madVR - high quality video renderer (GPU assisted)

[huhn]

with an unstable device it's unlikely to get accurate results so the 3% one could results in more accurate results if it is in the 400 nits range if the 3% is close to common knee points it unlikely to be every be used for real content in this case i would use a lower number.

[flossy_cake]

Originally Posted by VHT

Hello, Question about HDR settings and display peak luminance (nits) value. Do I put in eg. peak 3% window value or 100% fullscreen values?

If your display has ABL then I presume you'd want to pick a window size that represents the APL of typical video content. From memory most movies are around 15% APL so a 15% window might suit.

[VHT]

Originally Posted by flossy_cake

If your display has ABL then I presume you'd want to pick a window size that represents the APL of typical video content. From memory most movies are around 15% APL so a 15% window might suit.

Thanks! Will try with my OLED display. How about with projectors?

[flossy_cake]

Originally Posted by VHT

How about with projectors?

I have zero experience with projectors, but if it uses a technology which has an ABL that reduces peak brightness as APL increases, then I presume the same general logic would apply, unless there is some other behaviour specific to projectors that I'm not aware of.

And I've just realised that figure of 15% I gave you was for SDR movies -- the APL of HDR movies can vary with each movie having a different MaxFALL (Maximum Frame Average Light Level). I'm not sure how to take that into account... perhaps we could take an average MaxFALL across a random selection of movies?

[flossy_cake]

Actually MaxFALL doesn't seem very useful either — imagine if there was a fade to white scene that pushes the MaxFALL to a high value.

[nevcairiel]

Originally Posted by flossy_cake

Actually MaxFALL doesn't seem very useful either — imagine if there was a fade to white scene that pushes the MaxFALL to a high value.

A fade to white would likely fade to diffuse (SDR) white, not max brightness HDR white, though. You don't want to blind people, assuming a screen could actually do that.

[flossy_cake]

Originally Posted by nevcairiel

A fade to white would likely fade to diffuse (SDR) white, not max brightness HDR white, though. You don't want to blind people, assuming a screen could actually do that.

Yeah that's true, and as quoted here diffuse white is 200 nits -- higher than I would expect the APL would be across the entire movie.

[Alexkral]

Originally Posted by nevcairiel

A fade to white would likely fade to diffuse (SDR) white, not max brightness HDR white, though. You don't want to blind people, assuming a screen could actually do that.

I've seen some of those though, e.g. in the film Mother! at 1:52:10 there's a fade to 533 nits white. The rest of the film is extremely dark, so I think the author's intention is precisely to blind the audience.

[flossy_cake]

Random question: how come when I set [deint=off] with 29.97i 3:2 cadence video I'm not seeing mice teeth?

[huhn]

because it is 2:2 or 6:4 anime or because you have software deint on.

[flossy_cake]

Originally Posted by huhn

because it is 2:2 or 6:4 anime or because you have software deint on.

I do not have software deint on & LAV Video Decoder's deint settings are both set to "auto".

Here are 2 clips with the following properties:

• MadVR & MediaInfo both report 29.97fps
• MadVR reports "cadence 3:2" when MadVR's film mode deint is activated
• My TV is refreshing @ 59.94hz and I visually observe the classic 3:2 "judder"
• I can step through each frame with ctrl+arrows and count 24 unique images before the timecode ticks over 1 second

https://drive.google.com/file/d/1Gv9...ew?usp=sharing
https://drive.google.com/file/d/1Pgs...ew?usp=sharing

[huhn]

soft telecine or what ever this has no interlaced frame in it.

[flossy_cake]

Originally Posted by huhn

soft telecine or what ever this has no interlaced frame in it.

Yes that's what I thought too!

But the first clip (Cowboy Bebop) immediately follows the intro sequence which contains mixed cadences of 1:1 and 3:2. So it really is a 29.97i stream not soft telecine.

Is it possible to be using variable framerates? I don't mean mixed cadences, I mean the MPEG stream may be flagging "these frames are 29.97i, but these frames are 23.976p" and then leaving it up to the DVD player to generate the interlaced output?

edit: but that still doesn't explain why MadVR & MediaInfo are both reporting 29.97i, AND, if I play a typical 23.976p file with [deint=film] MadVR reports "cadence 2:2" instead of "cadence 3:2" with those test clips. Maybe the 29.97i flag is just some leftover incorrect metadata based on the beginning of the stream before it dynamically changed to 23.976p mid stream?

edit: and also, playing them on my PC monitor refresh at 119.88hz shows 3:2 judder as well, which it wouldn't if it was soft telecine 23.976p since that goes evenly into 119.88 just like when playing a 1080p23.976 file there is no 3:2 judder. But I still think you are right and it is probably soft-telecine, because I can't think of any other explanation why there aren't mice teeth. There must be some explanation for how soft-telecine variable framerate stuff is handled by the decoder that's making it get presented in this fashion. I just have no clue what it is.

[Siso]

Which is the good way to prevent black clipping on an IPS display? With custom levels in madvr 0,5-255 in my case, or to adjust the black output offset in the 3dlut to 90% or something?

[flossy_cake]

Originally Posted by Siso

Which is the efficient way to prevent black clipping on an IPS display? With custom levels in madvr 0,5-255 in my case, or to adjust the black output offset in the 3dlut to 90% or something?

What do you mean by "efficient" - low GPU consumption? In that case just use a GPU meter while factoring in the current clock speed. The solution you are currently using should use a trivial amount of GPU processing.

From memory, 3D LUT has the advantage of adjusting a much wider range of colours but doesn't have as much precise adjustability within a narrow range of colours, otherwise the 3D LUT has to be very large. This could be bad if you want to fine-tune a small range of colours near-black.

