LAV Filters - DirectShow Media Splitter and Decoders

[v0lt]

Hi,

I put together a small test version.
I remember someone asking about speex support, and because i put together a cross-compiling environment today, i thought i would just build the speex library and see if it worked.
...

http://samples.mplayerhq.hu/A-codecs/speex/speex_openacm.avi
LAV Audio decoder does not work with system AVI splitter.

[THX-UltraII]

I have some questions about the configuration settings of my JVC RS45 projector with LAV filters+MPC-HC.

1. HDMI Input
I can select a. Standard (16-235), b. Enhanced (0-255) or c. Super White (16-255) in the JVC RS45 projector menu.
What is the setting that should be used when using a HTPC with:
- Blu-Ray content to play
- Media Player Classic Home Cinema
- madVR as output renderer (set to PC levels 0-255 as RGB output levels)
- LAV video filter (RGB set to 'untouched' in LAV video filter)
- NVIDIA GTX460 with default 1080p23 profile BUT with n vidia fix during installation (for further information about this fix check http://forums.nvidia.com/index.php?showtopic=216099&view=findpost&p=1328716)

2. Color Space
I can select YCbCr (4:4:4), YCbCr (4:2:2) or RGB. It seems that only RGB seems to show the correct colours. I thought Blu-Ray specification was 4:2:2?
no ideas anyone?

[Kurtnoise]

I put together a small test version.
I remember someone asking about speex support
why not using speex decoder from libav ?

[nevcairiel]

why not using speex decoder from libav ?

It doesn't have one. It only has support for libspeex, which i use now.

[aufkrawall]

I have some questions about the configuration settings of my JVC RS45 projector with LAV filters+MPC-HC.

1. HDMI Input
I can select a. Standard (16-235), b. Enhanced (0-255) or c. Super White (16-255) in the JVC RS45 projector menu.
What is the setting that should be used when using a HTPC with:
- Blu-Ray content to play
- Media Player Classic Home Cinema
- madVR as output renderer (set to PC levels 0-255 as RGB output levels)
- LAV video filter (RGB set to 'untouched' in LAV video filter)
- NVIDIA GTX460 with default 1080p23 profile BUT with n vidia fix during installation (for further information about this fix check http://forums.nvidia.com/index.php?showtopic=216099&view=findpost&p=1328716)

2. Color Space
I can select YCbCr (4:4:4), YCbCr (4:2:2) or RGB. It seems that only RGB seems to show the correct colours. I thought Blu-Ray specification was 4:2:2?
madVR should output RGB to your projector, that could explain why only RGB looks right.
Most BD videos are still 4:2:0 but it doesn't matter since it gets converted to RGB anyway.

[dukey]

nevcairiel, been looking at this VC1 thing for the bd source filter. Looking at LAV, it looks like you are only using an allocator of 256000 bytes or 0.25~ meg for VC1. If for VC1 entire frames have to be sent to the decoder, does that mean it might just die if frames are over that size? I can see key frames maybe being larger .. ? Unless I am missing some critical piece of understanding.

[nevcairiel]

Unless I am missing some critical piece of understanding.

You are.
There is no fixed packet size, the allocator just has some dummy value set, but since the source filter controls the allocator, it'll just have the packets matching the frame size perfectly, no extra allocation is done. Every packet has a different size.

For no VBR format you can use a static packet size, unless you specify it to be way huge, which isn't a good idea. However you have to tell the allocator some value in that allocator properties structure, could as well be "1" or something. :p
Most other splitters use a static size, and just increase that static size once a frame exceeds it. I used to do that as well, but implemented a better method recently.

[dukey]

Alright, thanks :)

[THX-UltraII]

madVR should output RGB to your projector, that could explain why only RGB looks right.
Most BD videos are still 4:2:0 but it doesn't matter since it gets converted to RGB anyway.

thanks. And HDMI?

