Media Player .NET (MPDN) - D3D HQ GPU Video Renderer [v2.49.0/v1.31.0 27 Dec 2018]

[Anima123]

It means downscaling I believe.

[aufkrawall]

Yes, but it shouldn't use Triangle (bilinear) since SSIM is in the script chain after SSSR.

Edit: With new Fluid Motion there are still sometimes delayed frames reported:
http://abload.de/img/delayed8iuh8.png (http://abload.de/image.php?img=delayed8iuh8.png)
Happens mostly when the GPU load is very high, but queues are always full.

[Shiandow]

Yes, but it shouldn't use Triangle (bilinear) since SSIM is in the script chain after SSSR.

If you want that you should use SSIM inside the prescaler, not after SSSR. Right now it only tries to downscale after using SSSR. I'm working on a new Tag system that hopefully will make the distinction clearer. What you want would look something like this:
http://i.imgur.com/TTF4i7h.png


Edit: With new Fluid Motion there are still sometimes delayed frames reported:
[...]
Happens mostly when the GPU load is very high, but queues are always full.

Seems it's too busy with something to get the frame out in time. Probably not an easy fix, unless it turns out to be another timing issue.

[aufkrawall]

If you want that you should use SSIM inside the prescaler, not after SSSR. Right now it only tries to downscale after using SSSR. I'm working on a new Tag system that hopefully will make the distinction clearer. What you want would look something like this:
http://i.imgur.com/TTF4i7h.png

Could you please provide me with an example script? :o

[Shiandow]

Could you please provide me with an example script? :o

The following should do the trick:


<ListOfPresetOfITextureFilterIRenderScript xmlns:yaxlib="http://www.sinairv.com/yaxlib/">
<PresetOfITextureFilterIRenderScript Name="SSSR" Guid="de825cc3-332d-4848-9d5f-588774c4d8ae">
<Script yaxlib:realtype="Mpdn.Extensions.RenderScripts.Shiandow.SuperRes.SSSRUi">
<Settings>
<PrescalerGroup>
<SelectedIndex>0</SelectedIndex>
<Hotkey></Hotkey>
<Options>
<PresetOfITextureFilterIRenderScript Name="OpenCL + SSIM" Guid="be40a0fc-efde-473d-9e5d-219af6aeef83">
<Script yaxlib:realtype="Mpdn.Extensions.RenderScripts.Mpdn.ScriptChain.ScriptChainScript">
<Settings>
<Options>
<PresetOfITextureFilterIRenderScript Name="OpenCL NNEDI3" Guid="943b652d-977e-43db-a2fa-a97db3ec4dbd">
<Script yaxlib:realtype="Mpdn.Extensions.RenderScripts.Mpdn.OclNNedi3.OclNNedi3Scaler">
<Settings>
<Neurons1>Neurons16</Neurons1>
<Neurons2>Neurons16</Neurons2>
<ChromaScalers />
<ChromaScalerGuid>00000000-0000-0000-0000-000000000000</ChromaScalerGuid>
</Settings>
</Script>
</PresetOfITextureFilterIRenderScript>
<PresetOfITextureFilterIRenderScript Name="SSIM downscaler" Guid="7e600065-baec-44bd-9394-4c376a96b0c3">
<Script yaxlib:realtype="Mpdn.Extensions.RenderScripts.Shiandow.SSimDownscaling.SSimDownscalerUi">
<Settings>
<SoftDownscaling>False</SoftDownscaling>
<Strength>0.5</Strength>
</Settings>
</Script>
</PresetOfITextureFilterIRenderScript>
</Options>
</Settings>
</Script>
</PresetOfITextureFilterIRenderScript>
</Options>
</PrescalerGroup>
<Passes>2</Passes>
<LinearLight>False</LinearLight>
<OverSharp>0</OverSharp>
<Locality>4</Locality>
<Mode>Hybrid</Mode>
</Settings>
</Script>
</PresetOfITextureFilterIRenderScript>
</ListOfPresetOfITextureFilterIRenderScript>

[aufkrawall]

Dam, I again forgot that one can add script chains as prescalers. *facepalm*
Working now. :o

[Garteal]

Hi everyone,
Since recently I'm now also involved in the development of MPDN itself. I'm still getting familiar with the code, so any fundamental changes will be a ways off.

