Media Player Classic - BE Win32/x64

[huhn]

good idea to use that test.
you should try again with the nv12 version DXVA doesn't like RGB input and falls usual back to bilinear with that.

edit: looked at the wrong line it is nv12 so DXVA on 950 and your driver is totally useless. sad stuff nvidia...

edit2: interesting:
https://abload.de/img/d3d9buggyl1c6l.png

[Sunspark]

It was fun for me to compare scaling of the 1-bit image. Some algorithms are so bad at it.

For reference, this is what it looks like scaled on an Intel Broadwell (HD Graphics 6000) CPU using D3D11 decoding, D3D11 scaling. Appears to be closer to v0lt's catmull scaler than the lanczos one which has more rounded edges, though obviously not identical to either. So, actually pretty good for this type of challenge.

https://dl.dropboxusercontent.com/s/8wj0q4vwjvj252w/mpcvrdx11intel.png

[huhn]

i guess that's the worst result yet.

[Sunspark]

i guess that's the worst result yet.

I'm not so sure that it is.. it doesn't have spindly needles with white hotspots of lens flare or whatever that is, etc. The geometry is closer to the original pixel width/lengths, etc.

https://dl.dropboxusercontent.com/s/lrjq9md2962a7gu/origimage.png

This is v0lt's image of what nvidia does.. what the hell..

https://dl.dropboxusercontent.com/s/3b31m4bqcy6mdo6/v0lt.png

[huhn]

that sharp one is NN it gives a false positive terrible scaler.
volts is bad bilinear which is very low end.
your intel image is just distorted with tons of error in it but the scaler itself is not that bad my it is lanczos with 8 bit int math which results in these terrible bands esle it looks similar to lanczos 3.
my nvidia does bad to because it has some issues similar to the old jinc2w with a proper "kernel".

the jinc2m which is jinc2 with PI as "kernel" does best here. (happy little accident that scaler)

while you see heavy aliasing in the source it doesn't mean it should be there it's just a limitation of the resolution a real pixel isn't square as seen in the image they are "round" and supposed to be infinity small.

[v0lt]

your intel image is just distorted with tons of error in it but the scaler itself is not
Intel uses bicubic interpolation (one of the types). The "gradient distortion" is most likely caused by too much magnification (I think a highly optimized algorithm is being used). And perhaps they will not be with a small increase.

[v0lt]

I want to warn you, bw18x18_nv12.avi is not a simple interpolation test. Due to its small size, it can cause problems in some renderer modes (distortions on the right and bottom edges). Video frames of 32x32, 64x64, or 128x128 are likely to be more compatible.

[Sunspark]

I want to warn you, bw18x18_nv12.avi is not a simple interpolation test. Due to its small size, it can cause problems in some renderer modes (distortions on the right and bottom edges). Video frames of 32x32, 64x64, or 128x128 are likely to be more compatible.

You are 100% right.

I took the image and made it 2x, 4x and 8x. It really makes a difference. The more pixels it has to work with, the sharper it is.

Below image is 2x (36x36) DXVA2 scaled in MPCVR to Window. A lot clearer simply by doubling the source.

https://dl.dropboxusercontent.com/s/ks1a8fu2yqdc3d4/Screenshot%202023-02-20%20235354.png

Video links for your convenience:

2x 36x36 (https://dl.dropboxusercontent.com/s/59a9ot0y1ziwbh2/bw36x36_nv12.avi)

4x 72x72 (https://dl.dropboxusercontent.com/s/q38du03625pypkf/bw72x72_nv12.avi)

8x 144x144 (https://dl.dropboxusercontent.com/s/3ya8ntojq4ky573/bw144x144_nv12.avi)

[huhn]

the problem isn't the sharpness it is the error you should test when it stops happening.
nvidia is outputting 10 bit which could help.
intel is still doing terrible in this test the scaling factor is still enormous.
so while the scaler in theory is better it has a bad implementation which heavy errors in this test.
mpcVR does bicubic with no error and the performance impact should be very limited.

