LAV Filters - DirectShow Media Splitter and Decoders

[oguzpasa]

guyz im a newbie. so i dont know how to install that patched ffmpeg versions in http://git.1f0.de/gitweb?p=ffmpeg.git;a=summary
anybody can help?

[Mikey2]

Does the "Mixing" functionality correctly "down-mix" 7.1 (or 6.1 (DTS-ES)) sources to my 5.1 speaker setup?

Originally I thought that this was the case, but imouto.my's configuration guide suggests that it needs to be done in ffdshow's mixer to properly mix 7.1 down to 5.1 channels. (Unfortunately this is difficult, if not impossible, to objectively test this functionality.)

Also, now that I am on the subject, doesn't it make more sense to put the "Convert Output...", "Expand Mono to Stereo", and "Expand 6.1 to 7.1" options to be on the "Mixing" tab? (I almost forgot about those settings since I am messing around with the Mixing options...)

Finally, is "LFE Mix Level" the equivalent of "LFE Crossover" in ffdshow? If so, what is the crossover frequency?

[blackjack12]

Originally Posted by madshi

@blackjack12, what happens if you force deinterlacing off (press Ctrl+Alt+Shift+D multiple times) in madVR? Does that solve all problems?

Madshi,

As noted before and sorry for taking so long ...

The answer is yes ...

With deinterlacing set and done with MadVR, the MPEG2 files that are really progressive that for whatever reason indicate interlaced at the MadVR renderer, when you force deinterlacing off as you suggest ... all turns good ...

Hope that you and Nev can figure this out.

[nevcairiel]

Originally Posted by Mikey2

Originally I thought that this was the case, but imouto.my's configuration guide suggests that it needs to be done in ffdshow's mixer to properly mix 7.1 down to 5.1 channels. (Unfortunately this is difficult, if not impossible, to objectively test this functionality.)

The guide is old. LAVs new mixing functionality should work just fine when downmixing 6.1/7.1 to 5.1

Originally Posted by Mikey2

Also, now that I am on the subject, doesn't it make more sense to put the "Convert Output...", "Expand Mono to Stereo", and "Expand 6.1 to 7.1" options to be on the "Mixing" tab? (I almost forgot about those settings since I am messing around with the Mixing options...)

Probably, should restructure it some day.

Originally Posted by Mikey2

Finally, is "LFE Mix Level" the equivalent of "LFE Crossover" in ffdshow? If so, what is the crossover frequency?

No, its not LFE Crossover. It just controls if (and how strong) an existing LFE channel is mixed into the front channels when doing downmix to stereo, it does not affect anything if you downmix to 5.1 or any other format with LFE.

[madshi]

@blackjack12, thanks for the detailed reports. I appreciate it. I am a bit confused now, though, because your reports #12001 and #12006 and #12009 seem to contradict each other. So is the final conclusion now that everything is alright as long as you have deinterlacing forced off in madVR? But then I don't understand the screenshots from your #12001 report because there deinterlacing *IS* in fact turned off for all the "BAD" screenshots? So how does that fit to your report #12009?

If the incorrectly turned on deinterlacing is the cause of all the trouble, does that apply to the VC-1 and h264 problems, too?

Try changing the madExcept\processing\deinterlacing option to "if in doubt, deactivate deinterlacing". This should make madVR behave more similar to EVR.

So there are really 2 separate problems here:

(1) madVR activates deinterlacing although it isn't needed. This depends on your madVR settings. As explained above, changing them to "if in doubt, deactivate deinterlacing" seems to take care of the problem. Please understand, though, that there are broadcasts out there which need deinterlacing although they appear to be encoded progressively. That's why madVR has this setting in the first place. EVR plays such files with interlaced artifacts, madVR plays them fine (at least when using its own IVTC algorithm).

