madVR - high quality video renderer (GPU assisted)

[apgood]

Got 3D working in jriver mc using amd r9 270x. I needed to install crimson 16.1 hotfix, latest MadVR and LAV. The other thing I had to do in my projector is swap which eye comes first from "normal" to "swapped".

At first the video was a bit jerky but once I reduced all the scaling algorithms to no more than about super-xbr 100 or was fine.

[XTrojan]

Decrypting will likely be your responsibility, madVR does not use any protected paths, i.e. a HDCP display would not be required.
But as of today there are neither discs nor decryption software available. It's too early for those questions.


Currently no special HDR hardware support is required by madvr as it does not output any HDR metadata. Madshi explained it at bit here:
http://forum.doom9.org/showpost.php?p=1749826&postcount=34630

Full fixed function 10 bit HEVC decoding is currently only available on Nvidia GeForce GTX 750 SE, GTX 950, GTX 960. For AMD and Intel it will probably come with Polaris/Kaby Lake. A fast CPU might be able to at least do 2160p24 in software, though.

I will leave the answer about the different HDR variants to madshi but I'm pretty sure not all of them are supported at the moment. Nev and madshi will probably require at least the specs and samples and we might not get the latter before we can decrypt the discs (metadata might even be in playlist file, not HEVC SEI). Let's hope the new encryption will not be too big of a hurdle.

There's metadata, the samsung demo in a .mpeg container apparantly has HDR metadata, though i'm unsure if it's the final versions.

I do know the video is HDR as I tested it with a USB and my TV (JS9000), reports that it's getting HDR and changes backlight to max and dimming.

http://demo-uhd3d.com/categorie.php?tag=hdr
http://demo-uhd3d.com/fiche.php?cat=uhd&id=115

Can anyone check out that file, since mediainfo can't report the HDR info.

[Manni]

Got 3D working in jriver mc using amd r9 270x. I needed to install crimson 16.1 hotfix, latest MadVR and LAV. The other thing I had to do in my projector is swap which eye comes first from "normal" to "swapped".

At first the video was a bit jerky but once I reduced all the scaling algorithms to no more than about super-xbr 100 or was fine.

Thanks. Did you check with say chapter 2 of Gravity that the depth was correct? For example, when they circle around each other in space, is the depth correct for each of them and the rope/tether?

[madshi]

This is a bug in the AMD driver, and isn't something Nev or Madshi can do anything about.
Is it a constant delay? E.g. exactly one frame? Or does the delay differ sometimes, from movie to movie, or even from frame to frame? If it's a constant delay then it shouldn't be too hard to implement an option to delay one eye by 1-3 frames or so.

I have some quick feedback re: chroma superres.

From what I can see, the superres for chroma is quite similar to the superres for image refinement right now. However, while the algorithm is great for luma, it's not good for chroma. In luma, superres enhances small details and edges by increasing contrast (darks become darker, lights become lighter at image transients), and the same happens to the chroma layers. However, in chroma, light vs dark is actually green vs red, or blue vs yellow, etc. When the edges have their contrast increased, you get increased wrong-color ringing.

In actual content, I haven't found any live action image that actually looks different with SR on or off. In cartoon, it can help in many situations, but also increase artifacts and ringing in others. It's also quite dependent on source quality. All in all, I don't find the performance hit to be really worth it.
Ok, thanks for the feedback!

I'll have to do some testing to see when this changed, but I just watched my first PAL DVD in some time, and noticed that srcHeight is using the cropped values again, instead of the source height, so it ended up using my NTSC profile instead.

As a temporary fix I've had to change my profile rules to use uncroppedsrcHeight. (I still think that should be how srcHeight always works...)
Where did we do the previous discussion about this? In the bug tracker? I don't remember the conclusion of the our previous discussion, so I'm not sure right now if this behaviour is as intended or not.

Fair enough, but does it matter if we don't like Aero and have it disabled in W7?
It doesn't matter much if you only use fullscreen exclusive mode.

I think I've found the 3D issue for me in MC and it has something to do with "Bookmarks" and how playback is commenced (100% repeatable):
- If I play the 3D MKV from the beginning it works.
- If I seek while playing it works.
- If I double stop (reset book mark) then playback will commence from the beginning and 3D works
- If I stop/play (so playback commences from where I left off and not from the begining) it will go to SBS / using D3D9. If I try to force D3D11 in the madVR Settings it will crash MC.

My test setup was as simple as I could get it:
- Fresh install of madVR with default settings (but Windows Exclusive unchecked) No other changes
- Put Win 10 64-Bit Desktop into 3D
- Put the nVidia Control Panel into 1080/23hz/3D (so no switching is required by madVR)
- Commence playback testing
Ok, that's interesting. My best guess right now is that in the stop/play situation the decoder sends at least one frame in 2D and only switches to 3D a little bit later. That's just a guess, though. Can you create a debug log from the stop/play situation?