32 source nvidia DXVA2: https://abload.de/img/nvidia32source1gcw4.png
still much much better than intel and much bigger scaling factor because of resolution and still errors.

[huhn]

AMD is currently bilinear in d3d9 but no banding.
while d3d11 is catmull rom so bicubic which is pretty soft.
some displacement error and similar to intel banding issue.
https://abload.de/img/amd4eid5.png

so that leaves the original question why is the nvidia scaler only ok in d3d9 mode and why doesn't it work for 900 series and for video with an actual resolution like 480-2160p.

edit: found a bug. should be the GPU driver.
https://abload.de/img/amd26pfxg.png
d3d11 accepts RGB but thinks it is YCbCr 444.
it's also amazing seeing this rendering doing all this at like 19 mhz and 1% GPU load truly impressive.

[v0lt]

I made another 32x32 sample - pattern2_bw32x32_nv12.mkv (https://www.sendspace.com/file/otcxqt).
madVR DXVA upscaling is not buggy on this sample.

These patterns are designed to recognize and test how interpolation works. It is wrong to use it for quality assessment. "Intellectual" scalers can produce artifacts on these samples.

[v0lt]

I took the image and made it 2x, 4x and 8x. It really makes a difference. The more pixels it has to work with, the sharper it is.
OK. You no longer have the bugs that were on 18x18. But your test samples no longer have rhombus, now there are steps of 2, 4, 8 pixels. And it doesn't work as originally intended.

[huhn]

the 32x32 links to 18x18

[Sunspark]

Unrelated, but I was bemused to see that on my smartphone video player the 18x18 pattern scales like the Nvidia one with the crosses and sparkles.

I am thinking about making a Windows-to-Go USB drive to try on my Steam Deck to see what it looks like. It is an AMD RDNA2 APU, but customized for them.

[v0lt]

patterns_bw_for_checking_upscaling.zip (https://www.sendspace.com/file/otcxqt)
pattern1_bw18x18_nv12.mkv - old pattern 100 seconds in mkv format
pattern1_bw32x32_nv12.mkv - added white margins around the edges
pattern2_bw32x32_nv12.mkv - new pattern

[huhn]

nivdia 1060 512.95
https://abload.de/img/nvidiad3d9dxvascaler4oce6.png

still the same issues but as expected the position error is gone.
looks like it is placing the pixel wrong by half a sub pixel or something(not that i mean the chroma issue with that).

[JNW]

Why. Interesting and all .All work on shaders. Hell Huhn thinks pixels are round haha

[JNW]

Working on plasma membrane was so thick the walls between next pixel like a wall. So early on was horizontal 1366x768 not even 720p about 300,000 pixels . People jumped on LED cause cheaper and closer mesh but mostly gamers. Full bright like torch mode, factory till burn eyes out know idea of calibration.

[huhn]

Why. Interesting and all .All work on shaders. Hell Huhn thinks pixels are round haha

i'm talking about the digital realm not what a subpixel is and not what is done in practice but sure sub pixel yeah should be infinity small which who would have though is not doable...

[atomcu]

Is it possible playing 4K video on a HD laptop? I want to passthrough to 4K TV. But it is realy buggy. Maybe my laptop is too old. (12 years)

[Klaus1189]

What CPU is built in? Does it have a dedicated GPU?

[huhn]

you need driver that know what HDR is and that's it.

[atomcu]

Intel(R) Core(TM) i3 CPU M 380 @ 2.53GHz
Intel® HD Graphics


What CPU is built in? Does it have a dedicated GPU?

[v0lt]

Intel(R) Core(TM) i3 CPU M 380 @ 2.53GHz
Intel® HD Graphics
Forget 4k on this laptop.

[atomcu]

Forget 4k on this laptop.