1. HDMI Input
I can select a. Standard (16-235), b. Enhanced (0-255) or c. Super White (16-255) in the JVC RS45 projector menu.
What is the setting that should be used when using a HTPC with:
- Blu-Ray content to play
- Media Player Classic Home Cinema
- madVR as output renderer (set to PC levels 0-255 as RGB output levels)
- LAV video filter (RGB set to 'untouched' in LAV video filter)
- NVIDIA GTX460 with default 1080p23 profile BUT with n vidia fix during installation (for further information about this fix check http://forums.nvidia.com/index.php?s...post&p=1328716)

[mastan]

Out of interest I've run lavfilters sources through CppCheck and found these warnings:

decoder/LAVAudio/Media.cpp:274
incorrectLogicOperator : style : Expression always evaluates to false. Did you intend to use || instead?
if (num_channels < 1 && num_channels > 8)
return 0;

demuxer/LAVSplitter/PacketAllocator.cpp:36
duplicateBreak : style : The second of the two statements can never be executed, and so should be removed

return GetInterface((ILAVMediaSample *) this, ppv);
return S_OK;

demuxer/Demuxers/LAVFDemuxer.cpp:74
uninitVar : style : Member variable 'CLAVFDemuxer::m_pBluRay' is not initialized in the constructor.

And some other "not initialized in the constructor" warnings, which are too many to describe here.

There were other minor things like assignments to a variable that is never used, like ret in CBDDemuxer::Open(), but that's not that important.

Hope it helps!

Maybe use static code chekers sometimes?

[nevcairiel]

thanks. And HDMI?


Please stop spamming 3 different forums with your questions.
This is not a question related directly to LAV.

Maybe use static code chekers sometimes?

Static code analyzers are 99% useless, they usually produce more false positives then useful info, so filtering through those is more annoying then useful.
Especially cppcheck is really bad.

Especially the constructor things, not everything needs to be initialized in the constructor if you're conscious when you use it, and such things are just not caught by static analyzers.

[Pat357]

Nev,

This is what the MPC-HC AVI-splitter gives on a file with speex-audio :



MPC AVI Source::Audio 1

Media Type 0:
--------------------------
Audio: 0xa109 32000Hz stereo 44kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Audio {73647561-0000-0010-8000-00AA00389B71}
subtype: Unknown GUID Name {0000A109-0000-0010-8000-00AA00389B71}
formattype: FORMAT_WaveFormatEx {05589F81-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 4207
cbFormat: 132

WAVEFORMATEX:
wFormatTag: 0xa109
nChannels: 2
nSamplesPerSec: 32000
nAvgBytesPerSec: 5610
nBlockAlign: 561
wBitsPerSample: 10
cbSize: 114 (extra bytes)

pbFormat:
0000: 09 a1 02 00 00 7d 00 00 ea 15 00 00 31 02 0a 00 .¡...}..ê...1...
0010: 72 00|01 00 53 70 65 65 78 20 20 20 73 70 65 65 r...Speex spee
0020: 78 2d 31 2e 30 00 00 00 00 00 00 00 00 00 00 00 x-1.0...........
0030: 01 00 00 00 50 00 00 00 00 7d 00 00 02 00 00 00 ....P....}......
0040: 04 00 00 00 02 00 00 00 ff ff ff ff 80 02 00 00 ........ÿÿÿÿ€...
0050: 00 00 00 00 05 00 00 00 00 00 00 00 00 00 00 00 ................
0060: 00 00 00 00 09 43 6f 64 65 63 20 68 6f 6d 65 70 .....Codec homep
0070: 61 67 65 20 77 77 77 2e 6f 70 65 6e 61 63 6d 2e age www.openacm.
0080: 6f 72 67 20

Here's another file (speex in MKV) with the Haali splitter :


Haali Media Source::Audio

Media Type 0:
--------------------------
Audio: 0xa109 32000Hz stereo 30kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Audio {73647561-0000-0010-8000-00AA00389B71}
subtype: Unknown GUID Name {0000A109-0000-0010-8000-00AA00389B71}
formattype: FORMAT_WaveFormatEx {05589F81-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 0
bTemporalCompression: 0
lSampleSize: 1
cbFormat: 98

WAVEFORMATEX:
wFormatTag: 0xa109
nChannels: 2
nSamplesPerSec: 32000
nAvgBytesPerSec: 3800
nBlockAlign: 4
wBitsPerSample: 16
cbSize: 80 (extra bytes)

pbFormat:
0000: 09 a1 02 00 00 7d 00 00 d8 0e 00 00 04 00 10 00 .¡...}..Ø.......
0010: 50 00|53 70 65 65 78 20 20 20 31 2e 32 72 63 31 P.Speex 1.2rc1
0020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 01 00 ................
0030: 00 00 50 00 00 00 00 7d 00 00 02 00 00 00 04 00 ..P....}........
0040: 00 00 02 00 00 00 a0 73 00 00 80 02 00 00 00 00 ......*s..€.....
0050: 00 00 01 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
0060: 00 00
If you need more info, just let me know.

