madVR - high quality video renderer (GPU assisted)

[trip_let]

Is Catmull-Rom messed up for anyone else after image doubling (super-xbr or NNEDI3)? This is on v0.89.11, but I haven't been testing every new build.

On the latest build, Win10 x64, GT 650M/HD 4000 Optimus.

It should be obvious here (applies to any video but clear enough when working with something very low res):
http://screenshotcomparison.com/comparison/147572

Catmull-Rom and Bicubic 50 should be the same thing, right?

[Nezil]

I have an interesting, though probably not that unusual, requirement for managing zoom control and aspect ratios.

What I’m basically looking for, is a way of setting up profiles in madVR to automatically achieve the result that I’m looking for; any suggestions would be very much appreciated.

99.9% of the content I watch is Blu-ray content, which has black bars embedded in the video stream itself, and has a stream aspect ratio of 1920 x 1080. Having said that, there are several different aspect ratio cases commonly seen with Blu-ray content:


16:9 content; 1920 x 1080 real video data
1.85:1 content; 1920 x 1040 real video data
~2.35:1 content; 1920 x ~800 real video data
Movies with IMAX scenes - e.g. Interstellar
Movies with multiple aspect ratios - e.g. The Grand Budapest Hotel

My viewing setup consists of a JVC DLA-RS46 projector with a 2.35:1 screen. I have black velvet masking panels that attach to the sides of the screen, resulting in a 16:9 sized area for projection when needed as well. The JVC projector has motorised zoom and shift, and I have it set up with two memories


16:9 mode, where all 1920 x 1080 active pixels fill the centre of the screen (masking attached on the sides)
2.35:1 mode, where 1920 x 816 pixels fill the whole 2.35:1 screen, 132 pixels at the top and bottom of the projected image fall above and below the screen.

Before madVR’s zoom and black bar detection algorithms were added to the renderer, I simply used the lens memories to zoom the content manually. This worked great for the first 3 commonly found Blu-ray formats listed above, but type 4 resulted in annoying image data being projected above and below my screen for some scenes (assuming 2.35:1 projector zoom), and type 5 would result in the perception of black all around the video data for some scenes (assuming 16:9 projector zoom).

I would like madVR to behaving in the following way for each of the 5 content types (My projector setting in parenthesis):


No zoom (16:9 mode)
No zoom (16:9 mode)
No zoom (2.35:1 mode)
‘Crop’ all content to 1920 x 816 (2.35:1 mode)
Zoom all content to fill as much of the 1920 x 816 active pixels that fit on the projection screen, sometimes upscaling, sometimes downscaling (2.35:1 mode)

So far, I have created two different screen config profiles for my display, effectively corresponding to the lens memories setup in the projector:


Cropped 1920 x 816 visible area defined
No cropping

I also have 3 profiles for zoom control:

Default (for content types 1, 2 and 3)
- All zoom controls disabled
IMAX Scene Movies (for content type 4)
- Disable scaling for ... 25 lines or less
- Automatically detect black bars
- If their are big black bars ... zoom the bars away completely
Mixed Ratio Movies (for content type 5)
- Disable scaling for ... 10 lines or less
- Automatically detect black bars
- Zoom small black bars away

Currently, the only way that I can think of to use the Profile Rules, is to add '*.IMAX.*' / '*.multi.*' to the filenames to trigger selection of both sets of profiles, but I'd rather this be automated without having to change filenames.

I totally understand the reason that the zoom control works the way that it does, but I'm finding it difficult to setup rules and configurations that consistently achieve the desired effect for the content that I have.

This isn't a complaint, I'm just wondering if there is a better way to do what I'm trying to do.

[blueray101]

Can someone please tell the best setting for watching anime in 720p and 1080p for a high-end PC. I never watch anything below that.

[ryrynz]

Can someone please tell the best setting for watching anime in 720p and 1080p for a high-end PC. I never watch anything below that.

http://wiki.mikejung.biz/MadVR

[Vyral]

Can someone please tell the best setting for watching anime in 720p and 1080p for a high-end PC. I never watch anything below that.
This guide should help you get started : https://imouto.my/tutorials/madvr/

Then you'll have to play with the settings yourself.

[nijiko]

@madshi

Sorry for late. (Bcos it will make com dead.)

1,2,3 have no problem.
4,5 will popup Resample shader logic error (3).

[nijiko]

@madshi ,is test1 89.10?
I can't find 89.10 on the internet. When I see you publish 89.10, your link was turn to 89.11.
And in videohelp.com, 89.10 and 89.11 are both the same 89.11.

[nevcairiel]

0.89.10 was severly broken, and there was only a single fix in 0.89.11 to fix that problem in .10, so there is no reason to test .10 for anything.

[nijiko]

According to his words, test1 is the highest risk, meaning that it's the farest from 89.9.
But it works well with me, and it's showing 89.10. I never have a test for 89.10 release.
If I now turn to use 89.11, com will be fully freezed immed, again.
Then the problem is between test1(89.10) and 89.11.

[Sunset1982]

Hey guys,

I'm trying to find the best madVR configuration on my HTPC. I read a lot "madvr guides" and now I want to ask if someone could give me some suggestions for my system.

Syszem specs: i5 4590, 16GB DDR3 RAM, GTX970, Win 8.1
mainly used to upscale 1080p mkv to my 4K UHD OLED TV.


my settings atm:

artificial removing: default: low / during fade: high

chroma upscaling: NEEDI3, 16N
image downscaling: CR with AR and LL
image doubling: double luma if factor is 2.0x with NEEDI3, 32N, and double chroma with 2.0x and NEEDI3 16N
image upscaling: Jinc AR
Refinement: SuperRes St:3, sh:2, rad:066, once after upscaling ist complete


general: I use FSE and D3D11

Dithering: Ordered Dithering with cn and change dither every frame

trade quality for performance: "dont render frames when fade in out ist detected" is checked

With these settings i get 38ms and 1080p video upscaled to UHD without glitches and stuttering.


