madVR - General discussion thread

[NoTechi]

Well there are some new as well (https://en.wikipedia.org/wiki/List_of_3D_films_(2005_onwards)) ;)

NoTechi

[Alexkral]

Hugo and Guardians of the Galaxy Vol. 2 are the best for me, and the last one is from 2017. But it is true that quality and depth are often just decent.

[BetA13]

How can i explain to people what MadVR does?
i know its a stupid question, but i need some help..
I have a friend and was explaining to him why i use madvr, and he begun to discuss that u cant make the video better, there is nothing more then the source file, you cant add details etc etc..

WHile i do agree with him to a certain point, i cant seem to make him understand what madvr does and how it CAN enhance the picture, etc.. best comparsion would be the latest GoT epsiode. But he is a hard head.

So, anyone else had those discussions about quality and would lend me a hand here :)

I really dont know how to explain it what and why madvr can does what it does.

Thanks ...

[jbnpaul]

How can i explain to people what MadVR does?

i know its a stupid question, but i need some help..

I have a friend and was explaining to him why i use madvr, and he begun to discuss that u cant make the video better, there is nothing more then the source file, you cant add details etc etc..



WHile i do agree with him to a certain point, i cant seem to make him understand what madvr does and how it CAN enhance the picture, etc.. best comparsion would be the latest GoT epsiode. But he is a hard head.



So, anyone else had those discussions about quality and would lend me a hand here :)



I really dont know how to explain it what and why madvr can does what it does.



Thanks ...



Are there any comparison images available that would show what madvr can do with the long night?



Sent from my iPhone using Tapatalk

[Alexkral]

WHile i do agree with him to a certain point, i cant seem to make him understand what madvr does and how it CAN enhance the picture, etc..

I don't think that in general the objective is to enhance the picture, but to minimize the quality loss that implies some processing that is necessary, or to hide the poor quality of the picture itself. Sometimes the best way to do it is subjective, sometimes your system is not able to handle the best way, and there are infinite scenarios, so madVR offers you a lot of options, some of them really good, so you can do it which way you deem most convenient.

The exception to this would be tone and gamut mapping, which could be considered a more artistic question, and despite having more than 30 years of history has not yet found an ideal solution.

Anyway, madVR is not the only software that does this, e.g. mpv also has a lot of options for the same things, some of them really good too.

[Warner306]

I'd agree that the purpose of any video player is to minimize quality loss. But a good video player also restores quality lost through content creation and compression. Even content on a Blu-ray disc, encoded in Y'CbCr 4:2:0, is lower quality than a display is capable of presenting. The dominance of streaming video means video quality loss will be even more of an issue going forward, and the average streaming media player puts very little effort into improving it.

Video processors have to carry out a number of basic video processing calculations just to get the image to the display. Every media player will handle those calculations differently, which is where the variance in image quality comes in. On PC alone, you are likely to see slight differences in color at the display amongst different video players because everything is converted to RGB and processed in RGB, which further complicates the job of the video player.

[mclingo]

not sure I can quite agree exactly with that Warner, but I suspect its just your wording I disagree with, you cant actually add quality to an image, you can essentially only fool the eye to to mask issues, like dithering, sharpening or banding removal, perhaps in a decade or so AI might be able to add quality to an image but right now its about minimising loss through the transformations / conversions that occur from original media to player to output.

I agree more with Alexkral.

[nevcairiel]

Loss would mean that there is information present that might be lost if you process it wrong. But upscaling, for example, doesn't really do that. It creates new information - it adds quality. Of course you can also do scaling entirely wrong and destroy quality, but lets assume that everyone manages to at least build a working Bilenear scaler that preserves at least all information.

Preventing loss would be something like doing color conversion correctly, dithering properly when needed, things like that. Such conversions have a similar amount of information afterwards, just in a different format (ie. RGB instead of YUV, etc).
Adding quality would be things like advanced upscaling. You have more information after then you had before - higher resolution, no chroma subsampling.

[BetA13]

thank you for those answers :)
nevcairiel has a goooooooooood point.. Wich i never thought of like that. Yes, there are MORE Information afterwards (upscaling, supersample, etc), wich would be a good point, i think i do understand a bit better now..

BUT PLEASE; keep it coming, i really like to hear some more :)

Cheers

[Alexkral]

I don't see it that way. Upscaling always means a loss of quality, not because you are losing information, but because you are forced to create new information. It's not that you have the option to do it. You are not adding quality, but minimizing that loss of quality by creating new information in a way that the loss of quality is less perceptible.

[mclingo]

Loss would mean that there is information present that might be lost if you process it wrong. But upscaling, for example, doesn't really do that. It creates new information - it adds quality. Of course you can also do scaling entirely wrong and destroy quality, but lets assume that everyone manages to at least build a working Bilenear scaler that preserves at least all information.