[egur]

Static code analyzers are 99% useless, they usually produce more false positives then useful info, so filtering through those is more annoying then useful.

I can't agree with that, a properly set static analysis tool can do a lot of good and give very little false positives.
CppCheck is a simple tool but there are others which are better.

A good practice is to raise compiler warnings to level 4 on a regular basis and prior to release set it to 'EnableAllWarning' and use static analysis (whatever you can afford).
In gcc, raise warning level to -Wall. which is similar to level 4 in VS.

This way new issues are quickly found as the code is always clean.

[nevcairiel]

Compiler Warnings are one thing, but static analysis can't keep up with some more advanced constructs.
Its a tool that can be used occasionally, but its nearly useless if you run it on code that you don't know and take the messages at face value.

Although i would never use CppCheck for it.

[JEEB]

Compiler Warnings are one thing, but static analysis can't keep up with some more advanced constructs.
I like MSVC's and Clang's static analyzers that go through compiled output together with the source, even if there would be some false positives.

Purely source-based analyzers such as cppcheck can of course only go as far...

[aufkrawall]

thanks. And HDMI?

1. HDMI Input
I can select a. Standard (16-235), b. Enhanced (0-255) or c. Super White (16-255) in the JVC RS45 projector menu.
What is the setting that should be used when using a HTPC with:
- Blu-Ray content to play
- Media Player Classic Home Cinema
- madVR as output renderer (set to PC levels 0-255 as RGB output levels)
- LAV video filter (RGB set to 'untouched' in LAV video filter)
- NVIDIA GTX460 with default 1080p23 profile BUT with n vidia fix during installation (for further information about this fix check http://forums.nvidia.com/index.php?s...post&p=1328716)
You've modified your display driver inf to use full range, so 0-255 should be right.
I don't have any HDMI experience, but simply using what works/looks right would have answered your questions too, I guess. ;)

Let us stop trying out nev's bright mood. :sly:

[nevcairiel]

Note for everyone doing their own LAV builds!

Yesterday i added libspeex to my build, which required me to add a new binary dependency on libspeex itself.
At first it was included in the repository, but that was not a good idea, so i removed it again.

To build LAVs ffmpeg now, at least with my build scripts, you'll need to install libspeex into your local MSYS/MinGW build environment.

You can get the pre-compiled libraries here:
http://files.1f0.de/mingw/

I'll most likely add 1-2 more libraries in the future, which i'll all place there.
I may even produce my own build-env setup, as i'm not too happy with the current one anymore.

Edit:
I reverted those changes until i can provide a proper complete build env, too much trouble otherwise.

[SamuriHL]

A build env setup would ROCK. I like to maintain my own custom installer to update J River's private copy of LAV so having the build env auto setup would be nice. Do we just drop these in and we're good? Thanks!

[nevcairiel]

Just unpack them in the folders as outlined on the downloads page, and it should work.

[SamuriHL]

Ok, thanks!

[SamuriHL]

You changed the install script on me? Cause I can't get the files to register now. :p I guess you must be including something new that I don't have in my install script modification.

[nevcairiel]

Actually that was my fault, the libspeex package i uploaded was for dynamic linking, not static linking.
You need to remove the libspeex.dll.a from the "lib" dir where you just unpacked it to (just the ones with dll in them, not libspeex.a), otherwise you'll also need to put the libspeex-1.dll into the LAV folder.

I posted corrected library zips, just when you already unpacked it, it won't remove the old files ;)

[SamuriHL]

Oh, kuel, thanks! I'll try that out now.

[SamuriHL]

That fixed it. Thanks, Nev!

[Pat357]

Nev,

The few samples I have with speex-audio played all fine with lav-splitter + lav-audio.

I hope the info from MPC-AVI splitter and Haali splitter I posted contain some usefull info for you to make speex-audio work with other splitters too.
If you need more info or samples, just ask.

[nevcairiel]

I added support for other AVI Splitters, but for some reason the first few seconds of audio are corrupted with other AVI splitters.
Maybe the file i tried is weird, but it works with LAV, so i dunno. Not that i care much, use LAV or go away. :D

[Pat357]

I took the binary package from http://henry.fushizen.eu/builds/LAV%20Filters/#!/view=details/lang=en/sort=na (latest 589c957bf655) to test the speex-decoder, but LAV-audio doesn't want to connect to the splitter anymore ..?

see http://www.mediafire.com/conv/34bd1fd4f578d1529d10e7cc015ced6642edc7c6bfbd34f145d599ff917395056g.jpg

There must be something wrong with his compilation...