Does anyone see any settings which could be changed for better quality? or any ideas of chaning settings?

the guy in this guide http://wiki.mikejung.biz/MadVR#DXVA2_Focused_2 uses DXVA under imageupscaling and downscaling and therefor using higher Neurons under doubling and chroma. what do you think about that?

Any suggestions for different settings than mine for 4K upsacling are welcome!

[madshi]

Do you have any idea on how madvr would fair with Nvidia Quadro M6000?
No.

I found that enabling it in JRiver Media Center really messes things up when videos contain multiple aspect ratios.

For example, I have a video which is supposed to be displayed as 16:9, but it is horizontally compressed.
This video contains three different aspect ratios:
16:9 full-screen
4:3 pillarboxed (12:9)
16:9 windowboxed (16:9 letterboxed in a 4:3 frame)

Because it is horizontally compressed, it actually displays as approximately:
15:9 full-screen
11:9 pillarboxed
15:9 windowboxed (15:9 letterboxed in an 11:9 frame)

If I enable aspect ratio correction in JRMC what happens is that you get:

15:9 full-screen → 16:9 full-screen (correct)
11:9 pillarboxed → stretched to 16:9 full-screen (very wrong!)
15:9 windowboxed → 16:9 full-screen (correct)

This could quite well be a bug in madVR, but I'm not sure. How exactly can I reproduce this on my PC?

Does SuperRes do any deblocking? Trying to figure out what it does exactly.
No, it doesn't deblock. SuperRes tries to improve the upscaled image, by fixing errors the upscaling algorithm might have done, and by sharpening the image if the upscaling algorithm produced a soft output.

Don't disable deband fade-in-out performance trade-off option, it requires very long queues to prevent frame drops.
That completely solved the problem, many thanks!:thanks:
Good call, aufkrawall.

Alternatively, probably increasing the CPU and/or GPU queue size(s) would also help. Turning the fade-in-out trade quality option off does sometimes require queues to be a bit larger than usual to get smooth playback.

When I pause a video I can see that the image goes back a few frames.

I noticed this while presenting 8 frames in advance.
Setting it down to 1 helps tremendously but sometimes I can still see this glitch.
It might be slightly improved in the next build, but it will still be there. I consider this a cosmetical glitch and as such not very important at the moment. Remember, madVR is still not at v1.0, so some imperfections are to be expected. I might look at things like that after v1.0 release, or shortly before, but now is not the time.

I get some visible tearing when using a custom 71.928 Hz refresh rate in fullscreen exclude and windowed mode. HD7700 Windows 10.
Then either your GPU drivers or the OS is at fault. In theory it should not be possible to get tearing in Windows 8.1 or Windows 10, when using Direct3D. You could try using a different GPU driver. Maybe it helps.

Too bad that as soon as DXVA deinterlacing gets activated, one still gets the lossy conversion, even when using DXVA copyback. DXVA copyback is not the hardware accelerated solution that gives the best quality in all cases like suggested in the posts above then.
Maybe some day we'll get DXVA deinterlaced copyback, but not now. Usually DXVA deinterlacing is needed only for sports, music concerts and very rare/weird DVD encodes. And for those image quality isn't always the most important, anyway. So I don't consider this a show stopper. Although, obviously, it's not an ideal situation atm. FWIW, last time I checked, with an AMD GPU you get lossless quality. But not with Intel or NVidia. So if DXVA deinterlacing is very important to you, you may want to get an AMD GPU for now...

1,2,3 have no problem.
4,5 will popup Resample shader logic error (3).
Try this build, it's based on v0.89.11/12 and should have the problem fixed:

http://madshi.net/madVR8911sneaker.rar

I'm trying to find the best madVR configuration on my HTPC. [...]

chroma upscaling: NEEDI3, 16N
image downscaling: CR with AR and LL
image doubling: double luma if factor is 2.0x with NEEDI3, 32N, and double chroma with 2.0x and NEEDI3 16N
image upscaling: Jinc AR
Refinement: SuperRes St:3, sh:2, rad:066, once after upscaling ist complete
The most important is luma doubling and SuperRes (and debanding, if your sources need it). Try to get 64 neurons for luma doubling. You can use cheaper algos for all other scaling settings (except SuperRes). Don't use DXVA scaling, I think it will disable luma doubling.

I am using madVR v0.89.11 and MPC-BE v1.4.6.892. I'm not sure how to cut a small sample but it happens with all of my DVDs. Here is a debug log.

http://www.megafileupload.com/qjnU/madVR_-_log.txt
Do you have any form of aspect ratio correction enabled?
^

As 6233638 already asked: Do you have any aspect ratio overwrite active in MPC-BE? For your DVD madVR is reporting a desired target resolution of 720x540 at first, but MPC-BE tells madVR to upscale that to 1219x719, which is also the resolution of the MPC-BE window. Since the aspect ratio of 1219x719 doesn't match 720x540, madVR believes that MPC-BE is switched to "stretch to window" mode, and then behaves like that later on. I can judge from the log that probably MPC-BE is not really in "stretch to window" mode, but instead you've probably activated a 16:9 aspect ratio override. But madVR cannot know that. madVR later asks for 853x480 instead of 720x540, which then is the same AR as 1219x719, so at that moment both "stretch to window" and "touch window from inside" would fit, but at that moment madVR has already decided that MPC-BE is probably in "stretch to window" mode.