But maybe I'm wrong and 3D LUTs are big now and can do these kinds of precise adjustments. From memory 3D LUTs will get upscaled at runtime where it calculates intermediate tones between each row in the table. But that won't necessarily get you the precision you want - depends how it's interpolating between rows in the table.

At the time I was studying all this, only a 10-bit or greater 1D LUT with 256 rows gave me the necessary precision to finely tune shadow detail to my liking. If you had the same precision 3D LUT the table would be massive, but again maybe that's not a problem anymore these days maybe someone can correct me on that.

The other problem I have with 3D LUT is that it can't be applied to everything, so you'll only get the correction when in an application that supports it. If you want a global customised boost to shadow detail that applies to all apps then you can do that with my app Calibration Tools which uses 1D LUT (video card gamma table). There was some discussion about that on the previous 2 pages. Try not to overboost the shadows though as 1D LUTs cant adjust brightness linearity without also affecting hue and saturation. Basically the more you boost the shadows in a 1D LUT, the more blue and sickly skin tones will start to look in the low end (0-10% range).

[flossy_cake]

Originally Posted by flossy_cake

Try not to overboost the shadows though as 1D LUTs cant adjust brightness linearity without also affecting hue and saturation. Basically the more you boost the shadows in a 1D LUT, the more blue and sickly skin tones will start to look in the low end (0-10% range).

Just to clarify this in case it sounds like I'm talking rubbish. Suppose you have a skin tone RGB (0.15, 0.10, 0.05) which has an average brightness value of around 10%, such as might be found in a "low light" scene. Because 1D LUT operates on subpixels not entire pixels, if we use 1D LUT to boost R+G+B at the 5% level, for this particular colour only the blue subpixel (0.05) gets brighter.

So in a 1D LUT we tried to brighten all channels around 5% to increase shadow detail, but a side effect was that skin tones at 10% got brighter, bluish and desaturated. 3D LUT doesn't have this problem (but has that other problem of not enough granularity or precision to make precisely shaped curves in very small ranges of colours...unless the table is very big).

The best way to adjust shadow detail imo is to use a pixel shader, take the lightness value (such as L* or Y) of the pixel and apply a curve to it, then convert it back to RGB and it should get brighter without affecting its hue and saturation very much.

Here is an article showing how putting curves in a 1D LUT screws up the saturation and hues: http://www.colormancer.com/whitepape...omparison.html

You can replicate it in Caltools by opening the first original image from above and selecting S-Curve "high" preset in the editor and toggling it on and off with the "preview" checkbox. I've also put some test images in /images/miscellanous folder of the app (skin ramp & skin check png's) where you can see how much you're mucking up the colours.

[Drew Neilson]

Will MadVR be rewritten as a Media Foundation Transform so that Media Foundation players can use it?

My experience with MadVR is limited by the fact that my only PC--a 2017 Microsoft Surface Pro--doesn't really have enough horsepower to support MadVR, according to my limited experimentation with it in JRiver Media Center. However, I experimented with it enough to see its quality. I'm currently researching PC parts and I might build a gaming HTPC in the near future. I'd like to be able to use clients for Twitch and YouTube and have those clients use MadVR processing for video playback. I am assuming, perhaps incorrectly, that modern video playback applications use Media Foundation and not DirectShow. That's why I'm asking if MadVR will be rewritten as a Media Foundation Transform, so that it will work with Media Foundation.

It would be cool to have the option to make a web browser, such as Microsoft Edge, apply MadVR processing to videos on the web, such as at YouTube or at Twitch's sites. It'd also be nice to have the option to make third-party YouTube and Twitch apps from the Microsoft Store use MadVR.

[SirMaster]

Originally Posted by Drew Neilson

Will MadVR be rewritten as a Media Foundation Transform so that Media Foundation players can use it?

My experience with MadVR is limited by the fact that my only PC--a 2017 Microsoft Surface Pro--doesn't really have enough horsepower to support MadVR, according to my limited experimentation with it in JRiver Media Center. However, I experimented with it enough to see its quality. I'm currently researching PC parts and I might build a gaming HTPC in the near future. I'd like to be able to use clients for Twitch and YouTube and have those clients use MadVR processing for video playback. I am assuming, perhaps incorrectly, that modern video playback applications use Media Foundation and not DirectShow. That's why I'm asking if MadVR will be rewritten as a Media Foundation Transform, so that it will work with Media Foundation.

It would be cool to have the option to make a web browser, such as Microsoft Edge, apply MadVR processing to videos on the web, such as at YouTube or at Twitch's sites. It'd also be nice to have the option to make third-party YouTube and Twitch apps from the Microsoft Store use MadVR.

You can definitely play YouTube and twitch via MPC-BE/HC and thus with madVR processing.

This is how I play YouTube in my home theater. madVR tone-maps YouTube HDR and upscales everything to 4K for me.

[Drew Neilson]

Originally Posted by Drew Neilson

I am assuming, perhaps incorrectly, that modern video playback applications use Media Foundation and not DirectShow. That's why I'm asking if MadVR will be rewritten as a Media Foundation Transform, so that it will work with Media Foundation.

I meant to ask the following:
I'm asking if MadVR will be rewritten as a Media Foundation Transform, so that Media Foundation will use it like how DirectShow currently uses it, and its processing will be applied to video played by *any* Media Foundation application--web browsers, video players from the Microsoft Store, and video players from elsewhere.

Originally Posted by SirMaster

You can definitely play YouTube and twitch via MPC-BE/HC and thus with madVR processing.

This is how I play YouTube in my home theater. madVR tone-maps YouTube HDR and upscales everything to 4K for me.

Thanks. I will check out MPC-BE/HC.

[nevcairiel]

There are practically no standalone players using Media Foundation that would support the use of madVR. And even if there were, the chances of such an addition happening to madVR are basically zero.