[nevcairiel]

If you want to test the decoder in combination with LAV Splitter, i posted a build yesterday for that, which i know is working. :)
The change to also support other splitters was not available in the public Git until right now.

Recent versions require an additional library to make compilation work again, so its possible some people that do "automated" builds have it messed up right now.
Maybe i should change my mind and move the speex library back into the Git repository. :p

Edit:
And i did. I'll try to go back to keeping the libs out of my sources once i have a better plan for a build env.
To keep everything clean i purged two commits from history, so sorry about breaking some checkouts. :P

[karl_lillevold]

I am trying to get madVR to work along with the LAV video decoder using Intel QuickSync decoding, running in MPC-HC with Lucid Virtu. It works great with EVR Custom Pres, but with madVR there is no video. Graphstudio exhibits the same behavior - opens a video window, plays audio, but no video.

I have a SandyBridge 2500K, and I am using a discrete ATI Radeon 6450 because I cannot get the integrated Intel HD Graphics to display colors well on my HDTV.

If I remove the ATI card (so Lucid Virtu is not being used), playback works as expected (but poor color saturation).

if I use DXVA Copy-back on the ATI card, (and disable Lucid Virtu), playback also works.

I know this is perhaps an odd combination, but I would prefer QuickSync decoding, due to its better efficiency. (It can play Crowd.Run without skipping frames, the ATI card cannot. The fanless ATI card also runs hot while decoding Blu-ray resolutions).

[wanezhiling]

http://i.imgur.com/GFgiZ.png

I checked "Force Deinterlacing" and used LAV CUVID(50/60p output) to play back a 30p progressive video, but fps was still 30.

nev, this area has no effect on CUDA deinterlacing?

[egur]

I am trying to get madVR to work along with the LAV video decoder using Intel QuickSync decoding, running in MPC-HC with Lucid Virtu. It works great with EVR Custom Pres, but with madVR there is no video. Graphstudio exhibits the same behavior - opens a video window, plays audio, but no video.

I have a SandyBridge 2500K, and I am using a discrete ATI Radeon 6450 because I cannot get the integrated Intel HD Graphics to display colors well on my HDTV.

If I remove the ATI card (so Lucid Virtu is not being used), playback works as expected (but poor color saturation).

if I use DXVA Copy-back on the ATI card, (and disable Lucid Virtu), playback also works.

I know this is perhaps an odd combination, but I would prefer QuickSync decoding, due to its better efficiency. (It can play Crowd.Run without skipping frames, the ATI card cannot. The fanless ATI card also runs hot while decoding Blu-ray resolutions).

Using Lucid Virtu implies the following:
* All DirectX/DXVA will go to the iGPU.
* Display buffers are copied to the dGPU.
* MadVR will run on the iGPU which is not optimal even if it works.

You can try faking a display by following these instructions (http://forum.doom9.org/showthread.php?p=1532786#post1532786).
I have a very similar setup (i7-2600, Radeon HD 6950, H67 board).
By faking the display and not using Virtu you get the advantage that MadVR will run on the dGPU (AMD) which is probably stronger (assuming you connected the display to it).
This trick might also enable other QuickSync enabled SW to run without a display connected. This is an unfortunate DirectX limitation - you can't enumerate a headless GPU. It was fixed in DX10 but DXVA (and Quicksync) relies on DX9.

I can try to reproduce your issue but I need more details:
* OS version (64 bit?)
* Virtu version
* Intel driver version.

Regarding display colors, my own HTPC is connected to a high quality plasma display via the iGPU. The colors are correct. It sounds like the output is in the range of 16-235 which will produce a milky image (blacks are grey, color is less saturated, image has low contrast). Try checking the values in graphics properties->Display->Monitor/TV setting
The TV should be set to full range (0-255). I'm assuming HDMI, DVI or DisplayPort output.

[karl_lillevold]

egur: thanks for the quick reply and fake display tips! Worked like a charm. Intel QuickSync with madVR now displaying on my main screen without any problems and looks great! Lucid Virtu was the problem. I was hoping to get rid of it completely, but for some reason even with the fake display, unless I enable it for dvdfab transcoding, it just sits there and never starts transcoding. It says QuickSync is available, but the transcoding pipeline hangs. My audio-less test clips also hang with this setup.