I did, however, manage to make some small improvements to MPDN's rendering. To quote the changelog:

Niiice! Going to to be using this build for playback when I get to it.

I also tried your new deband method, and it definitely produces better results.
Old (https://abload.de/img/old-dtxunu.png) // New (https://abload.de/img/new-dzguwp.png)
ScreenshotComparison (http://screenshotcomparison.com/comparison/189490)

Edit: Btw enabling "add grain" results in a compile error. Somethings undefined.

[Shiandow]

I also tried your new deband method, and it definitely produces better results.
Old (https://abload.de/img/old-dtxunu.png) // New (https://abload.de/img/new-dzguwp.png)
ScreenshotComparison (http://screenshotcomparison.com/comparison/189490)

Edit: Btw enabling "add grain" results in a compile error. Somethings undefined.

You're early, I hadn't even pushed that to the official repository yet. Glad you like the results though. That compilation error should be fixed in the official version. I'm probably removing the 'add grain' option entirely (unless it turns out to be way more popular than anticipated, but then I'd still make some changes).

[aufkrawall]

New deband, is it christmas already?
Thx a lot, Shiandow. :)

[Anima123]

Is the new deband filter in render script or not? I haven't noticed it yet.

[Shiandow]

Is the new deband filter in render script or not? I haven't noticed it yet.

It's on my personal github, it hasn't really been officially released yet.

[ryrynz]

I also tried your new deband method, and it definitely produces better results.
Old (https://abload.de/img/old-dtxunu.png) // New (https://abload.de/img/new-dzguwp.png)
ScreenshotComparison (http://screenshotcomparison.com/comparison/189490)

Probably not a good test there as there's no real banding in that image to speak of. Most of the change occurs in the top left where I can see that the pixels are averaged out more but it's basically imperceptible.

[huhn]

both images use an deband filter so no banding.

[Garteal]

Yup what ^ said.

Speaking of which, Shiandow, is there anything that can be done against banding flickering? It's quite noticeable in motion, especially on sources with horrible banding.

[Shiandow]

Speaking of which, Shiandow, is there anything that can be done against banding flickering? It's quite noticeable in motion, especially on sources with horrible banding.

Not easily, usually the source itself is flickering as well. It might also be a consequence of the recursive debanding method I use. Or it may even be inherent to debanding large gradients, since those change quite a bit with small changes.

Anyway, if possible I'd like to wait with reviewing the new debanding method until there's actually an official release of it.

[aufkrawall]

Here someone created some juddertest videos:
http://www.oc-burner.de/ftp/Videos/juddertest/
I sometimes notice jumping even though MPDN doesn't report any anomalies. madVR looks smoother to me.

[ryrynz]

Yup what ^ said.


I forgot when I started posting that I was looking at two debanded images.. any way, what I said still stands that the resulting images are near identical.
I'm sure you could find something that shows more significantly the improvements in the new version :P

[Zachs]

Just released Shiandow's latest improvements v2.48.0. Enjoy!

[Shiandow]

Hi all,

Please let me know if this latest release fixes the remaining timing related issues with fluid motion.

Also, I've noticed that the difference between SSSR in gamma light, and SSSR in linear light is a lot smaller than I originally though, could anyone confirm or deny this?

[Hrxn]

FYI, the automatic update check of MPDN is currently not working: 500 - Internal Server Error

[Magik Mark]

Hi Shiandow!

Is there hope to add "Stop, Pause, Play & Exit" OSD buttons during FSE?

Thanks a lot for your work here. We really appreciate it

[Shiandow]

FYI, the automatic update check of MPDN is currently not working: 500 - Internal Server Error

Seems to be some trouble with the server, the links on the first page should still work.


Is there hope to add "Stop, Pause, Play & Exit" OSD buttons during FSE?


That should be possible, I'll have a look when I have some time. It could probably be done with an extension, if anyone else wants to give it a try.

[Magik Mark]

That should be possible, I'll have a look when I have some time. It could probably be done with an extension, if anyone else wants to give it a try.

Thanks a Lot! Looking forward to it

[aufkrawall]

Please let me know if this latest release fixes the remaining timing related issues with fluid motion.

It seems to, have to try a few more times to confirm.