MPC-HC in the meanwhile has added support for madVR's new zooming APIs, so with MPC-HC there's no doubt, anymore, what MPC-HC wants. But MPC-BE doesn't support those madVR APIs, so madVR has to guess what MPC-BE wants, and in your case obviously madVR has guessed incorrectly. You can probably fix that by removing the aspect ratio override.

Is Catmull-Rom messed up for anyone else
Ooops, good catch - thanks for letting me know! Will be fixed in the next build.

I have an interesting, though probably not that unusual, requirement for managing zoom control and aspect ratios.

What I’m basically looking for, is a way of setting up profiles in madVR to automatically achieve the result that I’m looking for; any suggestions would be very much appreciated.

99.9% of the content I watch is Blu-ray content, which has black bars embedded in the video stream itself, and has a stream aspect ratio of 1920 x 1080. Having said that, there are several different aspect ratio cases commonly seen with Blu-ray content:


16:9 content; 1920 x 1080 real video data
1.85:1 content; 1920 x 1040 real video data
~2.35:1 content; 1920 x ~800 real video data
Movies with IMAX scenes - e.g. Interstellar
Movies with multiple aspect ratios - e.g. The Grand Budapest Hotel

My viewing setup consists of a JVC DLA-RS46 projector with a 2.35:1 screen.
So far everything 100% identical to me... ;)

I have black velvet masking panels that attach to the sides of the screen, resulting in a 16:9 sized area for projection when needed as well. The JVC projector has motorised zoom and shift, and I have it set up with two memories


16:9 mode, where all 1920 x 1080 active pixels fill the centre of the screen (masking attached on the sides)
2.35:1 mode, where 1920 x 816 pixels fill the whole 2.35:1 screen, 132 pixels at the top and bottom of the projected image fall above and below the screen.

Before madVR’s zoom and black bar detection algorithms were added to the renderer, I simply used the lens memories to zoom the content manually. This worked great for the first 3 commonly found Blu-ray formats listed above, but type 4 resulted in annoying image data being projected above and below my screen for some scenes (assuming 2.35:1 projector zoom), and type 5 would result in the perception of black all around the video data for some scenes (assuming 16:9 projector zoom).

I would like madVR to behaving in the following way for each of the 5 content types (My projector setting in parenthesis):


No zoom (16:9 mode)
No zoom (16:9 mode)
No zoom (2.35:1 mode)
‘Crop’ all content to 1920 x 816 (2.35:1 mode)
Zoom all content to fill as much of the 1920 x 816 active pixels that fit on the projection screen, sometimes upscaling, sometimes downscaling (2.35:1 mode)

So far, I have created two different screen config profiles for my display, effectively corresponding to the lens memories setup in the projector:


Cropped 1920 x 816 visible area defined
No cropping

I also have 3 profiles for zoom control:

Default (for content types 1, 2 and 3)
- All zoom controls disabled
IMAX Scene Movies (for content type 4)
- Disable scaling for ... 25 lines or less
- Automatically detect black bars
- If their are big black bars ... zoom the bars away completely
Mixed Ratio Movies (for content type 5)
- Disable scaling for ... 10 lines or less
- Automatically detect black bars
- Zoom small black bars away

Currently, the only way that I can think of to use the Profile Rules, is to add '*.IMAX.*' / '*.multi.*' to the filenames to trigger selection of both sets of profiles, but I'd rather this be automated without having to change filenames.

I totally understand the reason that the zoom control works the way that it does, but I'm finding it difficult to setup rules and configurations that consistently achieve the desired effect for the content that I have.

This isn't a complaint, I'm just wondering if there is a better way to do what I'm trying to do.
First of all, although all the controls in zoom control etc seem to be relatively complicated, they were designed to handle all sorts of display setups, including CIH, CIW and CIA front projection setups. Your overall configuration seems to be ok, but a bit more complicated than necessary.

BTW, v0.89.12 will be able to automatically activate your JVC lens memories for you, so there's even less work you have to do manually now! Here's what I'd recommend, using the upcoming v0.89.12:

1) device -> properties

- Connect your JVC to the LAN.
- Select a proper IP address, or enable DHCP.
- In the JVC setup enable IP control via LAN.
- In madVR -> device -> properties press "find projector".
- Activate "pause playback while changing lens memories".

2) device -> screen config

- Create a profile for 16:9, no masking, activate lens memory 1.
- Create a profile for 21:9, proper masking, activate lens memory 2.

if (fileName = "*multi*") or (ar > 1.9) "21:9" else "16:9"

3) processing -> zoom control

Create a "default" profile for all movies, including IMAX, with the following options checked:

- disable scaling if image size changes by only: [whatever you like]
- automatically detect hard coded black bars
- if black bars change pick one zoom factor: ... which doesn't show any black bars
- crop black bars

Create a "multi" profile for Grand Budapest Hotel and similar movies, with the same settings as above, just disable the "if black bars change pick one zoom factor".

Normally, you wouldn't need profiles for zoom control. The big problem is that you want madVR to behave differently for IMAX movies compared to Grand Budapest Hotel, and madVR cannot know which is which. So you have to file name tag either IMAX movies or Grand Budapest Hotel. The key option which makes all the difference is the "if black bars change pick one zoom factor", which you want enabled for IMAX and disabled for Grand Budapest Hotel.

if (fileName = "*multi*") "multi" else "default"

[6233638]

This could quite well be a bug in madVR, but I'm not sure. How exactly can I reproduce this on my PC?I didn't think that it was a bug, just a quirk of enabling JRMC's aspect ratio correction because it seems to blindly scale the input to the selected output aspect ratio, and Zoom Control changes what the input is as the aspect ratio in the video changes.