Wrt. colors on Intel: there are problems with the "Full" color range setting. It goes back to Limited by itself. I see numerous threads on avsforum discussing this problem. When I set it "Full", black-level test videos do indicate it is using the full range, at least temporarily, but I still get washed out colors, compared to ATI. Latest driver 8.15.10.2696. Win7 x64. Virtu 112. ASUS Z68 (maximus iv gene-z/gen3) MB. Connected to Onkyo TX-SR505 to Samsung DLP HDTV via HDMI.

[Avish]

Egur, I'm gonna try out the fake display trick, so could you please tell me what bios options I have to set for it to work?? My setup is same as yours except I have N460GTX Cyclone. :)

[egur]

The BIOS has to be set to enable the iGPU. If you installed Intel's graphics driver and it shows up in the device manager, than the BIOS setting are fine.
Some boards will disable the iGPU (in BIOS) when a dGPU is present.

Regarding the limited color range, I'll check around, my own TV is connected directly to the iGPU and my projector is connected though an Onkio 608 via an Nvidia dGPU on an older platform (core 2 quad).

Update:
I checked my driver setting (connected to Panasonic TV 50V20) and the settings were on 'limited'. Changing it to 'full' caused the colors to become washed out (i.e. RGB in 16-235 range). So in my case the labels are reversed.
I set the range to full and rebooted. It returned to 'Limited' like previously reported.
I'll report the 3 issues, this should be a simple fix.

I performed the tests using the desktop to eliminate renderer setting problems.

Does this also happen in IvyBridge?

[ryrynz]

It really is awesome that you're here to report these sorts of problems and get them resolved quickly for us Eric, thanks.

[Cyber-Mav]

is hardware acceleration for xvid and divx working yet?

[wanezhiling]

MPEG4 ASP works fine if you have a 300 series driver.

[karl_lillevold]

egur: thanks for the quick reply and fake display tips! Worked like a charm. Intel QuickSync with madVR now displaying on my main screen without any problems and looks great!
The only and very annoying problem now is that whenever I switch away to another input on my AVR or connect from work via RDT, Windows loses the multiple desktop setup information, and I have to redo the fake display setup every time... There is a 40+ page topic on this general "KVM" switching problem on answers.microsoft.com (http://answers.microsoft.com/en-us/windows/forum/windows_7-hardware/kvm-switch-windows-7-creates-monitor-switching/84f41660-1933-4109-9b13-1ea8a1c27be7). There is no good solution to this :(

Update: and not only does Win7 lose the multi desktop setup under the above conditions. Just switching from 60 to 23.976Hz with an ATI hotkey, which I frequently do, I have to re-extend the desktop. I don't have to go through the Detect stage, just choose extend desktop again. And then when I extend the desktop, it switches back to 60Hz. This is becoming an unrealistic solution.

[karl_lillevold]

Update:
I checked my driver setting (connected to Panasonic TV 50V20) and the settings were on 'limited'. Changing it to 'full' caused the colors to become washed out (i.e. RGB in 16-235 range). So in my case the labels are reversed.
I set the range to full and rebooted. It returned to 'Limited' like previously reported.
I'll report the 3 issues, this should be a simple fix.

I performed the tests using the desktop to eliminate renderer setting problems.

Does this also happen in IvyBridge?
Good findings. I will run some more tests as well. I also have an IvyBridge dev platform at work, provided to RealNetworks by Intel, but only computer monitors to connect it to.

[egur]

The only and very annoying problem now is that whenever I switch away to another input on my AVR or connect from work via RDT, Windows loses the multiple desktop setup information, and I have to redo the fake display setup every time... There is a 40+ page topic on this general "KVM" switching problem on answers.microsoft.com (http://answers.microsoft.com/en-us/windows/forum/windows_7-hardware/kvm-switch-windows-7-creates-monitor-switching/84f41660-1933-4109-9b13-1ea8a1c27be7). There is no good solution to this :(

Update: and not only does Win7 lose the multi desktop setup under the above conditions. Just switching from 60 to 23.976Hz with an ATI hotkey, which I frequently do, I have to re-extend the desktop. I don't have to go through the Detect stage, just choose extend desktop again. And then when I extend the desktop, it switches back to 60Hz. This is becoming an unrealistic solution.

I guess this hack isn't robust enough for setup. It's working for me on my dev platform, but I don't change resolutions or frame rate so I encounter these issues.
There's the option to use a VGA dummy (Google it).
I hope the driver or MSDK will support HW acceleration without a screen. I don't have any other information on the matter.