Also, I've noticed that the difference between SSSR in gamma light, and SSSR in linear light is a lot smaller than I originally though, could anyone confirm or deny this?
In madVR with SuperRes, the difference is only very noticable with large scaling factors (doubling/quadrupling/...) and content with fine bright or dark structures.

[Anima123]

It seems the audio volume control didn't apply when starting a video file, maybe just before the buffer been filled.

Edit: Just noticed that it happened with the Sanear Audio renderer, not with the good ole directsound audio renderer.

[Zachs]

In that case it's most likely a Sanear bug.

[Shiandow]

Seems the new debanding algorithm may have to wait a bit. I just figured out what the technique I'd partially rediscovered is called (pyramid decomposition). Using it properly I managed to get an algorithm with even better quality in about a day. It would take too long to get the new version completely finished, so I will probably just release an experimental version so you guys can compare.

[kingpage]

There's new upscaling algorithm called RAISR (Rapid and Accurate Image Super Resolution).

Romano, et al ‎2016 Paper (https://arxiv.org/pdf/1606.01299v3.pdf)

Supplementary Results (https://drive.google.com/file/d/0BzCe024Ewz8ab2RKUFVFZGJ4OWc/view)

Google Research Blog Article (https://research.googleblog.com/2016/11/enhance-raisr-sharp-images-with-machine.html)

It seems more than fast enough for real-time video rendering and yet comparable to the very best upscaling techniques out there (such as A+ & SRCNN).

They report a runtime of about 20-30ms for 630K-pixel upscaled images on a Xeon 6-core CPU. They also implemented another version that used mobile GPU and ran at 200Mpix/s, which is like 10ms at 1080p or 40ms at 2160p.

Based on the above, it should be a lot faster on a modern desktop GPU (<5ms). Anyone with the know-how interested to write a script to see how it performs in MPDN?

Edit: temporal artifacts could be a problem as it's never been used for video rendering, perhaps some additional filter is needed to ensure consistent changes to each frame. The authors did say their new method not only doesn't introduce artefacts, it can reduce them via learning. What do you think?

In terms of getting the trained filters and/or code, the lead author can be reached here (http://www.cs.technion.ac.il/~yromano/#software).

[Shiandow]

It might be fast, but so far I've not seen any results from it that actually impress me. In most cases their image seems similar to what I can get using SSSR.

[kingpage]

What about in terms of artifacts? SSSR is very sharp, but it seems to exacerbate compression artifacts and film grain, etc. RAISR is supposed to be able to "suppress the aliasing in addition to alleviating moderate compression artifacts", by using moderate (95 quality JPEG) level of compression in training. I loved SSSR but for some movies, it seems to oversharpen everything. How do you use SSSR (for both upscaling e.g. from 720p to 1080p and native resolution e.g. 1080p)?

I used to use Jinc for both upscaling and downscaling, now I use Sinc-Jinc for that. I still don't know whether I like NEDDI3, SSSR, Super-xBR or SuperRes better. How would you describe their differences?

BTW, I've seen screenshots of render chain OSD, is there a hotkey for that?

Edit: I came across another one with great potential. FSRCNN (http://mmlab.ie.cuhk.edu.hk/projects/FSRCNN.html), even better (higher PSNR & SSIM) than SRCNN but much faster (seems to be quite a bit faster than RAISR from a quick glance). There's matlab code and training code and dataset included. I would love to be able to implement these myself, but I don't really know where to start. Is there somewhere I can learn how to write render scripts for MPDN or in general?

[BetA13]

what about in terms of artifacts? Sssr is very sharp, but it seems to exacerbate compression artifacts and film grain, etc. Raisr is supposed to be able to "suppress the aliasing in addition to alleviating moderate compression artifacts", by using moderate (95 quality jpeg) level of compression in training. I loved sssr but for some movies, it seems to oversharpen everything. How do you use sssr (for both upscaling e.g. From 720p to 1080p and native resolution e.g. 1080p)?

I used to use jinc for both upscaling and downscaling, now i use sinc-jinc for that. I still don't know whether i like neddi3, sssr, super-xbr or superres better. How would you describe their differences?