Preventing loss would be something like doing color conversion correctly, dithering properly when needed, things like that. Such conversions have a similar amount of information afterwards, just in a different format (ie. RGB instead of YUV, etc).
Adding quality would be things like advanced upscaling. You have more information after then you had before - higher resolution, no chroma subsampling.

it adds "data" but it cannot increase the "quality" of the image only the perceived quality of the image as you cant add information to an image that wasnt there in the first place.

I guess it depends on how you define "quality" for me, a native 4k image is better "quality" than a native 1080p image as there is more real information captured, the quality of an image on one level is purely about the information captured and you cant add that later, you can only approximate that information, therefore you are not increasing the quality of the image, only the resolution and perceived quality. However if you capture a blurry 4k image a non blurry 1080p image could be defined as a better quality image, so there is room for interpretation.

I wouldl still state that MADVR is not increasing the quality of the image, it is adding new data but no new information about the subject captured, therefore it cannot be increasing the quality of the image.

[Asmodian]

A 1080p source upscaled to 4K with NGU very high looks better than the original 1080p source viewed at the same size. Just writing this off as irrelevant seems counter productive. madVR's chroma upsampling is way better than simply using bilinear upscaling before conversion to RGB. You can say Nearest Neighbor is the best option because it does not change the source pixels but since the resulting image very obviously looks worse than NGU AA very high I will stick to using NGU. :p

The resulting image from madVR looks better, what it this effect if not improved quality? Sometimes you have to drop the theory and look at the results with common sense.

Remember the source is not 100% quality, the original scene viewed with our eyes is 100% quality, and we want to use the source to try to get as close to the look of the latter as possible. The original source pixels are not special things that must be preserved for some reason, they are simply the data set we have available to try to approximate the original scene from.

[Warner306]

thank you for those answers :)
nevcairiel has a goooooooooood point.. Wich i never thought of like that. Yes, there are MORE Information afterwards (upscaling, supersample, etc), wich would be a good point, i think i do understand a bit better now..

BUT PLEASE; keep it coming, i really like to hear some more :)

Cheers

This article provides a test that is both a good test of preserving the quality of the source through color space conversions and adding quality in the compressed chroma channel:

http://spearsandmunsil.com/portfolio-item/choosing-a-color-space/

And this post shows how some devices can go wrong when doing the simplest of video processing, such as outputting an 8-bit RGB test pattern with the same input and output RGB values, with no other work to do:

https://forums.geforce.com/default/topic/895245/shield-tv/picture-quality-and-color-space-evaluation/post/4818394/#4818394

This is how the Nvidia Shield used to score on the video benchmark test:

https://forums.geforce.com/default/topic/895245/shield-tv/picture-quality-and-color-space-evaluation/1/

I think of the use of subtle noise reduction, artifact removal and chroma upscaling as restoring quality lost through the camera lens or from when the source master is converted for distribution. It looked better on the mastering monitor and some attempt has been made to restore any lost detail (mostly due to subsampling and compression).

[Alexkral]

A 1080p source upscaled to 4K with NGU very high looks better than the original 1080p source viewed at the same size.
That does not make any sense to me. I could test it a hundred times with NGU very high, FSRCNNX, waifu ... and I'm sure the original 1080p source would always look better to me. Up to some point, even downscaling it would look better, it would lose information but it would be sharper.

And the original source pixels are the only reference we have of the original scene, so trying to preserve them is the best way to aproximate to it. From there, any creative option is always going to be debatable.

[nevcairiel]

That does not make any sense to me. I could test it a hundred times with NGU very high, FSRCNNX, waifu ... and I'm sure the original 1080p source would always look better to me. Up to some point, even downscaling it would look better, it would lose information but it would be sharper.


Its practically impossible to make a fair comparison, either way.
A display has a native resolution, so you would never have one single display that can show the same image, in the same size, in both 1080p or 4K without some upscaling.

However, tests have shown that a properly upscaled image on a 4K TV can look better in controlled settings then a native 1080p image on 1080p display. Upscaling can increase the perceived details, especially when the source is already pretty high quality. This might be "fake" quality, but what users prefer is what wins either way.

Ultimately this is irrelevant anyway, since you have one display with one resolution, and all content needs to be shown on it. Running a display at anything but native resolution would just be fooling yourself, since the display will do the upscaling then.


And the original source pixels are the only reference we have of the original scene, so trying to preserve them is the best way to aproximate to it.

Thats called "Nearest Neighbor", and its super ugly for any normal content, because you get a lot of aliasing. The image looks perceivably low resolution. (Its even worse if you need to downscale, entire information is just lost)

So you start upsampling the image in some way. Any artificial limitations you impose on the upscaling process will always make it worse. So something like "but it must preserve the original pixel" will result in a worse result either way. If you upsample to a ratio not a power of 2, this would not work anyway, since the original pixel don't fall onto positions where there are pixel in the output resolution.