[egur]

Wrt. colors on Intel: there are problems with the "Full" color range setting. It goes back to Limited by itself. I see numerous threads on avsforum discussing this problem. When I set it "Full", black-level test videos do indicate it is using the full range, at least temporarily, but I still get washed out colors, ...
What exactly did you do?
How did you test this?

FYI, blacker-than-black (Y < 16) pixels when converted to RGB in a PC system will become RGB(0,0,0). Same goes for whiter than white (Y > 235) they become RGB(255,255,255).

The only way to notice these out of range pixels is to use procamp (Contrast and brightness) while pixels are still in YUV colorspace. Raising the brightness will move real backs (Y =16) to a higher value, turning it to grey and the blacker-than-black pixels will stay black.

[Pat357]

What exactly did you do?
How did you test this?

FYI, blacker-than-black (Y < 16) pixels when converted to RGB in a PC system will become RGB(0,0,0). Same goes for whiter than white (Y > 235) they become RGB(255,255,255).

The only way to notice these out of range pixels is to use procamp (Contrast and brightness) while pixels are still in YUV colorspace. Raising the brightness will move real backs (Y =16) to a higher value, turning it to grey and the blacker-than-black pixels will stay black.

One can easily test the difference between full and limited with MadVR : IIRC, the short Ctrl+Alt+Shift+Y will switch between TV (limited) and full (PC).
With output-levels to full, madVR will expand the YUV Y=16 -> Y=235 to resp. RGB(0,0,0) and RGB(255,255,255).
This also means that there is no BTB or WTW when using "full range" output levels in MadVR.
This is good, because video/movie content is almost always limited (only Y16 -> Y235).
The expansion done by MadVR makes also sure we optimally use the whole range : otherwise the regions Y=0-16 and Y=235-255 and there respective RGB ranges would get wasted, because the is no useful info in these ranges for video/movie content.
The above is only valid if ALL your interfaces (HDMI, AVR,TV,..) are all set to FULL range and no other level conversion is done after MadVR expanded the levels.

[malmsteen81]

@nevcairiel

hi, in future release can you add an auto mode for hardware deinterlacing?

thanks

Andrea

[wanezhiling]

LAV only has one hardware deinterlacing -- CUDA deinterlacing.

[malmsteen81]

i mean film/video mode

[karl_lillevold]

What exactly did you do?
How did you test this?

I realize now the clip I used was a TV levels test clip... Thanks for pointing out this error. I just now found the AVSHD-709 collection (http://www.avsforum.com/avs-vb/showthread.php?t=948496). In the "1-Black Clipping" video in this collection, all levels below 17 appeared black on my HDTV.

[egur]

To sum this up, the bug is that the full/limited labels are reversed and that the setting doesn't stick?
Anyone has anything to add (see a different problem)?

BTW, since PCs output RGB (desktop is always RGB), it's always better to do the calibration without video - just a bitmap pattern. This eliminates bad settings in the renderer/player.

[Leapo]

Running into a pretty major issue with fallback detection when using LAV with Windows Media Center.

If I put LAV Video into DXVA (copy-back) mode, and run Windows Media Center in a window, everything works perfectly. LAV uses DXVA to decode the video. If I fullscreen Windows Media Center while the video is already playing, it continues to use hardware acceleration without a problem.

Now, here's where it gets weird. If I start Windows Media Center, fullscreen the window first, THEN start the same video, LAV Video immediately falls back to software decoding. The only difference is that the Media Center window was already fullscreen when I clicked "play".

Any chance you could look into this? LAV would finally make the AMD graphics card in my media center usable for hardware accelerated decoding if this worked correctly!

[nevcairiel]

BTW, since PCs output RGB (desktop is always RGB), it's always better to do the calibration without video - just a bitmap pattern. This eliminates bad settings in the renderer/player.

If your HTPC is the main source for the TV, you should calibrate for your final video playback chain.
Of course making sure its configured properly is the first step.

[nevcairiel]

Any chance you could look into this? LAV would finally make the AMD graphics card in my media center usable for hardware accelerated decoding if this worked correctly!

Use DXVA Native.
While in theory DXVA-CB could be made to work, it would only work if LAV Video is connected directly to the renderer itself, not allowing post-processing in any case, so DXVA-CB does not offer any advantages over DXVA native anymore.
Thats the limitations the Exclusive mode WMC runs under places on you, can only use DXVA if you get direct contact with the renderer.