I will get you a sample later today though - I'll try to find a sequence where it switches between the three aspect ratios in close succession.

It would be great if you could figure out some kind of solution for this, but I recognize that it's an odd situation.

[nijiko]

Try this build, it's based on v0.89.11/12 and should have the problem fixed:

http://madshi.net/madVR8911sneaker.rar

No problem, too.
Then, what's the real problem on earth?

[madshi]

I didn't think that it was a bug, just a quirk of enabling JRMC's aspect ratio correction because it seems to blindly scale the input to the selected output aspect ratio, and Zoom Control changes what the input is as the aspect ratio in the video changes.

I will get you a sample later today though - I'll try to find a sequence where it switches between the three aspect ratios in close succession.

It would be great if you could figure out some kind of solution for this, but I recognize that it's an odd situation.
It could be JRMC's fault, or mine, I don't know. Either way, shouldn't be hard to get it fixed. If it's JRMC's fault, I know who to contact... :)

No problem, too.
Then, what's the real problem on earth?
There's no problem (anymore). Will be fixed in v0.89.12.

[nijiko]

There's no problem (anymore). Will be fixed in v0.89.12.

Now I use HC 181/190 + LAV 66.28 + this one madVR, and work fine.
I wonder what's wrong in 89.11 release ver?
Also, hope 89.12 to publish asap.

[madshi]

I wonder what's wrong in 89.11 release ver?
A bug in madVR, obviously. Although your OS/GPU driver is to blame for freezing the PC. The bug should have been handled more gracefully by your OS/GPU driver.

[madshi]

madVR v0.89.12 released

http://madshi.net/madVR.zip

* added auto lens memory activation via "IP Control" for JVC & Sony projectors
* added several new profile rule script variables
* added IMadVRCommand::SendCommandDouble("setArOverride") support
* when moving subtitle, margins are now relative to video height
* fixed: memory leak introduced in v0.89.10
* fixed: Catmull-Rom was broken since v0.89.10
* fixed: potential cause for "old frame" flickering when using smooth motion
* fixed: potential cause for "old frame" flickering in new windowed/FSE modes
In the next few weeks only expect bugfixes and small improvements, no big new features. Need to concentrate on my daily job for a while now.

[6233638]

* added several new profile rule script variablesThanks for the changes. (http://bugs.madshi.net/view.php?id=349)
Profile selection is working well for me with Zoom Control now.

[Ver Greeneyes]

* fixed: potential cause for "old frame" flickering in new windowed/FSE modesInteresting. Is this specific to D3D9? I don't remember ever seeing "old frame" flickering with D3D11, though I have had some issues with queues not filling up (still haven't had time to try and reproduce that). I've only seen it with streams that drop frames, using the D3D9 path.

[madshi]

I could only reproduce the "old frame" flickering when doing frame stepping under very specific circumstances in DX11, but the underlying issue is not specific to anything. It applies both to ("new") windowed and FSE modes, both DX9 and DX11. It's only one potential cause of the problem, though, and only rarely occurred, so I've no idea whether having fixed this will bring any improvements for anybody other than in the specific frame stepping case. It might, or maybe not.

[aufkrawall]

D3D11 windowed fullscreen flickering is still there.
Not a surprise, since it's most likely a NV Windows 10 driver bug with the way how madVR does things with DX11.

My script now fully works as intended, thank you.
I think I will now also use SuperRes for low-res cartoon scaling instead of Adaptive Sharpen, since the repair effect of SR is really nice when you don't use 256 neurons (who would ever do this?).

[XMonarchY]

Is it normal having to reset madVR to default settings every 2-3 builds to NOT experience issues? Every so often, updating madVR files results in me having severe presentation errors that go away after I reset settings to defaults.

[Olivier C.]

@madshi :
i have a question about the "scale in linear light" option in a downscale luma context.
I realized that this option is very very GPU demanding, even on very powerful graphic card like GTX970 or Zotac 980Ti (both tested).
To give an idea, the GTX970 can vary from 23% to 78% GPU on a 4K movie with this option disabled/enabled respectively.

I suppose it's more complicated than doing exp(1/gamma) on luma but i wonder which kind of processing is behind this option, if i can ask ?

[sneaker_ger]

Thx for the new version. Memory is looking good.

I could only reproduce the "old frame" flickering when doing frame stepping under very specific circumstances in DX11, but the underlying issue is not specific to anything. It applies both to ("new") windowed and FSE modes, both DX9 and DX11. It's only one potential cause of the problem, though, and only rarely occurred, so I've no idea whether having fixed this will bring any improvements for anybody other than in the specific frame stepping case. It might, or maybe not.
I actually had a problem with out-of-order frames when pausing in D3D9 FSE mode. At the time it was introduced I was busy and since it was more of a cosmetic issue I never bothered reporting. It is fixed now with v0.89.12, though it appears the pausing is not instant but takes an additional 1 or 2 frames to be finished. (Was it always this way? I can't remember)

[michkrol]

Thanks for the new release. Works like a charm here. :cool:

Is it normal having to reset madVR to default settings every 2-3 builds to NOT experience issues? Every so often, updating madVR files results in me having severe presentation errors that go away after I reset settings to defaults.

I'm a happy madVR user for years now and had to reset setting like one time to fix an issue.
Granted we probably have completely different setups, it's still not normal to reset setting almost every other build.

EDIT:
@madshi :
i have a question about the "scale in linear light" option in a downscale luma context.
I realized that this option is very very GPU demanding, even on very powerful graphic card like GTX970 or Zotac 980Ti (both tested).
To give an idea, the GTX970 can vary from 23% to 78% GPU on a 4K movie with this option disabled/enabled respectively.