Btw, i've seen screenshots of render chain osd, is there a hotkey for that?



ctrl/strg - k

[Shiandow]

What about in terms of artifacts? SSSR is very sharp, but it seems to exacerbate compression artifacts and film grain, etc. RAISR is supposed to be able to "suppress the aliasing in addition to alleviating moderate compression artifacts", by using moderate (95 quality JPEG) level of compression in training. I loved SSSR but for some movies, it seems to oversharpen everything. How do you use SSSR (for both upscaling e.g. from 720p to 1080p and native resolution e.g. 1080p)?

Lowering compression artefacts would be nice, but is not really the job of a scaler if you ask me.

The default settings of SSSR are designed for scaling 720p to 1080p. Be sure to use the 'soft' or 'hybrid' modes, as the 'sharp' mode is indeed pretty sharp. In theory it can't oversharpen (it tries to ensure the contrast in the final image is the same as in the original) but you never know; using a softer prescaler (super-xbr, or even NEDI) could help.


I used to use Jinc for both upscaling and downscaling, now I use Sinc-Jinc for that. I still don't know whether I like NEDDI3, SSSR, Super-xBR or SuperRes better. How would you describe their differences?

They're sharper, without adding much aliasing or ringing (although SuperRes can have some problems with ringing). They each have their own advantages and disadvantages though, you really need to decide for yourself what you like best.


Edit: I came across another one with great potential. FSRCNN (http://mmlab.ie.cuhk.edu.hk/projects/FSRCNN.html), even better (higher PSNR & SSIM) than SRCNN but much faster (seems to be quite a bit faster than RAISR from a quick glance).

So far I'm more partial to methods based on this paper (https://pdfs.semanticscholar.org/7df0/39049948d54fd1f4d75526743f315cf4865f.pdf).


There's matlab code and training code and dataset included. I would love to be able to implement these myself, but I don't really know where to start. Is there somewhere I can learn how to write render scripts for MPDN or in general?

Unfortunately, no. That said most of the work is basically knowing how to write and use shaders, the easiest way to start with that is probably modifying existing shaders (like the ones in the MPDN extensions folder) to see what effect this has, then move on to writing your own shaders and finally learn how to connect everything to MPDN. Now I can help with that last step, but learning how to write shaders is going to take most of the time. Also, implementing something like FSRCNN is going to take some effort, you need to somehow fit the calculations into shaders (I suppose you could use OpenCL instead, but I'm not too sure you'll find it easier).

[kingpage]

So far I'm more partial to methods based on this paper (https://pdfs.semanticscholar.org/7df0/39049948d54fd1f4d75526743f315cf4865f.pdf).


https://github.com/huangzehao/Super-Resolution.Benckmark

Benchmark results tend to suggest external example-based methods are superior to internal example-based methods. I can sort of understand why it is harder to implement something like FSRCNN than a method that doesn't need training. Literature seems to favour convolutional networks.

I know SSSR stands for Structural Similarity (SSIM) SuperRes. Could you tell me which papers are SSSR, Super-xBR and SuperRes based on?

Edit: Found your post about SuperRes (http://forum.doom9.org/showthread.php?p=1685124#post1685124) which is based on Lukin et al. 2006 (http://www.graphicon.ru/html/2006/proceedings/papers/we08_94_LukinKrylovNasonov.pdf). xBR stands for scale by rules, Super-xBR is developed by Hylian 2015 (https://drive.google.com/file/d/0B_yrhrCRtu8GYkxreElSaktxS3M/view?pref=2&pli=1).

[BetA13]

hey there,

just a quick question:

is there any differnces in using dx11 api over dx9e?

do i loose any features? or is it more optimized?

the thing why i ask, is because fullscreen exclusive mode only works with dx9e here..when i try it with dx11 my screen goes fullscreen But i only see 2/3 of the picture, its kinda too big, dunno..

And i was just wondering, if i stay with dx9 since it works for me, is there anything thats different now besides the api?

THX..

[kingpage]

Is the forced image doubling option only available for NEDI? Is there a way to force it to use SSSR, SuperRes, etc for 1080p material rendering at native 1080p? Nothing seems to show up on the render chain OSD except chroma upscaling (or forced NEDI) unless the source material is <1080p.

[Shiandow]

just a quick question:

is there any differnces in using dx11 api over dx9e?

do i loose any features? or is it more optimized?