(2) I can reproduce the interlacing artifacts with my Radeon 3850. This is clearly a fault of the AMD DXVA deinterlacing. It should perform IVTC, but it does not. You can "fix" this by forcing madVR into "film" mode (either in the settings or by pressing Ctrl+Alt+Shift+T). That nicely solves the problem with the "Good Year" sample, too. Of course forcing madVR into film mode won't work for natively interlaced content (music concerts, sports etc).

[oguzpasa]

Originally Posted by oguzpasa

guyz im a newbie. so i dont know how to install that patched ffmpeg versions in http://git.1f0.de/gitweb?p=ffmpeg.git;a=summary
anybody can help?

any help???

[noee]

install git client, then clone the ffmpeg repo into your Lavfilters repo

#!/bin/bash
# cd lavfilters
git clone git://git.1f0.de/ffmpeg.git ffmpeg

[roytam1]

Originally Posted by noee

install git client, then clone the ffmpeg repo into your Lavfilters repo

#!/bin/bash
# cd lavfilters
git clone git://git.1f0.de/ffmpeg.git ffmpeg

use "git submodule init" to setup the sub-directories and use "git submodule update" to check them in.

[Deim0s]

nevcairiel,
When mixing in stereo, LAV Audio mixer uses an incorrect coefficient for LFE Mix Level.
Applied in the mixer LAV Audio coefficient 1.00, actually equals 0.7071.
File 5.1 mixes on a matrix:
Lt = 100*L + 0.7071*C + 0.7071*LF + 100*SL
Rt = 100*R + 0.7071*C + 0.7071*LF + 100*SR
http://i.imgur.com/wfYvg.png
Equal to the file, mixed in with the parameters of LAV Audio:
(GraphStudioNext -> LAVSplitter -> LAVAudio -> Dump)
Center Mix Level: 0.7071
Surround Mix Level: 1.00
LFE Mix Level: 1.00
http://i.imgur.com/y14CN.png

Test file from here:
http://www.microsoft.com/windows/windowsmedia/howto/articles/Multichannel.aspx

[nevcairiel]

Thats normal. The LFE channel is duplicated, from originally one channel to two channels, and this requires a reduction in volume by 3dB. This happens automatically for all channels that are duplicated when mixing them.

[Deim0s]

nevcairiel,
Yes, but the center, when selected in the LAV Audio coefficient of 0.7071, mixes, as planned, with a 0.7071.
For the LFE channel, to get a real coefficient 0.7071, you need to select the mixer LAV Audio coefficient 1.00 (using the dump file and audio editing).
It is possible to do, that would be in the mixer LAV Audio displayed the real coefficients that will be mixing?

[mkanet]

@Nev/@blackjack12

I'm curious if this issue is related to the issue I'm having for 1080i TV material (HBO, SHO, AMC, and other premium TV channels). This is where certain scenes in the video (usually during panning or movement of the camera back and forth will cause the video to stutter for a second or so).

This bug described above is well known in the MS Media Center community that has access to US premium digital cable TV content. What makes this content different than other digital TV recordings is they constantly switch between hard and soft telecine. This bug commonly referred to as the "29/59 bug" where (for whatever reason) the PC cant switch to soft telecine seamlessly.

I recently upgraded from a GT545 to a GT640; hoping the GPU would completely handle soft-telecine and hard-telecine switching smoothly. While the GT640 did significantly reduce the stutter, it's still visible enough to be distracting.

Unfortunately, madvr is not an option for me when watching TV content since my Media Center solution uses a custom EVR presenter which cant be changed; especially since it integrates a full blown graphics, menus, extra movie info onto the screen when necessary.

Originally Posted by blackjack12

While I love the LAV filters in general, have found that the deinterlacing performance of the LAV Video Decoder is inconsistent at best and poor at worst across a general mix of video types.

It is worst in allowing the AMD Radeon GPU’s to hardware deinterlace across all video formats. The hardware deinterlacing of the AMD cards appears in my tests to be the best across the board when it is used but the NVidia can produce very good results with LAV Video and CUVID.