When you force D3D11 in the madVR settings, do you see anything at all, before it crashes? E.g. if you enable the OSD (Ctrl+J) before activating D3D11 mode, do you see the OSD for a short time before the crash? If so, please let me know which refresh rate the OSD lists. Is it near to 23.976? Or is it near to 47.952?

I have a bug on my setup that cropped up recently, after the 90.x builds, not sure if anyone else experiences this:

Subtitles rendered in black bars will play at 1x speed even if video is set to 2x speed. Leads to desync.
Is it a new bug with v0.90.x? That would surprise me. Which subtitle renderer are we talking about?

I think you can create Row/Column Interleaved and Checkerboard frames from stereo frames and output as 2D, just like you output SBS as a fallback method. For passive 3D sets, Row/Column Interleaved output should result in higher image quality as compared to frame packed 3D where the TV will use its internal processing to convert to Interleaved.
That would produce higher quality only if the output would be full resolution. For that to work, I'd have to switch to e.g. a 2D 1920x2160 mode (for row interleaved). And I have no API to tell the GPU to do that, so the user would have to create a 1920x2160 custom resolution. There's also no API to signal "row interleaved" to the TV, so the user would still have to manually switch to TV to "row interleaved" mode, if the TV has such a manual option.

It's basically 23fps 2160p, but when you enable "side by side" or "over under", vertical or horizontal resolution gets cut in half as it combines two sides into 3D or over/under into 3D. Most TVs should support this
Maybe you're right about 4K sets, I don't know. But I don't think many 1080p sets will accept a 2160p signal, or am I wrong?

I have a question regarding the upcoming UHD Blurays.

Will Dolby HDR and HDR10 be supported? And how will ripping work with HDCP 2.2 protection?

I heard that currently Nvidia nor AMD doesn't support HDR, but upcoming AMD Polaris will.
The key question is whether you will be able to play UHD Blu-Rays with madVR *at all*. It comes with a new copy protection, which isn't cracked yet.

If you somehow manage to unprotect UHD Blu-Rays, then madVR should have no problem playing HDR10 content. At the moment HDR passthrough to the display is not supported yet, but probably will be sooner or later. Right now you can already let madVR convert HDR10 content to a format current displays can understand, and that in high quality.

As long as I don't have technical Dolby Vision documentation, I can't support it.

Got 3D working in jriver mc using amd r9 270x. I needed to install crimson 16.1 hotfix, latest MadVR and LAV. The other thing I had to do in my projector is swap which eye comes first from "normal" to "swapped".

At first the video was a bit jerky but once I reduced all the scaling algorithms to no more than about super-xbr 100 or was fine.
Sounds good. Try a few different movies. It's possible that the eye swapping depends on the movie. If that's the case, either MakeMKV or LAV need a minor fix/update to solve that issue.

[omarank]

That would produce higher quality only if the output would be full resolution. For that to work, I'd have to switch to e.g. a 2D 1920x2160 mode (for row interleaved). And I have no API to tell the GPU to do that, so the user would have to create a 1920x2160 custom resolution. There's also no API to signal "row interleaved" to the TV, so the user would still have to manually switch to TV to "row interleaved" mode, if the TV has such a manual option.

No, I believe there is no need to output 1920x2160. If I understand it correctly, you will have to fit View 1 in every odd line and View 2 in every even line in a regular 1920x1080 2D frame. So the vertical resolution of both views will be reduced to 540. When you output this kind of frame to a Passive 3D TV (Row Interleaved Type), the user shouldn’t even need to tell the TV about the signal type. The user can just put on their 3D glasses and enjoy 3D. Similarly, for Column Interleaved Type Passive 3D TV, the horizontal resolution of both views will be reduced to 960 when you output Column Interleaved 1920x1080 2D frames.

I said the user would get higher quality this way because when a Passive TV accepts frame packed 3D, the stereo frames will pass through its 3D processing engine when getting converted to the Interleaved format. From what I have seen, when 3D is enabled in any 3D TV, the image quality suddenly degrades due to whatever processing they do to the frames. In the above suggested method, the TV will display the Row Interleaved frames like normal 2D images and Passive 3D glasses will do the magic. Also, 10 bit output will work as well.

Row Interleaved, Column Interleaved can be kept as 3D output options in madVR settings. Passive 3D TV users can select the appropriate type depending upon what kind of Passive TV they have. For Active 3D TV users, frame packed 3D will be the best option.

[XTrojan]

Maybe you're right about 4K sets, I don't know. But I don't think many 1080p sets will accept a 2160p signal, or am I wrong?
My 6y old Samsung supports this as well, the reason people wanted Frame-packed back then was because it cuts resolution in half, if it's a 1080p TV the quality becomes much worse.