Ultimately anyone can do what they prefer, of course. And there is plenty options for everyone.

[Asmodian]

Up to some point, even downscaling it would look better, it would lose information but it would be sharper.

It is only sharper if you view it at a smaller size so details occupy less of your vision. It is less sharp after downscaling if you view it at the same angular size in your field of view. A thumbnail of an out of focus photo can look fine, but that does not mean that downscaling would help the original image look any sharper.

[Alexkral]

Yes, that's right, I was incorrectly assuming a different angular size. Anyway, I was not referring to preserve the exact original pixels, but to preserve the way they should look at a different image size. I don't know much about upscaling, in tone mapping for example this is done sometimes by matching the perceived brightness or contrast. But this is probably very close to what you were saying.:)

[juffi]

MadVR seems to switch display mode every time I start and stop playback with MPC-BE. I have disabled "switch to matching display mode" and "restore original display mode".

How to prevent that happening? I have an Optoma UHD51 which has exceptionally long and annoying display mode change delay.

[mclingo]

anyone been watching Doom Patrol, banding and streaming artifacts horrific, even with MADVR at HIGH.

[chros]

Forget about Intel :) go for AMD this time

If anybody is happened to be bored before Monday's announcements at Computex: Intel is dead (https://www.youtube.com/watch?v=zdCi_r_lm7o) :)

... Intel ... what they did to the PC market in the last decade due to lack of competition.

I just came across a 1.5 years old similar video about nVidia (https://www.youtube.com/watch?v=Jd1bp9eSfwo) as well. Really interesting, especially the conclusions at the end.

Enjoy!

[huhn]

someone really likes click bait videos here...
i mean intel can't produce enough CPU to meet demand that'S a clear sign of a dead company!

[chros]

someone really likes click bait videos here...
i mean intel can't produce enough CPU to meet demand that'S a clear sign of a dead company!
:) Yeah, maybe the word "dead" is a bit strong :)

[huhn]

yeah maybe every single word in the title is miss leading,
so why should anyone even trust/hear out a word of a person that is clearly lying with the title?

[mclingo]

something ive wondered about before but never really got an answer to. In windows audio control panel my HTPC keeps defaulting to STEREO 16bit 48khz instead of 7.1 24bit 192 khz which I want it set to. When playing a movie using KODI or something else and bitstreaming what quality does it bitsteam at, your windows setting or does it automatically output the cards max setting, does whats set in the control panel acually matter at all, could I just leave it set at stereo 16/48?

I setup a batch file to run at logon which switches it back to 7.1 but it doesnt always work and I often forget about it, its quite annoying and I havent yet found a way around this, anyone else?

[mclingo]

or does it bitstream at the movies encoded rate regardless of what windows it set to?

[SamuriHL]

Bitstreaming bypasses the windows settings completely. It's sent directly to your receiver so whatever it's encoded as and whatever your receiver does with that have nothing to do with windows settings at all.

[mclingo]

cool thanks, i guess my only concern is youtubewhich wont bitstream, but thats still not surround anyway is it, is youtube still just outputting in stereo on a PC desktop?

[huhn]

youtube uses multi channel opus so you are unlikely to bitstream that.

[mclingo]

you cant get any surround on youtube with desktop PC can you?, I heard firefox supports it but I dont use it, chrome doesnt.

[huhn]

no clue what you are talking about. i can clearly play multi channel on my PC.

[mclingo]

you can play multi channel videos on youtube and you get full surround sound using google chrome?, when I play stuff it simply sums the rears to the front, sure your amp is just upmixing to virtual surround, pretty sure youtube is only stereo using chrome on a pc.

[mclingo]

https://www.youtube.com/watch?v=bzv63Q2beg0

[huhn]

i tested it using mpc-hc it's working fine.

[chros]

https://www.youtube.com/watch?v=bzv63Q2beg0
That video only has 2ch sound at 1080p, using MPC-BE:
Video: H.264/AVC 1920x1080 25fps [VideoHandler - MPEG4 Video (H264) (AVC (H.264), High Profile, Level 4.0, 1920x1080, 25.000 fps)]
Audio: AAC 44100Hz 2.0 chn [SoundHandler - MPEG-4 Audio (AAC, 44.1 kHz, 2.0 chn, 16 bit)]

Can you link one that indeed has 6ch?

