go to

Change Resolution -> button below resolution list -> Create User Specified Resolution

In the upper Display Mode box (1920x1080) select 23 Hz as frequency and progressive, expand Timings below, set Standard from automatic to manual and make sure those are the according numbers:

active pixels horizontal: 1920 vertical: 1080
front porch horizontal: 638 vertical: 4
sync width horizontal: 44 vertical: 5
total pixels horizontal: 2750 vertical: 1124
polarity horizontal: positive vertical: positive
refresh rate: 23.976

Test, save and make sure the new resolution is active. The tray icon should offer 23 Hz now with right click (although it lists the new resolution as 24 Hz in user specified res list).

Good Luck