Today with 4k sets that weakness isn't there anymore, 4k monoscopic tv sets don't have that weakness either.

With that in mind, frame-packed is the original format for 3D, if you can get it to playback perfectly without glitches i'd choose frame-packed, the only issue is compability as 3D MVC mode for AMD/Nvidia drivers is limited and seems to crash with 10bit output etc. The advantage with Side-by-Side 3D is that the TV handles everything and madvr can do same algoritms as it does to a 2D video.

Edit: You don't need to switch to 1920x2160, you just switch to 3840x2160, but when you switch to side by side the TV will "combine" right+left into one 3D video, by combining them both you lose half the resolution, which makes it 1920x2160.

Think of it as two 2D images, the 3D effect is one image above the other, that's what the TV basically does, but when doing so they both become same picture, aka half of the resolution gets lost either horizontally or vertically depending if it's side-by-side or over-under.

What stereoscopic player does is rendering the MVC and 2D at same time on left/right and then letting the TV convert them both into 3D. I'll try to explain with pictures later

[madshi]

No, I believe there is no need to output 1920x2160. If I understand it correctly, you will have to fit View 1 in every odd line and View 2 in every even line in a regular 1920x1080 2D frame. So the vertical resolution of both views will be reduced to 540. When you output this kind of frame to a Passive 3D TV (Row Interleaved Type), the user shouldn’t even need to tell the TV about the signal type. The user can just put on their 3D glasses and enjoy 3D. Similarly, for Column Interleaved Type Passive 3D TV, the horizontal resolution of both views will be reduced to 960 when you output Column Interleaved 1920x1080 2D frames.

I said the user would get higher quality this way because when a Passive TV accepts frame packed 3D, the stereo frames will pass through its 3D processing engine when getting converted to the Interleaved format. From what I have seen, when 3D is enabled in any 3D TV, the image quality suddenly degrades due to whatever processing they do to the frames. In the above suggested method, the TV will display the Row Interleaved frames like normal 2D images and Passive 3D glasses will do the magic. Also, 10 bit output will work as well.

Row Interleaved, Column Interleaved can be kept as 3D output options in madVR settings. Passive 3D TV users can select the appropriate type depending upon what kind of Passive TV they have. For Active 3D TV users, frame packed 3D will be the best option.
Oh ok. I guess I didn't understand how passive 3D works. So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Implementing that should be relatively easy.

@Aleksoid1978, would that do the trick for your TV?

My 6y old Samsung supports this as well, the reason people wanted Frame-packed back then was because it cuts resolution in half, if it's a 1080p TV the quality becomes much worse.
You mean your old Samsung supports 1080p side-by-side, but it doesn't support 3840x2160p23 side-by-side, does it?

Today with 4k sets that weakness isn't there anymore, 4k monoscopic tv sets don't have that weakness either.
Sure. For 4k sets side-by-side should work just fine. But the big majority of users are still using 1080p sets. Reading your previous posts I thought you were talking about sending 2160p23 side-by-side to 1080p sets. And I would be quite surprised if 1080p TVs accepted 2160p23 data. Do they?

Think of it as two 2D images, the 3D effect is one image above the other, that's what the TV basically does, but when doing so they both become same picture, aka half of the resolution gets lost either horizontally or vertically depending if it's side-by-side or over-under.

What stereoscopic player does is rendering the MVC and 2D at same time on left/right and then letting the TV convert them both into 3D. I'll try to explain with pictures later
I'm rather confused. madVR already does side-by-side rendering by default *right now* if your TV doesn't support frame packed output, or if the OS has frame packed output disabled. So why are you explaining to me how side-by-side works when madVR already does it?

[huhn]

It's basically 23fps 2160p, but when you enable "side by side" or "over under", vertical or horizontal resolution gets cut in half as it combines two sides into 3D or over/under into 3D. Most TVs should support this, it doesn't require 3D planes either since one side = one eye, and subtitles are rendered at both. The downside is that 4k is needed otherwise it's gonna be something like 960x1080 or 1920x540, no software is needed as the TV does all of the work, the player only needs to show the MVC stream at right or left and 2D stream at other, with subtitles at both.

my TV clearly said no to 3D at UHD no matter what. all options are simply grayed out. and i can't blame the screen this is not a spec.

[XTrojan]

Oh ok. I guess I didn't understand how passive 3D works. So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Implementing that should be relatively easy.

@Aleksoid1978, would that do the trick for your TV?


You mean your old Samsung supports 1080p side-by-side, but it doesn't support 3840x2160p23 side-by-side, does it?


Sure. For 4k sets side-by-side should work just fine. But the big majority of users are still using 1080p sets. Reading your previous posts I thought you were talking about sending 2160p23 side-by-side to 1080p sets. And I would be quite surprised if 1080p TVs accepted 2160p23 data. Do they?