Seems interesting. Are you sure the GPU is not changing frequencies to save power?
On a quick test I get 35% usage for Catmull-Rom without linear light and 41% with LL for image downscaling on a 24fps 4K viedo and that's on a Geforce 750Ti(!).
Could you post your settings, especially scaling algorithms used, smooth motion (on/off), dithering used and screenshot of debug OSD?

[jmonier]

The "auto lens memory" seems to have problems in my setup: JVC RS-4910 and Zoomplayer. It worked a couple of times after I shutdown Zoomplayer and started it up again, but now even that doesn't seem to work. Even when it did work, it never paused playback while it was changing. (I DID verify that I could pause/play Zoomplayer from the madVR tray icon.)

I've worked a lot with the JVC tcp/ip control. I have been able to make it work reliably 99.9% of the time, but I still consider it to be a real mess done by someone who didn't really understand network Sockets. It's actually less reliable (along with the RS-232 control) in the latest JVC models than it was before.

I'll be happy to work with you on this but I need some direction on where to look. Right now I have no idea as to how to proceed.

[pirlouy]

Is it normal having to reset madVR to default settings every 2-3 builds to NOT experience issues?
Yes. You also have to reset it every Sunday, December 25th and February 29th.

Each Time I read this topic, I regret it.

[DigitalLF]

* added auto lens memory activation via "IP Control" for JVC & Sony projectors

MadShi are you f*cking kidding me! you don't know how happy this makes me!!!!! THANK YOU THANK YOU THANK YOU!!!

Something is wrong in my setup. MadVR finds my projector on 192.168.0.50, same net as i use for all my stuff, it says its a Sony and it is, but nothing is happening! i got a Sony VPL-VW500ES. i have tried "activate lens memory: 1,2,3"

Mine are called: "1.85:1", "2.35:1", "Custom 1"

[madshi]

Is it normal having to reset madVR to default settings every 2-3 builds to NOT experience issues?
No.

The "auto lens memory" seems to have problems in my setup: JVC RS-4910 and Zoomplayer. It worked a couple of times after I shutdown Zoomplayer and started it up again, but now even that doesn't seem to work. Even when it did work, it never paused playback while it was changing.
Something in my setup. MadVR finds my projector on 192.168.0.50, same net as i use for all my stuff, it says its a Sony and it is, but nothing is happening! i got a Sony VPL-VW500ES. i have tried "activate lens memory: 1,2,3"
A couple of questions for both of you:

1) Do the profiles switch as intended? You can check by opening the settings while playing the movie and then checking which profile is drawn in bold.
2) Do you have black bar presentation enabled?
3) Please check with the following tool whether IP Control generally works reliably with your projector or not:

http://madshi.net/IpControlTest.rar

The problem could be a simple logic bug, madVR not even trying to activate the lens memory for some reason. Or it could be a problem with the IP Control code somehow.

Some additional questions just for @DigitalLF, because I don't have a Sony projector to test with here:

1) Does the "find" function reliably detect your Sony projector, without you having to enter an IP address etc?
2) Does the lens memory "abort" function work properly? It's supposed to stop/cancel/abort the activation of a lens memory immediately.
3) Does activating lens memory 1, then before it fully reached its final position, activating lens memory 2 work? Ideally pressing lens memory 2 once (while lens memory 1 is still being processed) should abort activating lens memory 1 and immediately start activating lens memory 2 instead. Does that fully work as expected?
4) If would be great if there was a way to find out when exactly activation of a new lens memory is fully finished. So could you please: a) Activate a new lens memory. b) Check the log how quickly success (or failure) is reported. c) Immediately press "read lens memory". Does "read lens memory" still report the previous lens memory number? Or already the new one we're currently in the process of activating? In the first case, please press "read lens memory" multiple times, maybe once every second or so, to check when exactly "read lens memory" switches to the new lens memory number. With a bit of luck maybe it will switch to the new lens memory number only after the lens memory activation has fully run through?

(P.S: The names of the lens memories don't matter.)

[har3inger]

Is Catmull-Rom messed up for anyone else after image doubling (super-xbr or NNEDI3)? This is on v0.89.11, but I haven't been testing every new build.

On the latest build, Win10 x64, GT 650M/HD 4000 Optimus.

It should be obvious here (applies to any video but clear enough when working with something very low res):
http://screenshotcomparison.com/comparison/147572

Catmull-Rom and Bicubic 50 should be the same thing, right?

Wait, so if catmull-rom and bicubic 50 are the same thing, why do both options exist?

[SecurityBunny]

I could only reproduce the "old frame" flickering when doing frame stepping under very specific circumstances in DX11, but the underlying issue is not specific to anything. It applies both to ("new") windowed and FSE modes, both DX9 and DX11. It's only one potential cause of the problem, though, and only rarely occurred, so I've no idea whether having fixed this will bring any improvements for anybody other than in the specific frame stepping case. It might, or maybe not.

The problem I was experiencing with every first subtitle line flickering seems to be fixed in 0.89.12. Thanks. :)

DX11 (new) windowed fullscreen path, smooth motion enabled.
MadVR 0.89.12 x64
XySubFilter 3.1.0.746 x64
MPC-HC 1.7.9.190 x64

The only other "flickering" I can reproduce is if I play a video with subtitles currently being displayed and then rapidly change the video aspect ratio, such as locking the media player to the side of my screen at half resolution. The subtitles flicker for a second when repositioning back into place with DX11. With DX9, the subtitles don't flicker but instead the video stretches into the new aspect ratio for the second.