Basic Testing:
Windows 7/Window 8 Preview (64 bit)
MPC-HC 1.6.3.5818 (32 bit) – some testing with 64 bit MPC-HC but not much and not focus (for 64 bit, EVR-CP used, no MadVR, no ArcSoft filters)
MadVR 82.5 (with and w/o latest ringing mods)
NVidia 560Ti and GT525M
NVidia Drivers 301.42 and 302.82 (Win 8) – LAV CUVID
Radeon HD 4890 and 6570
AMD Drivers 12.6 – LAV None, copy-back and native
LAV 51.3(also tested latest builds to this date)
LAV Splitter and LAV Audio used in all testing
ArcSoft Video Decoder 2.28.444.129 X86

All combinations that can be used with LAV and MadVR were looked at. Aggressive, frame rate, mode, software, etc. and details can be provided if needed but the test files are best way to look at. These were on 4 “desktop” type units and one laptop.

Using the ArcSoft Video Decoder provided the best and most consistent deinterlacing performance in my tests while the NVidia produced good to superb results depending on the video formats tested. Using the de-interlacing in MadVR worked with NVidia but found the internal GPU only with LAV Video and passing to the renderer yielded better results. For the AMD GPU’s tested the MadVR de-interlacing did not work on several files. Only software (YADIF) de-interlacing worked as noted briefly by someone earlier.

File clips that are good to test de-interlacing can be found in the Google Drive link listed below and downloaded for testing. These can be used to easily see issues quickly. All files are .mkv types prepared from the native and purchased video material using MakeMKV and then remuxing with the latest version of MKVMerge/Toolnix. The audio has been removed just to keep file size as small as possible. Watching the rolling credits is a great test of deinterlacing performance and quality in many cases.

Problematic file types but inconsistent: MPEG2i (DVD, check the rolling credits for large visual differences), VC-1i, H264i (AVC and all produced with Premiere Pro from and with interlaced content)

Google Drive link for excellent examples (they must be downloaded, viewing the preview in the YouTube format is useless):
https://docs.google.com/folder/d/0B_...RpN2dsOVU/edit

Hope these are helpful in testing and optimizing deinterlacing performance with the LAV Video Filters in the future.

[blackjack12]

Originally Posted by madshi

@blackjack12, thanks for the detailed reports. I appreciate it. I am a bit confused now, though, because your reports #12001 and #12006 and #12009 seem to contradict each other. So is the final conclusion now that everything is alright as long as you have deinterlacing forced off in madVR? But then I don't understand the screenshots from your #12001 report because there deinterlacing *IS* in fact turned off for all the "BAD" screenshots? So how does that fit to your report #12009?

If the incorrectly turned on deinterlacing is the cause of all the trouble, does that apply to the VC-1 and h264 problems, too?

Try changing the madExcept\processing\deinterlacing option to "if in doubt, deactivate deinterlacing". This should make madVR behave more similar to EVR.

So there are really 2 separate problems here:

(1) madVR activates deinterlacing although it isn't needed. This depends on your madVR settings. As explained above, changing them to "if in doubt, deactivate deinterlacing" seems to take care of the problem. Please understand, though, that there are broadcasts out there which need deinterlacing although they appear to be encoded progressively. That's why madVR has this setting in the first place. EVR plays such files with interlaced artifacts, madVR plays them fine (at least when using its own IVTC algorithm).

(2) I can reproduce the interlacing artifacts with my Radeon 3850. This is clearly a fault of the AMD DXVA deinterlacing. It should perform IVTC, but it does not. You can "fix" this by forcing madVR into "film" mode (either in the settings or by pressing Ctrl+Alt+Shift+T). That nicely solves the problem with the "Good Year" sample, too. Of course forcing madVR into film mode won't work for natively interlaced content (music concerts, sports etc).

Madshi,

First it is good that you were finally able to duplicate what I have seen for a long time and continue to see across 5 systems with a range of CPU’s and GPU’s.