I'm rather confused. madVR already does side-by-side rendering by default *right now* if your TV doesn't support frame packed output, or if the OS has frame packed output disabled. So why are you explaining to me how side-by-side works when madVR already does it?

Because it's not a SBS encoding, it's a remux that's 2D but contains a MVC stream. I haven't checked newest MadVR, but I thought you needed SBS encoding for it to work? Can MadVR render the 2D and MVC streams seperate on right/left without engaging frame-packed mode but letting the TV do it? I have to test that, sorry :s.

And no 1080p sets can't playback 3D content without cutting resolution in half, making it look bad, frame-packing is still the goto for 1080p TV sets and only mode without degrading picture quality, Personally I just dislike frame-packing due to compability issues and buggyness of external content such as subtitles.

[sneaker_ger]

madVR's 3D mode (in collaboration with the latest LAV nightly) currently only works with mvc-in-mkv, nothing else. It does not work with side-by-side or top-bottom encoding. madVR will output said mvc-in-mkv over HDMI with the full 1080p resolution.

[madshi]

Because it's not a SBS encoding, it's a remux that's 2D but contains a MVC stream. I haven't checked newest MadVR, but I thought you needed SBS encoding for it to work? Can MadVR render the 2D and MVC streams seperate on right/left without engaging frame-packed mode but letting the TV do it?
Yes. If the decoder sends 3D to madVR, and if you have 3D enabled in madVR, then madVR tries to do frame packed rendering. If that doesn't work, madVR renders this MVC-MKV as side-by-side. It was this way right from v0.90.0.

And no 1080p sets can't playback 3D content without cutting resolution in half, making it look bad, frame-packing is still the goto for 1080p TV sets and only mode without degrading picture quality, Personally I just dislike frame-packing due to compability issues and buggyness of external content such as subtitles.
It seems to me you haven't actually tested any of this yet with madVR. So please do, and then report your findings. As far as I can see, there should be zero difference between frame packed and side-by-side rendering when using madVR. madVR always draws subtitles and OSD on both left and right eye views in the same way.

madVR's 3D mode currently only works with mvc-in-mkv, nothing else. It does not work with side-by-side or top-bottom encoding. madVR will output said mvc-in-mkv as side-by-side over HDMI (that's the way 3D works for HDMI).
Frame packed is somewhat different to side-by-side. I think frame packed does transport the images as some sort of side-by-side, but I think there's a gap between the left/right eye images.

[sneaker_ger]

I stand corrected. Always thought HDMI 3D is side-by-side only but wiki confirms top-bottom and frame-packed are possible as well.

[huhn]

madVR can output it as side by side too.

frame packing isn't side by side.

so the current madVR version can do both.

when 3d is enabled in madVR and in the OS you get frame packing 3D.

when 3d is enabled in madVR and disabled in the OS you get side by side.

[CarlosCaco]

I having a strange problem, if upscale a 720p video to 1366x768p monitor none of my upscaling refinements is activated, i don´t see the effects on image e and OSD don´t show any changes on reders times, on others versions below 89.19 this issue don´t happen

[omarank]

Oh ok. I guess I didn't understand how passive 3D works. So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Yes, right.

Before you implement these 3D methods, I would suggest to create a test build which outputs any one type, say Row Interlaced, and get the 3D output tested by users having the same type of Passive TV. Maybe Aleksoid1978 can help in the testing.

[jmone]

When you force D3D11 in the madVR settings, do you see anything at all, before it crashes? E.g. if you enable the OSD (Ctrl+J) before activating D3D11 mode, do you see the OSD for a short time before the crash? If so, please let me know which refresh rate the OSD lists. Is it near to 23.976? Or is it near to 47.952?

It is in 23.976 and one static SBS Image. Log at:
https://dl.dropboxusercontent.com/u/59011278/madVR%20-%20log%203D%20Crash.zip

[Prinz]

Oh ok. I guess I didn't understand how passive 3D works. So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Implementing that should be relatively easy.

One thing to add: 4K TV's with passive 3D can show full 1080p resolution this way, because they have double the lines of 1080p.

[Ver Greeneyes]

Do 4k TVs with passive 3D have polarization filters applied as a coating, alternating between columns? Otherwise, I wonder how glasses would know which lines to show.

[Manni]

Is it a constant delay? E.g. exactly one frame? Or does the delay differ sometimes, from movie to movie, or even from frame to frame? If it's a constant delay then it shouldn't be too hard to implement an option to delay one eye by 1-3 frames or so.


I can't say if the delay is constant or not. The way it's best described is in the thread I already linked to a couple of times. 3D is fine on static shot and gets wrong with motion due to the delay between the two eyes.