[mclingo]

that video wasnt a test video, i posted it as it explains in detail why youtube doesnt have surround sound on PC/laptops in a browser.

perhaps i'm not making myself clear here, I was wondering why so many people where saying they are getting full surround sound from youtube in a browser when it does not appear to be technically possible, i've only managed to get 2.0 with all other channels summing to front, i cant even get phased rear as part of the dolby quad matrix (see video)

[nussman]

If you want an expample for „not technically possible“: 4K 444 RGB 60hz / 10bit

[mclingo]

thats a bit random, whats your point?

[nevcairiel]

The point is that it is very well technically possible to do multi-channel on a PC, and that YouTube does provide multi-channel audio. If a particular browser (or any browser) doesn't play that, its the browsers problem. Many people consume YouTube through media frontends on a HTPC, which doesn't involve a browser for playback.

This video for example serves 5.1 AAC: (and yes, this is also available with "real" content, not just this test video)
https://www.youtube.com/watch?v=iTHtXgnhqfM

If you play that with MPC-HC, you get 5.1 audio that you cannot bitstream.

[mclingo]

again, i'm simply stating its not technically possible to do 5.1/7.1 or any type of surround in desktop Chrome, I am aware it can be gained in using other methods, I cant even get my rear channels to process anything at all in chrome, this is ALL i'm seaking to get clarification on.

How youtube works is covered i the video I posted, essentially what it states is that when something is uploaded to youtube the original video is held "somewhere" but when its streamed its re-encoded in dolby stereo in most PC browsers. However he states that some platforms could be getting access to the original 5.1/7.1 source but people could be thinking they are getting 5.1/7.1 when they are really only getting phased dolby 2.0 back channels or their audio equipment is upmixing to 5.1/7.1.


I'll ask again, given that there are a lot of people out there stating they are getting surround sound through CHROME BROWSER on their desktop PCs, how are they achieving this if I cant, and I cant get anything out of my rears at all, or are they just mistaken?

[nevcairiel]

You are the only one talking about Chrome. Noone else even mentioned a particular kind of browser. The original point was that non-bitstreamable multi-channel audio exists and some people play that. Not that it doesn't work on your system with your particular software.

[richardpl]

There are certainly multi-channel ambisonics on youtube.
They usually play as stereo via browser.

[mclingo]

i give up.

[Asmodian]

You already know the answer to your question, both the first and second one. What are you giving up?

No, no one else gets multichannel out of Chrome. Youtube is a different story, they are not synonymous.

[chros]

This video for example serves 5.1 AAC: (and yes, this is also available with "real" content, not just this test video)
https://www.youtube.com/watch?v=iTHtXgnhqfM

If you play that with MPC-HC, you get 5.1 audio that you cannot bitstream.
Thanks, it works in MPC-HC, but not in MPB-BE (only 2ch audio).

[ashlar42]

Thanks, it works in MPC-HC, but not in MPB-BE (only 2ch audio).Just loaded it in MPC-BE, disabling the built-in YouTube web page parser, selected the multichannel AAC track and it played it without a problem.

[leeperry]

kinda mVR-related but anyway it's the chitchat topic so here goes, this PDF makes very clear that REC.709 calls for D65 white point: https://app.spectracal.com/Documents/White%20Papers/Calibrating%20Color-Critical%20Monitors.pdf
6500 Kelvin being the neutral white/gray color specified by the BT.709 industry standard for HDTV images
yet, I often see video's being shot at 5200K: https://www.youtube.com/watch?v=X-aaDs5heQ4&feature=youtu.be&t=46

what gives :confused:

final result is plenty red, though: https://youtu.be/2S24-y0Ij3Y?t=31

I guess they prolly shoot at 48fps in order to control motion-blur when downgrading to 24

[nevcairiel]

Its called post-processing or mastering. That process can change anything in the video between the camera and your display. :p

[leeperry]

Breaking news lol, I just wonder why they don't record in D65 in the first place huh? Even with high bit-depth 3DLUT's, is 5200K>6500K conversion lossless?

I've seen it on EU / US & korean videos, must be a reason to it. I guess they use digital cameras with REC.709 gamut so why "daylight" 5200K :confused:

Anyhow, indeed motion blur is mostly inexistant in that 24fps final video so I guess recording in 48fps does indeed wonders if post-processing to 24 is right :)

[huhn]

depending on the point of view you can do this "lossless" ignoring rounding/dithering bit deep not the issue here. changing the white point effects the CR if you have to match another white point you will lose CR because you can't reach the whites white anymore. if a video is more red or not has nothing to do with this it just looks like this because they wanted it to look like this.

[leeperry]

indeed, theoretically they are crippling native CR but of course they will increase it artificially in every scene afterwards during mastering.

that screenshot says "SDI" & "4K" and REC.2020 also calls for D65: https://en.wikipedia.org/wiki/Rec._2020

they go 48fps so they know what they're doing and I guess there's a good reason to 5200K, usually koreans sit on top of technology so they woulda gone D65 if that was a better choice....oh well :D