Frankly shocked that nobody has reported before because the biggest issues continue to be with standard DVD ripped material (not illegally downloaded) that has been directly converted using MakeMKV. This retains the original MPEG2 video material of which the majority is progressive. I am seeing issues with every configuration from the older Radeon 4890 to newer NVidia 560Ti and GT525M

There are no contradictions, just 2 situations.

1. Using LAV and GPU only for deinterlacing – deinterlacing is turned off with MadVR.
2. Using deinterlacing done with MadVR.

The latest post was the first tests using the deinterlacing capabilities with MadVR.

To summarize findings with the Radeon 4890 GPU using MadVR deinterlacing.
(Radeon 6570 appears similar but it is a newer DirectX11 capable card with the newest non-legacy drivers)

If set the MadVR dieinterlacing to “if in doubt deactivate” all appears good … But … with the “Good Year” sample, the deinterlacing turns on again after about 10 seconds! … Just as reported in post 12004 by sneaker_ger. I will look into the sample and original for any issues.

Have also now found a number of other DVD ripped samples that are reporting as interlaced. They actually show interlaced in media info as well. They look bad with the Radeon processors and MadVR deinterlacing. Better with the NVidia/CUVID and will report on later.

Converting everything with Handbrake solves all the problems when content is converted to H264 progressive, but lose some resolution.

Will let you know what I find with the NVidia cards and systems as well. They showed the same issues.

The bottom line is if/when MadVR has deinterlacing set correctly, the results are superb... but how do we get that to happen with a single setting that is good across the board?

When I find a "good" settings list or other issues will post all setup screens.

Note: Is there any way that it is/or would be possible to set the MadVR parameters “offline” without having to have a video running.


QUICK UPDATE for Radeon 6570 systems using MadVR deinterlacing - "if in doubt deactivate deinterlacing" checked:

File formats tested with a lot of different materials.
H264 progressive - GOOD
H264i interlaced - GOOD
VC-1 progressive - GOOD
VC-1i interlaced - GOOD
MPEG2 progressive - GOOD
MPEG2i interlaced - BAD

MPEG2i - This is native/true interlaced MPEG2 content from a MakeMKV DVD rip. With MadVR, interlacing it is BAD. If force off and let LAV and hardware only deinterlace looks great. Preliminary same conclusions with NVidia but the MPEG2i does not look as bad ... will post some new samples and screen shots later.

Looks like all good if use "if in doubt deactivate deinterlacing" checked ... except MPEG2i content. It appears that MadVR deinterlacing is not handling the MPEG2i content correctly or well. If it is turned off and the hardware only is used to deinterlace it looks great. Looks like a conflict or "double" deinterlace type effects, really don't know.

[Pat357]

Originally Posted by Deim0s

nevcairiel,
Yes, but the center, when selected in the LAV Audio coefficient of 0.7071, mixes, as planned, with a 0.7071.
For the LFE channel, to get a real coefficient 0.7071, you need to select the mixer LAV Audio coefficient 1.00 (using the dump file and audio editing).
It is possible to do, that would be in the mixer LAV Audio displayed the real coefficients that will be mixing?

Why would the "LFE" channel need to be handled differently than the "CENTER" ? Both have 1 source in a 5.1/7.1 composition
and after down-mixing to stereo they have both 2 sources.
This explains the 0.7071 coefficient to preserve the total power [W] from each channel in the new setup.

[Deim0s]

Pat357,
Sorry for my bad English.
I mean, the selected coefficient for LFE channel in the mixer LAV Audio, does not correspond with the real coefficient that is mixing.
For example, to mix the LFE channel with real coefficient 0.7071, the mixer LAV Audio must selected 1.00.

[mkanet]

Nev, I just wanted you to know that the "Remove Soft-telecine" option doesnt have any affect on the frame rate reported by the decoder (when viewing in Reclock). I am using aggressive deinterlace, adaptive algorithm, high-quality processing)

I have tried both half checked, and full checked option. They both do exactly the same thing as older versions without this option.