Can you reproduce this on your AMD rig? We have the same OS (Win 8.1), so you should have the same issue. If you don't, please give details of drivers, etc.

[Asmodian]

One thing to add: 4K TV's with passive 3D can show full 1080p resolution this way, because they have double the lines of 1080p.

True, but they need to receive frame packed 3D, otherwise they never get the full resolution video. 1920x2160 or 3840x1080 input is not going to work.

Do 4k TVs with passive 3D have polarization filters applied as a coating, alternating between columns? Otherwise, I wonder how glasses would know which lines to show.

Yes, different rotational polarizations are usually used, I believe.

[Prinz]

True, but they need to receive frame packed 3D, otherwise they never get the full resolution video. 1920x2160 or 3840x1080 input is not going to work.

As far as I understand we are speaking of not telling the TV that it's 3D, instead sending it as 2D with interleaved lines. So madVR does what the TV would do. And in this case madVR could send the full 1080p 3D to a 4K passive TV in normal 3840x2160p resolution with interleaved Pictures.

[6233638]

Where did we do the previous discussion about this? In the bug tracker? I don't remember the conclusion of the our previous discussion, so I'm not sure right now if this behaviour is as intended or not.I had a chance to look this up, and it seems that I was mistaken - the behavior I saw with the "wrong" profile being selected is actually working as intended.
http://bugs.madshi.net/view.php?id=349

The srcHeight/Width behavior was only changed for Zoom Control profiles (since that actually broke things) and not all profiles.
It's not what I would personally prefer, but I think your decision to handle it this way does make sense, so I'll keep using uncroppedSrcHeight/Width for the affected profiles.

[Aleksoid1978]

So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Implementing that should be relatively easy.


I think you are right. You can try render as usually 2D picture and alternate lines for the left and right eyes(1 - left, 2 - right, 3 - left, 4 - right and so on.) Passive TV must from such image do 3D in half resolution for height(1080 -> 540).

You can try make test build - and i check it on my TV.

[huhn]

Oh ok. I guess I didn't understand how passive 3D works. So basically you're saying that a passive 3D TV when looked at without glasses has lines 1, 3 and 5 for left eye and lines 2, 4 and 6 for right eye? So I would simply render as if it were 2D and the TV wouldn't even have to know that I'm sending 3D, the user would simply put on his glasses and that's all?

Implementing that should be relatively easy.

@Aleksoid1978, would that do the trick for your TV?

this may work but what the TV is subsampling. after that there is nothing worth calling 3D/in the chroma channel.

this issue doesn't exist with top bottom and side by side.

[omarank]

this may work but what the TV is subsampling. after that there is nothing worth calling 3D/in the chroma channel.

this issue doesn't exist with top bottom and side by side.
PC mode, Game mode or Graphics mode, which doesn’t downsample chroma, can be used in the TV. 3D output this way should still be better, rather much better, than after the unnecessary processing done by the TV when its 3D processing engine is active.

Row/ Column Interleaved output is anyway going to be optional in the settings. Passive TV users can always use frame packed 3D output to compare and decide which output method they like the most.

Do 4k TVs with passive 3D have polarization filters applied as a coating, alternating between columns? Otherwise, I wonder how glasses would know which lines to show.
Yes, not only just 4K Passive TVs, all Passive 3D TVs have a special coating which, for example in case of Row Interleaved Type, polarizes the light coming from odd lines in one way and from even lines in the opposite way. Passive 3D glasses have corresponding polarization filters which ensure that one eye sees just the odd lines and the other eye sees the even lines only. As the odd lines comprise View 1 and even lines View 2, the user gets 3D vision when they put on the Passive glasses.

[apgood]

Thanks. Did you check with say chapter 2 of Gravity that the depth was correct? For example, when they circle around each other in space, is the depth correct for each of them and the rope/tether?
Haven't converted Gravity to mkv yet but PAN and hobbit unexpected journey. The 3D layers look correct and there are no out of sequence frames or anything like that.

[4h4h270]

frame drop is from reclock, disabled, everything goes fine.

-----------------

I have a problem, I always get 1 frame drop every 30+sec, I checked with gpu-z the load is only 40%~50%.
I tried lighter algorithm and raise every quene size, nothing changed.

[aufkrawall]

Do you have also drops with other renderers? I'd try EVR CP of MPC HC and display rendertime graphs with ctrl + j.

[apgood]

Sounds good. Try a few different movies. It's possible that the eye swapping depends on the movie. If that's the case, either MakeMKV or LAV need a minor fix/update to solve that issue.

Only got 2 3D movies in mkv (rest still in iso format), but tested them in Mede8er as well and they need the eye swapping setting set there as well, so I suspect it might be something in MakeMKV that I did when I created them.

If I get a chance I'll try creating some more to see (and make sure I'm running the latest version of MakeMKV).