As i said my aim is passing through to TV. But it seems my laptop's specs is not enough.

[lvqcl]

my aim is passing through to TV.
AFAIK you need media server program for this. MPC-BE is a classic video player.

[v0lt]

MPC VR SuperRes mod (https://github.com/emoose/VideoRenderer/releases/tag/rtx-1.0)
To watch SD video on a 4k display. :-)
Requires Nvidia RTX graphics card (instructions in the release notes). It will probably work on Intel Arc video cards (see patch (https://github.com/emoose/VideoRenderer/commit/a684e0adff37fbdbe217db464df77beb0e8ca514)).

I can’t say anything more, I don’t have the right hardware.


RTX Video Super Resolution FAQ (https://nvidia.custhelp.com/app/answers/detail/a_id/5448)
You need a GeForce RTX 40 or 30 series graphics card.
"Super Resolution" is enabled in the driver's "Adjust Video Image Settings".
Supported in the latest Chrome and Edge browsers (there is something to compare the work of the MPC VR SuperRes mod).

[Sunspark]

@atomcu Think about other possibilities. Your TV might have an ethernet port already, or USB ports. If it does, it's probably a smart tv and has its own video player. So you could put a flash drive with videos into the usb port, or connect a media server like Serviio to the router and the tv could navigate to it using DLNA and play off the network. In that latter scenario, the laptop could be the server but not the processor of the video.

Likewise, you don't have to scale both chroma and image on the laptop. You could for example, scale just chroma on the laptop at the native resolution of the video, and then let the TV's panel act as an upscaler to 4K. (To do this, video frame would be "Normal Size" instead of "Touch Window From Inside". You may need to also define resolution switching in fullscreen settings in the player to the video's native resolution so the TV knows what to do.)

[ryrynz]

MPC VR SuperRes mod (https://github.com/emoose/VideoRenderer/releases/tag/rtx-1.0)


Looks okay. A bit heavy handed.. if Nvidia adds individual sharpening & denoising strength options and resolution options and continues to improve it adding support for more content it could be a winner. I personally won't be using it at this point and time but it's a solid start.

[Anima123]

RTX Video Super Resolution FAQ (https://nvidia.custhelp.com/app/answers/detail/a_id/5448)
You need a GeForce RTX 40 or 30 series graphics card.
"Super Resolution" is enabled in the driver's "Adjust Video Image Settings".
Supported in the latest Chrome and Edge browsers (there is something to compare the work of the MPC VR SuperRes mod).

Too bad I only have an RTX 2070 card at hand.

[ryrynz]

Too bad I only have an RTX 2070 card at hand.I read that support for 20 series should come later this year.

[atomcu]

Thnx for all your answers.
I have a Philips TV which supports Dolby Atmos. But from the TV it can play just lossy Atmos (DD+). No support for TrueHD Atmos. For this TrueHd Atmos media, I'm passing through to TV via HDMI cable.
Philips TV= 4K , Lossy Atmos (DD+)
Laptop>Soundbar>Philips TV= 1080P , TrueHD Atmos.

@atomcu Think about other possibilities. Your TV might have an ethernet port already, or USB ports. If it does, it's probably a smart tv and has its own video player. So you could put a flash drive with videos into the usb port, or connect a media server like Serviio to the router and the tv could navigate to it using DLNA and play off the network. In that latter scenario, the laptop could be the server but not the processor of the video.

Likewise, you don't have to scale both chroma and image on the laptop. You could for example, scale just chroma on the laptop at the native resolution of the video, and then let the TV's panel act as an upscaler to 4K. (To do this, video frame would be "Normal Size" instead of "Touch Window From Inside". You may need to also define resolution switching in fullscreen settings in the player to the video's native resolution so the TV knows what to do.)

[digi340]

Radio online in HiRes 24-bit/192kHz URL (with metadata): http://ingest-mia.radiomast.io/8a760c25-fb95-4fcb-9c0e-ca0f269a7360
In Microsoft Edge play , but in MPC-BE say " invalid"
Now is offline?