[rack04]

As 6233638 already asked: Do you have any aspect ratio overwrite active in MPC-BE? For your DVD madVR is reporting a desired target resolution of 720x540 at first, but MPC-BE tells madVR to upscale that to 1219x719, which is also the resolution of the MPC-BE window. Since the aspect ratio of 1219x719 doesn't match 720x540, madVR believes that MPC-BE is switched to "stretch to window" mode, and then behaves like that later on. I can judge from the log that probably MPC-BE is not really in "stretch to window" mode, but instead you've probably activated a 16:9 aspect ratio override. But madVR cannot know that. madVR later asks for 853x480 instead of 720x540, which then is the same AR as 1219x719, so at that moment both "stretch to window" and "touch window from inside" would fit, but at that moment madVR has already decided that MPC-BE is probably in "stretch to window" mode.

MPC-HC in the meanwhile has added support for madVR's new zooming APIs, so with MPC-HC there's no doubt, anymore, what MPC-HC wants. But MPC-BE doesn't support those madVR APIs, so madVR has to guess what MPC-BE wants, and in your case obviously madVR has guessed incorrectly. You can probably fix that by removing the aspect ratio override.

I don't think I have anything other than default selected in MPC-BE.

http://i11.photobucket.com/albums/a199/rack04/Untitled_zpsq40sxwvh.jpg (http://s11.photobucket.com/user/rack04/media/Untitled_zpsq40sxwvh.jpg.html)

[truexfan81]

Is it normal having to reset madVR to default settings every 2-3 builds to NOT experience issues? Every so often, updating madVR files results in me having severe presentation errors that go away after I reset settings to defaults.

i've been using madVR for as long as i can remember, i have never once had to reset to defaults after an update.

Now that it has gone 64bit making it where i can use 64bit MPC-HC,LAV i have 0 complaints about it, madshi has done a great job on it. :thanks: madshi keep up the great work

[jmonier]

A couple of questions for both of you:

1) Do the profiles switch as intended? You can check by opening the settings while playing the movie and then checking which profile is drawn in bold.

Since, in my case it DOES work occasionally (and sets the proper lens memory when it does), that part is working.

2) Do you have black bar presentation enabled?

What do you mean by this? There's no option that this suggests to me.

3) Please check with the following tool whether IP Control generally works reliably with your projector or not:

http://madshi.net/IpControlTest.rar

The tests that can be run with power off run fine. I'll run the others the next time I have the projector powered on.

As I said in my post, it works reliably with code that I have written. I DO know that it can be very picky, so that doesn't mean that it will run reliably with your code.

The problem could be a simple logic bug, madVR not even trying to activate the lens memory for some reason. Or it could be a problem with the IP Control code somehow.

Since it does work occasionally for me, I don't think that it can be a simple logic bug.

4) If would be great if there was a way to find out when exactly activation of a new lens memory is fully finished. So could you please: a) Activate a new lens memory. b) Check the log how quickly success (or failure) is reported. c) Immediately press "read lens memory". Does "read lens memory" still report the previous lens memory number? Or already the new one we're currently in the process of activating? In the first case, please press "read lens memory" multiple times, maybe once every second or so, to check when exactly "read lens memory" switches to the new lens memory number. With a bit of luck maybe it will switch to the new lens memory number only after the lens memory activation has fully run through?

If you're currently doing this for JVC, it's certainly the kind of thing that I've found to be unreliable with the RS-4910. I've never done this specifically, so it might or might not be a problem. A test build that simply leaves it paused indefinitely would verify this quickly.

If you want to do this kind of thing, you have to be VERY careful (especially make sure you get an ACK before you send the next command). One thing that caused me a lot of trouble at first was that the JVC closes the socket on it's own after 3-5 sec.

Note that I had code that ran fine with my RS-50, but I had to make a lot of changes so that it would work reliably with my RS-4910.

[DigitalLF]

1) Do the profiles switch as intended? You can check by opening the settings while playing the movie and then checking which profile is drawn in bold.

First off! I just saw that i should change my Meridian HD621 to a SWITCH instead of a Digital Projector! I have not seen that you added "SWITCH" sorry for that but I am not a pro in anyway but i really want to learn! But that did not help and no profile gets bold but before i changed it the profile for my HDMI Switch was bold, might just have been even on the bedroom TV. I just got a Atlona HDMI Splitter called AT-HDDA-4 and it's connected to output on my Meridian HD621 but my bedroom TV is not connected because the does not get any sound from the HDMI splitter before i had the splitter i always had to enable down-mix in LAV Filter but that does not help me now. So every night i want to use the bedroom TV i have to remove the HDMI cable from my HTPC and insert my bedroom TV there on the same port that is used for my HDMI Switch. I am working with Atlona to get a fix for this! And hopefully there will be a HDMI Matrix from Meridian in the future that got 2 HDMI outs. I know my system is complex. Sorry for that, But that's just so i can use all my stuff connected to my HDMI switch on both displays.


2) Do you have black bar presentation enabled?
Don't know. Willing to submit any print screens you need. or if you need access to my HTPC or if you want a virtual machine i can set that up for us.


3) Please check with the following tool whether IP Control generally works reliably with your projector or not:

http://madshi.net/IpControlTest.rar

The problem could be a simple logic bug, madVR not even trying to activate the lens memory for some reason. Or it could be a problem with

the IP Control code somehow.

"FIND" works!
"PING" works!
"TEST" works!
"READ POWERSTATE" works!
"POWER ON" works!
"POWER OFF" works!
"HDMI1" fails!
"HDMI2" fails!
"READ LENS MEMORY" works!
"MEMORY 1" works!
"MEMORY 2" works!
"MEMORY 3" works!
"ABORT" fails!
"READ COLORSPACE" works!
"BT.709" works!
"DCI" fails!