The only difference is how it tells windows to display the images, you shouldn't lose/gain any features by changing the renderer. Performance also shouldn't differ much.

[BetA13]

very good..

Thanks a lot, like allways, greetz Beta

[Shiandow]

Is the forced image doubling option only available for NEDI? Is there a way to force it to use SSSR, SuperRes, etc for 1080p material rendering at native 1080p? Nothing seems to show up on the render chain OSD except chroma upscaling (or forced NEDI) unless the source material is <1080p.

Currently there isn't a way to do that. Although it's fairly easy to change NNEDI3's code to make it always double if you want it to. All you'd have to do is remove the line that checks if the target size is bigger than the input.

[RyaNJ]

Hi Guys.

Since yesterday I've been getting an issue that I have absolutely no idea how to resolve. The issue prevents MDPN from even loading correctly - let alone playing anything. Every time when I attempt to load MPDN the following screen is displayed:

http://i.imgur.com/EBCS6c5.png

Followed by a bit of other information, none of which helps me figure out exactly what has apparently happened here. This box appears right after the optimising extensions message is shown. Upon closing MPDN the following message then appears:

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

However the program doesn't unload and has to be forcibly terminated.

What on earth has happened here and does anyone have any ideas how I can fix this?

[dipje]

All I know is I had the same thing, tried searching for it but couldn't find anything (it doesn't help that doom9's search-in-thread seems a bit quick to reject certain words these days).

Then I took the 32bit version, which worked fine.


To be honest - and it might be my old-skool coder thing and not getting all the dotNet stuff - but the 'extensions' are supplied as a bunch of files + support DLLs (Mpdn.Extensions.dll + Mpdn.Extensions.Framework.dll + a whole lot of stuff in the Libs\ folder). If I download the 'mpdn extensions' there are only single versions of these files, and I don't think it's possible to make a single DLL for both x86 and x64.. so the whole extensions seem to be compiled for x86, and of course they are not going to work with x64. Just me thinking out loud, no clue if this is true or not.

MPDN x64 just the bare installer (so, MediaPlayerDotNet_x64_2_48_0_3603_Installer.exe) works fine. The moment I try to install extensions (or some of the extensions) I get the same as you. Using the x86 version seems to 'fix' it.

[RyaNJ]

All I know is I had the same thing, tried searching for it but couldn't find anything (it doesn't help that doom9's search-in-thread seems a bit quick to reject certain words these days).

Then I took the 32bit version, which worked fine.


To be honest - and it might be my old-skool coder thing and not getting all the dotNet stuff - but the 'extensions' are supplied as a bunch of files + support DLLs (Mpdn.Extensions.dll + Mpdn.Extensions.Framework.dll + a whole lot of stuff in the Libs\ folder). If I download the 'mpdn extensions' there are only single versions of these files, and I don't think it's possible to make a single DLL for both x86 and x64.. so the whole extensions seem to be compiled for x86, and of course they are not going to work with x64. Just me thinking out loud, no clue if this is true or not.

MPDN x64 just the bare installer (so, MediaPlayerDotNet_x64_2_48_0_3603_Installer.exe) works fine. The moment I try to install extensions (or some of the extensions) I get the same as you. Using the x86 version seems to 'fix' it.

Using the full 32-bit install does seem to fix it, the issue seems to only exist on the 62-bit version. Well spotted.

Another bug to report here. If I set MDPN to auto-load a video from last played position on startup it continues playing from there - even if I attempted to launch another media player file externally.

I would have expected it to do something different, like open the file that was expected to be launched instead. Is this a bug or a feature?

[Shiandow]

Hi Guys.

Since yesterday I've been getting an issue that I have absolutely no idea how to resolve. The issue prevents MDPN from even loading correctly - let alone playing anything. Every time when I attempt to load MPDN the following screen is displayed:

[...]

Followed by a bit of other information, none of which helps me figure out exactly what has apparently happened here. This box appears right after the optimising extensions message is shown. Upon closing MPDN the following message then appears:

[...]

However the program doesn't unload and has to be forcibly terminated.

What on earth has happened here and does anyone have any ideas how I can fix this?

All I know is I had the same thing, tried searching for it but couldn't find anything (it doesn't help that doom9's search-in-thread seems a bit quick to reject certain words these days).

Then I took the 32bit version, which worked fine.