[foozoor]

Does madVR use image doubling with DVXA image upscaling?
In the OSD, I see chroma > DVXA and image > DVXA.

[mogli]

@foozoor
No, it does not.

[foozoor]

@foozoor
No, it does not.
Thanks! What about image enhancement and upscaling refinements?

[Manni]

Haven't converted Gravity to mkv yet but PAN and hobbit unexpected journey. The 3D layers look correct and there are no out of sequence frames or anything like that.

Thanks. This is either thanks to Crimson, or because you can swap left/right eye in your projector. I can't do this on my JVC.

I'll try Crimson when I have a chance if no one else confirms AMD working on Catalyst.

[huhn]

Thanks! What about image enhancement and upscaling refinements?

image enhancement yes
upscaling refinement no

[sat4all]

Hi madshi,

Debanding algo introduce dropping frames no matter what i set on level. I can fix this only by checking: don't rerender frames when fade in/out is detected. (Must be always checked?)
I've tried with my both rigs: main htpc with an amd hd 7950 and intel nuc i5 hsw.

[nussman]

Seems like your GPU is to weak for your settings without "don't render frames when fade in/out ..."?
Take a look at the rendering times in the OSD.

[XTrojan]

Using MPC-HC, Lav 0.66 and I can't get 3D to work Frame-packed or SBS, it simply won't activate nor engage Multiview mode for SBS.

Container is .mkv with MVC 3D stream in it

I noticed you added a nits mode, I know this is something similar to HDR, but what does it do to rec709 movies? I'm not using backlight at max on my TV since rec709 colors get weird above certain nits or strains eyes.

I'm using a JS9000 Tv that should handle most 3D formats.

I'm mostly trying to get Multiview SBS mode to work since framepacked is limited to 1080p23 8bit and all that stuff, Multiview SBS is registered 2D at the PC, it's when it passes to the TV it converts to 3D by merging left+right which is same thing as merging original track + MVC track in framepacked.

[XTrojan]

I did a image edit to explain Full SBS easier.

This image is what the PC reports, the PC always thinks the movie is 2D, the TV will merge these two images into 3D. The PC will thus never be restricted by 3D setting limit such as 1080p23.

http://i.imgur.com/um8Lal8.jpg

Regarding the "mountain effect" on the image, this is the industry standard, the more you stretch the images from eachother (2D and MVC) to each eye the more of a 3D effect you get but also introducing other weird things such as crosstalk. Most TVs have a "3D depth setting" which will basically try to stretch these two images even more in frame-packed or SBS.

also yes, windowed mode doesn't work that well since there's space in the middle and when the TV merges them it becomes buggy, fullscreen is needed to avoid scaling issues.

Either way i can't get Full SBS or Framepacked 3D to work with MadVR as my post above states.

Edit2: When I mean "stretch" it's basically adding distance from both images, 1 image = 1 eye, add too much distance between them and your eyes will become asymmetrical, lol.

[Manni]

Using MPC-HC, Lav 0.66 and I can't get 3D to work Frame-packed or SBS, it simply won't activate nor engage Multiview mode for SBS.

You need a recent nightly of LAV 0.67, 0.66 doesn't support 3D.

[Thunderbolt8]

bilateral chroma upsampling in madvr, is this actually adaptive bilateral sampling or fixed bilateral sampling?

[XTrojan]

You need a recent nightly of LAV 0.67, 0.66 doesn't support 3D.

Ahh, i'll try testing it tonight then.

[aufkrawall]

Hi madshi,

Debanding algo introduce dropping frames no matter what i set on level. I can fix this only by checking: don't rerender frames when fade in/out is detected. (Must be always checked?)
I've tried with my both rigs: main htpc with an amd hd 7950 and intel nuc i5 hsw.
You need a very long queue for disabling that option, which is normal.

[XTrojan]

Got it to work, but how do you create a separate profile for 3D?

2160p 3D -> GPU can't keep up, frame drops
1080p framepacked -> 2D movies crash or switch to 1080p when I want them to switch to 2160p.

[sat4all]

Seems like your GPU is to weak for your settings without "don't render frames when fade in/out ..."?
Take a look at the rendering times in the OSD.

Rendering time is ok:
HD 7950: 20 ms 1080p 》2160p with super-xbr image doubling and Reconstruction for chroma.
NUC: 3ms 1080p 》 1080p

[sat4all]

You need a very long queue for disabling that option, which is normal.

Thanks, I will give it a try.