1) Does the "find" function reliably detect your Sony projector, without you having to enter an IP address etc?
Yes it does, it finds both IP and "SONY".


2) Does the lens memory "abort" function work properly? It's supposed to stop/cancel/abort the activation of a lens memory immediately.
No, It does not!


3) Does activating lens memory 1, then before it fully reached its final position, activating lens memory 2 work? Ideally pressing lens memory 2 once (while lens memory 1 is still being processed) should abort activating lens memory 1 and immediately start activating lens memory 2 instead. Does that fully work as expected?
"MEMORY 1", "MEMORY 2" gives me a error. "activating lens memory 1 succeeded activating (new line) lens memory 2 failed" and stops the "MEMORY 1" zoom. But if i then press "MEMORY 2" again it zoom to "MEMORY 2" as it should.


4) If would be great if there was a way to find out when exactly activation of a new lens memory is fully finished. So could you please: a) Activate a new lens memory. b) Check the log how quickly success (or failure) is reported. c) Immediately press "read lens memory". Does "read lens memory" still report the previous lens memory number? Or already the new one we're currently in the process of activating? In the first case, please press "read lens memory" multiple times, maybe once every second or so, to check when exactly "read lens memory" switches to the new lens memory number. With a bit of luck maybe it will switch to the new lens memory number only after the lens memory activation has fully run through?

(P.S: The names of the lens memories don't matter.)
When i went from standard MEMORY 3 to MEMORY 1 it says MEMORY 1 directly.


P.S: it's 2.20 in the morning here so hopefully i have written all down but if anything is unclear i will try to do what i can to fix it. i have so much to learn from you guys before i saw that MadVR had gotten this WONDERFUL update i was about to buy a OPPO 103 today just to get rid of a problem i get when a movie pans sideways! And this is by far the longest forum post i have ever written in my 29 old life.

P.P.S: i really hope this text is understandable i have a lot of stuff against me so i tend to lose myself i text and dyslexia ain't helping.

[Olivier C.]

Seems interesting. Are you sure the GPU is not changing frequencies to save power?
On a quick test I get 35% usage for Catmull-Rom without linear light and 41% with LL for image downscaling on a 24fps 4K viedo and that's on a Geforce 750Ti(!).
Could you post your settings, especially scaling algorithms used, smooth motion (on/off), dithering used and screenshot of debug OSD?

Interesting, thanks for your reply and your test.
Yes, I am sure that the gpu is not changing frequencies (checked by gpu-z sensors).
I will post the informations you asked for, but as far as I can remember yet, i was using :

lanczos 4 taps for downscale luma
Jinc 3 taps for chroma upscale
Smooth motion disabled
Double luma disabled
Some trade quality for performance options

I will post further informations and screen.
Thanks for your help

[cvrkuth]

After install new version, my devices are gone?

http://i58.tinypic.com/5lui2p.jpg

[nijiko]

89.12 is well.

[Ruya]

Hi. For whatever reason, regardless of my settings, MadVR only ever uses D3D9 and not 11 (even in FSE mode). Drivers are up-to-date, and I've tried re-installing MadVR.

Using MPC-HC 1.7.9, OS is Windows 7 Ultimate x64 (Service Pack 1), GPU is AMD Radeon R9 290X.

[nevcairiel]

Hi. For whatever reason, regardless of my settings, MadVR only ever uses D3D9 and not 11 (even in FSE mode). Drivers are up-to-date, and I've tried re-installing MadVR.

Using MPC-HC 1.7.9, OS is Windows 7 Ultimate x64 (Service Pack 1), GPU is AMD Radeon R9 290X.

On Windows 7 you need to install the platform update, otherwise D3D11 is not available.

[madshi]

Wait, so if catmull-rom and bicubic 50 are the same thing, why do both options exist?
Because both are very well known. Also Bicubic 50 is just one of several Bicubic options. I could remove Bicubic 50, but then it wouldn't make the settings dialog any simpler. I could remove Catmull-Rom, but it's a name everybody knows.

The only other "flickering" I can reproduce is if I play a video with subtitles currently being displayed and then rapidly change the video aspect ratio, such as locking the media player to the side of my screen at half resolution. The subtitles flicker for a second when repositioning back into place with DX11. With DX9, the subtitles don't flicker but instead the video stretches into the new aspect ratio for the second.
Not important enough for me to worry about at this stage in the madVR development plan... :)

I don't think I have anything other than default selected in MPC-BE.
Strange. I can't reproduce any issues with this configuration on my PC. Does the madVR v0.89.12 OSD still report "stretch to window"? I did make a small change there, so maybe v0.89.12 behaves better? You really want to see "touch from inside" in the OSD.

Since, in my case it DOES work occasionally (and sets the proper lens memory when it does), that part is working.

[...]

Since it does work occasionally for me, I don't think that it can be a simple logic bug.
Not necessarily. The logic is very complex, tries to avoid setting the same lens memory multiple times etc. So there is a chance the logic could be at fault.

What do you mean by this? There's no option that this suggests to me.
Sorry, typo. I meant: "Do you have black bar *detection* activated"?

If you're currently doing this for JVC, it's certainly the kind of thing that I've found to be unreliable with the RS-4910. I've never done this specifically, so it might or might not be a problem. A test build that simply leaves it paused indefinitely would verify this quickly.

If you want to do this kind of thing, you have to be VERY careful (especially make sure you get an ACK before you send the next command). One thing that caused me a lot of trouble at first was that the JVC closes the socket on it's own after 3-5 sec.

Note that I had code that ran fine with my RS-50, but I had to make a lot of changes so that it would work reliably with my RS-4910.
Just give the test tool a try, please, to double check the reliability of Ip Control with your JVC, using my code. Thanks.