Using the full 32-bit install does seem to fix it, the issue seems to only exist on the 62-bit version. Well spotted.

My best guess is that the settings are somehow causing a problem. Which shouldn't happen obviously, but that might explain why it behaves differently for the 32- and 64-bit versions.

You could check this by removing the settings for the 64 bit version and see if that fixes things.


Another bug to report here. If I set MDPN to auto-load a video from last played position on startup it continues playing from there - even if I attempted to launch another media player file externally.

I would have expected it to do something different, like open the file that was expected to be launched instead. Is this a bug or a feature?

That's a feature of the playlist right? You should probably post this as an issue on the MPDN Extensions github, hopefully Garteal can find some time to fix it, if it's not working as intended.

[dipje]

My best guess is that the settings are somehow causing a problem. Which shouldn't happen obviously, but that might explain why it behaves differently for the 32- and 64-bit versions.

You could check this by removing the settings for the 64 bit version and see if that fixes things.

I had this issue while _trying out_ MPDN for the very first time. It was a clean install, never installed it before. 64bit was my first attempt with the extensions-installer. Had the same issue, unworkable. Tried searching around, found no solution besides cleaning stuff from the usual setting-places (looking for mpdn-related folders in %appdata%, %localappdata%, %programdata% and registry). Nothing fixed it besides going 32bit.

[Zachs]

It sounds to me like a corrupt windows installation. Installing it on a cleanly installed windows should fix it.
To test the theory, just try it out on a VM.

[EncodedMango]

I had this issue while _trying out_ MPDN for the very first time. It was a clean install, never installed it before. 64bit was my first attempt with the extensions-installer. Had the same issue, unworkable. Tried searching around, found no solution besides cleaning stuff from the usual setting-places (looking for mpdn-related folders in %appdata%, %localappdata%, %programdata% and registry).
Same here, though I haven't tried MPDN yet because of that 64 bit failure, never thought it was an x64 issue.

[dipje]

I see my post was lost again.

@Zachs you were right that the 64bit version works fine with extensions in a clean installed win10-anni version.

But my win10 install is far from corrupt . The fact that the 32bit version works fine proves that it can't be a disabled service or something.

Is there a way to get more debugging info ? See which extension fails to load and why ?

I'm guessing it's a DLL file in my system32 or my devenvironment is causing a conflict.

I'm apparently not the only one with the issue.

[trandoanhung1991]

I have a question.

I distinctly remember it being said that when a script chain is activated, it won't go through the included upscalers/downscalers. Is this not the case with script groups?

Because what I'm seeing is the downscaling script group, and on top of it the internal downscaler, whatever it's set to.

EDIT: Nevermind, brainfarted. I had only chroma scaling in the script groups, so luma scaling is still being done internally. D'oh!

[Hrxn]

[..]

I distinctly remember it being said that when a script chain is activated, it won't go through the included upscalers/downscalers. Is this not the case with script groups?

Because what I'm seeing is the downscaling script group, and on top of it the internal downscaler, whatever it's set to.

[..]

What do you mean by included scalers? Why would the script not use the scalers?

[Zachs]

I see my post was lost again.

@Zachs you were right that the 64bit version works fine with extensions in a clean installed win10-anni version.

But my win10 install is far from corrupt . The fact that the 32bit version works fine proves that it can't be a disabled service or something.

Is there a way to get more debugging info ? See which extension fails to load and why ?

I'm guessing it's a DLL file in my system32 or my devenvironment is causing a conflict.

I'm apparently not the only one with the issue.

Try running dbgview and see what error messages MPDN dump out when it encounters the error. Yes I agree it would be helpful if the loader can isolate failed extensions.

My feeling is something it requires in 64 bit is missing but present on 32. Gut feeling says it's the JavaScript engine failing to resolve dependencies.

[Sephiroth_FF]

Hi @Zachs, I am really sorry for bothering you again about this but unfortunately, NVIDIA has made thier HDR SDK exclusive to DirectX 11 and as far as I know MPDN is the only DirectX 11 video renderer. I would like to know if you can have a look at this SDK https://developer.nvidia.com/high-dynamic-range-display-development and see if it is feasible to integrate it in MPDN.

Once again sorry for my insistence
Thank you very much
Sephiroth_FF