[leeperry]

From what I can see, the superres for chroma is quite similar to the superres for image refinement right now. However, while the algorithm is great for luma, it's not good for chroma. In luma, superres enhances small details and edges by increasing contrast (darks become darker, lights become lighter at image transients), and the same happens to the chroma layers.
Actually I really enjoy how chroma & luma SR@4 work together on the very same EE as that makes PQ crispier, smoother and sharper without nasty halos. I'm not seeing any color fringing as I get with colored dithering. I also use Jinc AR for chroma so no amplified sxbr artifacts TYVM ^^

SuperRes is very visible in motion IME and the higher the fps the smoother it gets, too bad my HD7850 can't run them both together @4 on 1080p60 content :o

[madshi]

I having a strange problem, if upscale a 720p video to 1366x768p monitor none of my upscaling refinements is activated, i don´t see the effects on image e and OSD don´t show any changes on reders times, on others versions below 89.19 this issue don´t happen
Yeah, there's a specific upscaling factor under which I don't apply the refinement. E.g. if you upscale only 1 pixel, refinement doesn't really make too much sense.

The logic is this: Upscaling refinement is meant to make sure that the upscaled image looks as near to the original image as possible. Which means that sharpening and stuff is really only needed if you upscale by a certain amount. If you find the original image already too soft, you can use image enhancements to sharpen the source.

The whole sharpening logic might be improved in a future version. I'm not sure yet how to make it more logical/intuitive yet, though.

Before you implement these 3D methods, I would suggest to create a test build
I can't make a test build without implementing these 3D methods first.

It is in 23.976 and one static SBS Image. Log at:
https://dl.dropboxusercontent.com/u/59011278/madVR%20-%20log%203D%20Crash.zip
Thanks. But you didn't my question about refresh rates.

One thing to add: 4K TV's with passive 3D can show full 1080p resolution this way, because they have double the lines of 1080p.
So they switch left/right eye per 4K line? That would be rather nice.

I can't say if the delay is constant or not. The way it's best described is in the thread I already linked to a couple of times. 3D is fine on static shot and gets wrong with motion due to the delay between the two eyes.

Can you reproduce this on your AMD rig? We have the same OS (Win 8.1), so you should have the same issue. If you don't, please give details of drivers, etc.
For testing/development I currently have my old projector throwing a small image on my development painted wall. It's not high quality enough to judge whether there's a delay or not. I can just judge whether there's any 3D effect at all or not.

I had a chance to look this up, and it seems that I was mistaken - the behavior I saw with the "wrong" profile being selected is actually working as intended.
http://bugs.madshi.net/view.php?id=349

The srcHeight/Width behavior was only changed for Zoom Control profiles (since that actually broke things) and not all profiles.
It's not what I would personally prefer, but I think your decision to handle it this way does make sense, so I'll keep using uncroppedSrcHeight/Width for the affected profiles.
Thanks for looking that up. Yeah, after reading my comment there, I think I still have the same view, so I don't think I want to change anything there.

I think you are right. You can try render as usually 2D picture and alternate lines for the left and right eyes(1 - left, 2 - right, 3 - left, 4 - right and so on.) Passive TV must from such image do 3D in half resolution for height(1080 -> 540).

You can try make test build - and i check it on my TV.
I'll give it a try, we'll see how it goes. Not sure yet if I can implement it this weekend. Maybe yes, maybe next weekend.

When using PowerDVD, does your TV know when PowerDVD switches into 3D mode and when not? Or does the TV not know?

this may work but what the TV is subsampling. after that there is nothing worth calling 3D/in the chroma channel.

this issue doesn't exist with top bottom and side by side.
You're quite right. However, luckily for row interleaved 3D 4:2:2 doesn't have this issue because 4:2:2 is only subsampled horizontally. So there's only a problem if the TV converts back to 4:2:0. As long as it at least stays in 4:2:2, there's no problem (for row interleaved 3D). Column interleaved 3D would already suffer when going back from 4:4:4 to 4:2:2, but right now I don't think I'll implement column interleaved 3D, anyway.

Yes, not only just 4K Passive TVs, all Passive 3D TVs have a special coating which, for example in case of Row Interleaved Type, polarizes the light coming from odd lines in one way and from even lines in the opposite way. Passive 3D glasses have corresponding polarization filters which ensure that one eye sees just the odd lines and the other eye sees the even lines only. As the odd lines comprise View 1 and even lines View 2, the user gets 3D vision when they put on the Passive glasses.
Are there (m)any passive displays with column interleaved 3D? I think the usual LCD panels (from LG etc) are all row interleaved?

frame drop is from reclock, disabled, everything goes fine.

I have a problem, I always get 1 frame drop every 30+sec, I checked with gpu-z the load is only 40%~50%.
I tried lighter algorithm and raise every quene size, nothing changed.
Are you talking about the Ctrl+J OSD saying "1 frame drop every ..."? Or are you actually getting those frame drops (OSD increasing "dropped frames" statistic regularly every 30+ seconds)? The former is just a rough estimate by the OSD. The key statistic is the "dropped frames" line. As long as that doesn't increase that often, you're fine. If that one does increase regularly, you may have to use custom resolutions to optimize the display mode's refresh rate.