I can see framerate changing from 59.940 to 23.976 (back and forth) throughout each 1080i TV recording.

It's possible I misunderstood what this feature does; but, if it did what I was hoping for, it would definitely solve the issue I have with videos that cause the decoder to decoder to switch back and forth between 59.940 and 23.976.

Originally Posted by nevcairiel

Here is another version, i actually added an option on the options panel to control it now.

http://files.1f0.de/lavf/LAVFilters-...6-g76de254.zip

The "Remove Soft-Telecine" checkbox is the new one, and it has 3 states.
- Unchecked it behaves as any previous versions, it just sends the soft-telecine flags to the renderer
- Checked it removes all flags and adjusts the timestamp to 23.976
- In the "intermediate" state, it only removes the flags, but leaves the timestamps alone.

It would be great if you could test both the checked and the "half checked" mode, and see how they behave on your setup. I'm mostly interested to be able to determine which mode should be the default.
Specifically, i'm interested if the intermediate mode is enough to make it play properly, because its much safer to do then to try to fiddle with the timestamps.

Additionally, this version also includes a small fix for the "Travis" sample, it might work better now.

[madshi]

Originally Posted by blackjack12

Frankly shocked that nobody has reported before because the biggest issues continue to be with standard DVD ripped material (not illegally downloaded) that has been directly converted using MakeMKV. This retains the original MPEG2 video material of which the majority is progressive. I am seeing issues with every configuration from the older Radeon 4890 to newer NVidia 560Ti and GT525M

FWIW, the madVR default setting is "if in doubt deactivate deinterlacing". I'm not sure why you had it set differently in your initial tests.

Originally Posted by blackjack12

There are no contradictions, just 2 situations.

1. Using LAV and GPU only for deinterlacing – deinterlacing is turned off with MadVR.
2. Using deinterlacing done with MadVR.

But if you do the deinterlacing in LAV then it shouldn't matter which renderer you use. The result should be the same with every renderer because all renderers will get progressive images from LAV and will never know that the original content was interlaced.

Originally Posted by blackjack12

If set the MadVR dieinterlacing to “if in doubt deactivate” all appears good … But … with the “Good Year” sample, the deinterlacing turns on again after about 10 seconds! … Just as reported in post 12004 by sneaker_ger. I will look into the sample and original for any issues.

With "if in doubt deactivate", madVR switches deinterlacing on in the first moment an interlaced frame is sent from the decoder. This seems to be after 10 seconds with the "Good Year" sample. Probably EVR enables/disables deinterlacing on the fly, depending on which frame type is sent by the decoder. I could do that, too, in madVR, but I don't think it's the right way to handle the situation. I believe DXVA deinterlacing should be capable of handling such content. It's just too bad AMD fails on this.

Originally Posted by blackjack12

Have also now found a number of other DVD ripped samples that are reporting as interlaced. They actually show interlaced in media info as well. They look bad with the Radeon processors and MadVR deinterlacing. Better with the NVidia/CUVID and will report on later.

Which of the 3 problems is it? See bottom of my post.

Originally Posted by blackjack12

The bottom line is if/when MadVR has deinterlacing set correctly, the results are superb... but how do we get that to happen with a single setting that is good across the board?

The long term plan is to add an automatic video vs film vs progressive detection to madVR which decides based on content. That should allow correct results in all situations with all decoders, even if the files are encoded in a weird way. Currently every solution is nothing but a compromise. What madVR does is a compromise, what EVR does is a compromise, too.

Originally Posted by blackjack12

Note: Is there any way that it is/or would be possible to set the MadVR parameters “offline” without having to have a video running.

Not yet. On my to do list (like so many other things).