[digi340]

From time to time stream is offline , but on MPC-BE not play never.

[Shark007]

From time to time stream is offline , but on MPC-BE not play never.

If using LAV splitter, playback of the URL works in MPC-BE.

[Sunspark]

@atomcu Read up and see if your soundbar or TV received a firmware update to support TrueHD. Don't flash in firmware blindly to find out, read first to determine whether support exists in any version of it. Though, if you don't have a 7.1 speaker setup, benefits from having it available may not be much different from what you have now.

[Aleksoid1978]

From time to time stream is offline , but on MPC-BE not play never.

Right now he is playing, 30 minutes ago he refused. Very strange site, it is not even clear how to deal with it.
And there is no desire with such a floating problem.

[s0meone_new]

Hungarian translation updated:
https://www89.zippyshare.com/v/6eVSoyQ7/file.html

[Klaus1189]

Updated german (https://drive.google.com/file/d/1NwBzp70DqbsN-NeKUsrorOgUedPyb_nu/view?usp=share_link) translation :)

[v0lt]

@s0meone_new, @Klaus1189
Thanks. Updated.

https://disk.yandex.ru/d/WoH6AFKzmEejEA
My quick implementation of Super Resolution option.
Only for Nvidia. To apply the option, you must restart playback. Not tested by me.

[Klaus1189]

Is there any info if it is used in OSD?

[v0lt]

Is there any info if it is used in OSD?
It will be pointless. The VideoProcessorSetStreamExtension method returns S_OK even if the video card does not support Super Resolution.

I fixed the typo and updated the build - mpcvr_1997_super_resolution_2.zip (https://disk.yandex.ru/d/WoH6AFKzmEejEA).

[Klaus1189]

Too bad, it is not possible. My wooden eyes don't help there :D

[v0lt]

I used patches for Super Resolution and got picture changes for Intel UHD 750.
ModeNone+ScalingDefault vs ModePreproc+ScalingSuperResolution
https://imgsli.com/MTU5ODAz
Perhaps Super Resolution does not work and the picture was changed by Mode Preproc. But I won't check it today.

Checked. ModePreproc and ScalingSuperResolution do not work separately.

MpcVideoRenderer-0.6.6.2001_git2023.03.04-596c319 (https://disk.yandex.ru/d/X0EVMKP4TcmnHQ)

[ryrynz]

They need to calm down on that denoising for real.

[s0meone_new]

Hungarian translation updated:
https://www65.zippyshare.com/v/ygc2O29p/file.html

[huhn]

I used patches for Super Resolution and got picture changes for Intel UHD 750.
ModeNone+ScalingDefault vs ModePreproc+ScalingSuperResolution
https://imgsli.com/MTU5ODAz
Perhaps Super Resolution does not work and the picture was changed by Mode Preproc. But I won't check it today.

MpcVideoRenderer-0.6.6.2001_git2023.03.04-596c319 (https://disk.yandex.ru/d/X0EVMKP4TcmnHQ)

can you play around with the video settings in the intel driver?

[v0lt]

@s0meone_new
Thanks. Updated.

Perhaps Super Resolution does not work and the picture was changed by Mode Preproc.
Checked. ModePreproc and ScalingSuperResolution do not work separately.

Is there any info if it is used in OSD?
Made in MpcVideoRenderer-0.6.6.2002. But whether it will work depends on the hardware and the driver.

[Shark007]

whether it will work depends on the hardware and the driver.
I tested with Intel UHD 630 - working.

https://i.ibb.co/7JQQcbb/stats.jpg (https://ibb.co/7JQQcbb)

[Klaus1189]

RTX 4090 Founder Edition same, but I can not see any meaningful difference, at least with my tested files.

If you have any demo file which can show a measningful difference also my wooden eyes ;) can tell, please post link.