Only got 2 3D movies in mkv (rest still in iso format), but tested them in Mede8er as well and they need the eye swapping setting set there as well, so I suspect it might be something in MakeMKV that I did when I created them.

If I get a chance I'll try creating some more to see (and make sure I'm running the latest version of MakeMKV).
As mentioned earlier, it could be a bug in MakeMKV or in LAV. I suppose probably nevcairiel will get to the bottom fo that. If you have this problem for *ALL* movies, that would be somewhat weird, though. It would suggest that either LAV or madVR always output frames in the wrong order. In that case it could also be a bug in madVR.

Thanks. This is either thanks to Crimson, or because you can swap left/right eye in your projector. I can't do this on my JVC.

I'll try Crimson when I have a chance if no one else confirms AMD working on Catalyst.
Please let me know the results. If you still have a delay with Crimson, I can try to implement a delay option. I'd rather avoid spending the time, though, if Crimson already fixes that issue.

Debanding algo introduce dropping frames no matter what i set on level. I can fix this only by checking: don't rerender frames when fade in/out is detected. (Must be always checked?)
You need a very long queue for disabling that option, which is normal.
^ aufkrawall is right. That specific options requires you to increase your GPU/CPU queue sizes a bit.

I noticed you added a nits mode, I know this is something similar to HDR, but what does it do to rec709 movies?
Nothing, it's only for HDR content.

I'm using a JS9000 Tv that should handle most 3D formats.

I'm mostly trying to get Multiview SBS mode to work since framepacked is limited to 1080p23 8bit and all that stuff, Multiview SBS is registered 2D at the PC, it's when it passes to the TV it converts to 3D by merging left+right which is same thing as merging original track + MVC track in framepacked.
Your display is active 3D, though, right? In that case I think frame packed 3D should be the best option.

Got it to work, but how do you create a separate profile for 3D?

2160p 3D -> GPU can't keep up, frame drops
1080p framepacked -> 2D movies crash or switch to 1080p when I want them to switch to 2160p.
The next build will have a "3D" profile boolean variable.

I'll try to improve display mode switching for 3D in the next build. The crash is a bug in the NVidia drivers, though. We should be able to work around in your setup by avoiding to play 2D movies with frame packed output. Simply switching to 1080p23 for frame packed 3D output, and switching to 2160pXX for 2D playback should do the trick, I think.

bilateral chroma upsampling in madvr, is this actually adaptive bilateral sampling or fixed bilateral sampling?
What's the actual difference? Your question implies that "adaptive bilateral sampling" and "fixed bilateral sampling" were known terms that exactly describe a specific algorithm, but I don't think it's as simple as that.

Actually I really enjoy how chroma & luma SR@4 work together on the very same EE as that makes PQ crispier, smoother and sharper without nasty halos. I'm not seeing any color fringing as I get with colored dithering. I also use Jinc AR for chroma so no amplified sxbr artifacts TYVM ^^

SuperRes is very visible in motion IME and the higher the fps the smoother it gets, too bad my HD7850 can't run them both together @4 on 1080p60 content :o
Have you tried "reconstruction - soft" instead of Jinc?

[Prinz]

So they switch left/right eye per 4K line? That would be rather nice.

Yes they do exactly that.

[omarank]

Column interleaved 3D would already suffer when going back from 4:4:4 to 4:2:2, but right now I don't think I'll implement column interleaved 3D, anyway.


Are there (m)any passive displays with column interleaved 3D? I think the usual LCD panels (from LG etc) are all row interleaved?

I don’t know. Well, the thing is, any software that offers Row Interleaved output, offers Column Interleaved as well. Below is a quote from AMD HD3D_Whitepaper (https://www.google.co.in/url?sa=t&rct=j&q=&esrc=s&source=web&cd=12&ved=0ahUKEwj2ivuX8K7KAhVMA44KHf8wAiM4ChAWCCAwAQ&url=https%3A%2F%2Fwww.amd.com%2FDocuments%2FHD3D_Whitepaper.pdf&usg=AFQjCNGcXp5WJyo377npml7vNBKKXWHAWQ&bvm=bv.112064104,d.c2E&cad=rja)that you may find interesting:

“The column interleaving mode is convenient as some auto-stereoscopic panels (such as those developed by Sharp and others) use a parallax barrier to ensure that alternating columns are viewed by the left and right eyes. Content that is column-interleaved can thus be presented without any decoding.”

[Thunderbolt8]

What's the actual difference? Your question implies that "adaptive bilateral sampling" and "fixed bilateral sampling" were known terms that exactly describe a specific algorithm, but I don't think it's as simple as --that.I dont know, but at least the conclusion here might suggest this: http://www.interdigital.com/download/54313edae262284525000224