Originally Posted by blackjack12

QUICK UPDATE for Radeon 6570 systems using MadVR deinterlacing - "if in doubt deactivate deinterlacing" checked:

File formats tested with a lot of different materials.
H264 progressive - GOOD
H264i interlaced - GOOD
VC-1 progressive - GOOD
VC-1i interlaced - GOOD
MPEG2 progressive - GOOD
MPEG2i interlaced - BAD

Which of the 3 problems is it? See:

-------

Generally, if you report problems with specific samples with LAV + madVR, please always check if madVR reports deinterlacing to be on/off, and whether overriding that decision by pressing Ctrl+Alt+Shift+D fixes things. You would noticably increase the usefulness of your reports if you mentioned right away what the real problem is. There are 3 basic possibilities:

(1) Image quality is bad with deinterlacing off, but good with deinterlacing on. The problem is that it's not turned on automatically.
(2) Image quality is bad with deinterlacing on, but good with deinterlacing off. The problem is that it's turned on automatically.
(3) Image quality is bad in any case, no matter if deinterlacing is activated or not.

For every problem you see, it would be nice if you could specify which of the 3 problems we're talking about. Also, it would be helpful to know if EVR plays the same file correctly with the same decoder settings.

[blackjack12]

Originally Posted by madshi

Which of the 3 problems is it? See:

-------

Generally, if you report problems with specific samples with LAV + madVR, please always check if madVR reports deinterlacing to be on/off, and whether overriding that decision by pressing Ctrl+Alt+Shift+D fixes things. You would noticably increase the usefulness of your reports if you mentioned right away what the real problem is. There are 3 basic possibilities:

(1) Image quality is bad with deinterlacing off, but good with deinterlacing on. The problem is that it's not turned on automatically.
(2) Image quality is bad with deinterlacing on, but good with deinterlacing off. The problem is that it's turned on automatically.
(3) Image quality is bad in any case, no matter if deinterlacing is activated or not.

For every problem you see, it would be nice if you could specify which of the 3 problems we're talking about. Also, it would be helpful to know if EVR plays the same file correctly with the same decoder settings.

Will do ... will post good and short samples for all file types that I know can be problematic so you can use for testing as well. You only need a new MPEG2i sample which is now the problem child with Radeon GPU. (Prelim result - think is okay with NVidia)

One quick question. With the AMD Radeon driver, just want to confirm how it should be setup for best results? As reported, I have always had auto deinterlace on with pull down and no other post processing.

Nev ... I think I sent you on a somewhat diverted goose chase and apologize. But really want LAV and MadVR to work flawlessly together for everything ... the results are superb when they all work ...

[madshi]

Yes, in the AMD drivers, activate all those funny deinterlacing options, including pulldown, IVTC and whatever you find there. But disable most other funny stuff like noise reduction, detail enhancement etc, unless you're a fan of such algorithms. Samples for the troublesome videos would be nice, if they play fine with LAV + EVR. If this seems to be a madVR problem, though, it might make sense to move it to the madVR thread. Unless the problem only occurs with LAV + madVR, then I'm not sure where to discuss it.

[blackjack12]

Originally Posted by madshi

Yes, in the AMD drivers, activate all those funny deinterlacing options, including pulldown, IVTC and whatever you find there. But disable most other funny stuff like noise reduction, detail enhancement etc, unless you're a fan of such algorithms. Samples for the troublesome videos would be nice, if they play fine with LAV + EVR. If this seems to be a madVR problem, though, it might make sense to move it to the madVR thread. Unless the problem only occurs with LAV + madVR, then I'm not sure where to discuss it.

Madshi,

Will move to MadVR forum.

Will update on AMD Radeon MPEG2i interlaced issues after update to the latest 12.8 Catalyst driver that was released a couple of days ago.

Definite issues with MPEG2i content that is corrected when forcing interlacing off in MadVR. The material is interlaced but MadVR is not deinterlacing correctly or there is some kind of conflict. You can literally see frame jump when forcing MadVR deinterlace off. ...Then all is perfect. (again AMD Radeon only)