Do you have any other software which could potentially be issuing ip control commands to your JVC while madVR is running? That could cause trouble.

First off! I just saw that i should change my Meridian HD621 to a SWITCH instead of a Digital Projector! I have not seen that you added "SWITCH" sorry for that but I am not a pro in anyway but i really want to learn! But that did not help and no profile gets bold but before i changed it the profile for my HDMI Switch was bold, might just have been even on the bedroom TV.
I didn't mean which *device* was bold. I mean the "screen config" profiles. You do have different "screen config" profiles setup, so you could activate different lens memories, right? When you open the settings dialog while madVR is doing video playback, one of the sceen config *profiles* should be bold. Is it always the correct one?

"HDMI1" fails!
"HDMI2" fails!
Strange!! I had these tested by a VW1100 user and it worked for him. Anyway, it's not really important, I don't plan to do input reading/switching in madVR, anyway.

"MEMORY 1", "MEMORY 2" gives me a error. "activating lens memory 1 succeeded activating (new line) lens memory 2 failed" and stops the "MEMORY 1" zoom. But if i then press "MEMORY 2" again it zoom to "MEMORY 2" as it should.
Ok, two more questions about this:

1) What happens if you activate mem2, and then while it's still in the process of switching to mem2, if you activate mem2 *again*? Does it abort the operation then? Or does the 2nd mem2 button press have no effect?

2) What happens if you activate mem2, wait until it's fully done, then activate mem2 again? Does the projector do anything at all? Or is it clever enough to know that nothing needs to be done?

Here's a new build of the test tool:

http://madshi.net/IpControlTest.rar

With this new build please also repeat the following 2 tests:

1) Does abort work correctly now?
2) If you press mem2, and then mem1, before mem2 is fully done, does it properly abort mem2 and automatically activate mem1 now, with just the one button press?

After install new version, my devices are gone?
That's weird. I can only guess that your settings configuration got corrupted/damaged somehow. Does going back to the old version fix the problem? Please run a checkdisk/scandisk to make sure your file system is healthy.

Hi. For whatever reason, regardless of my settings, MadVR only ever uses D3D9 and not 11 (even in FSE mode). Drivers are up-to-date, and I've tried re-installing MadVR.

Using MPC-HC 1.7.9, OS is Windows 7 Ultimate x64 (Service Pack 1), GPU is AMD Radeon R9 290X.
On Windows 7 you need to install the platform update, otherwise D3D11 is not available.
^

And also Aero/desktop composition needs to be enabled. And "present several frames in advance" needs to be checked in the fullscreen exclusive mode settings.

[markanini]

Can I do better with settings for AMD HD7700 on Windows 10?

Two profiles, "LoFPS" handling everything below 31 FPS, "HiFPS" for framrates above.

"HiFPS"
Chroma upscaling: Bicubic75 + AR
Image downscaling: Catrom, linear-light
Image doubling: off
Image upscaling Jinc + AR
Upscaling refinement: off

"LoFPS"
Chroma upscaling: Bicubic75 + AR
Image downscaling: Catrom, linear-light
Image doubling: NNEDI3 64, Quadrupling: Super-XBR75
Image upscaling Lanczos3 + AR
Upscaling refinement: SuperRes enabled (default 3, 2, 0.66) "refine only once" and "apply superres first" checked. Finesharp enabled w/default settings if the source needs it.

The rest are default except for a calibration by 3DLUT.

[aufkrawall]

Are you sure this GPU can handle NNEDI3 64 + super-xbr + SuperRes 3 passes?

[markanini]

Are you sure this GPU can handle NNEDI3 64 + super-xbr + SuperRes 3 passes?

So far it's worked without dropped frames. I might not have played a difficult enough source yet. What would you suggest I'd drop in that case?

[aufkrawall]

Probably 32 neurons would be good in this case, since SuperRes wih 3 passes will most likely repair a lot of the scaling errors, at least if they look like aliasing.

[SweetLow]

And also Aero/desktop composition needs to be enabled. And "present several frames in advance" needs to be checked in the fullscreen exclusive mode settings.
May be you put link to this http://forum.doom9.org/showthread.php?p=1724688#post1724688 from first post of the thread?

[rack04]

Strange. I can't reproduce any issues with this configuration on my PC. Does the madVR v0.89.12 OSD still report "stretch to window"? I did make a small change there, so maybe v0.89.12 behaves better? You really want to see "touch from inside" in the OSD.

madVR v0.89.12.0 seems to work as expected. Here is the new log in case you're interested.

http://www.megafileupload.com/hW65/madVR_-_log.txt

[aufkrawall]

Performance improvements of NNEDI3 and SuperRes seem to pay off, I can watch 1080p60 -> WQHD with NNEDI3 32 chroma, Jinc AR image, SuperRes 3 passes and maximum deband + SM without dropped frames on my GTX 980 OC.
I'm relatively sure that this wasn't possible in the past.

[Asmodian]

http://wiki.mikejung.biz/MadVR

Wow, this guide seems insane to me. Really quite bad, using DSR with madVR is bad (it costs performance and hurts quality), NNEDI3 128 for chroma scaling?!? Debanding High/High?! CUDA decoding doesn't help render times...

Please don't use that guide.

Warner306's madVR setup (http://forum.kodi.tv/showthread.php?tid=222576&pid=1966571#pid1966571) is better, I don't agree with it entirely but it isn't insane or wrong anywhere.

[huhn]

i comment on that guide before. it's really strange to say the least...

totally agree with Asmodian.

blindly using hardware deinterlacing and totally ignoring powerstates yeah what ever...