madVR - high quality video renderer (GPU assisted) - Page 2325

tp4tissue · 11th October 2017, 21:55

Hi guyz.

Just borrowed a gpu that can playback 4K remuxes.

Now.. The color conversion..

Is there a way to set this to a reference setting ?

Because what's happening is, I'm playing with the nits setting in hdr to sdr, along with Luminance compression On vs Off, and I'm really not sure how to eyeball this,

I can only compare it to the 1080p remuxes i have of the same movie, but the difference between mastering makes the task kind of a confounded.

How are you guys setting the hdr to sdr , just eyeballing it?

So far with Luminace compression OFF and nits set at 270, seems as close to the 1080p remux as I can get using these settings alone.

Asmodian · 11th October 2017, 22:45

You do not want to match the sdr masters, they are simply too different.

I used to use the madVR defaults with the nits set by eye. Deciding how to calibrate for HDR is still tricky and mastering doesn’t seem to match what standards we have.

leeperry · 12th October 2017, 00:19

OMG, santa-madshi is back at it and he's quite a bit early too

been running tests with the new settings on 1/4 DVD and 1/4 1080p noisy footage @1080p and so far:
-random noise doesn't help at all
-compression artifacts does miracles but 1 isn't quite enough and 2 is too soft so I'd appreciate more granularity here please
-compression artifacts chroma doesn't improve picture clarity when using quad NGU high + both chroma & luma SR + SSIM 2D 100% LL AB25% on my 4:2:2 Sammy TV
-because those new settings are upfront in the "processing" tree we can't quite make profiles for them apparently, hopefully I can leave "compression artifacts" at 1 and call it a day?

tp4tissue · 12th October 2017, 00:36

Quote:

Originally Posted by Asmodian

You do not want to match the sdr masters, they are simply too different.

I used to use the madVR defaults with the nits set by eye. Deciding how to calibrate for HDR is still tricky and mastering doesn’t seem to match what standards we have.

I think compressing the highlights might be the wrong approach..

Because that's like saying, if you had a high pitch noise OUTSIDE of what your speaker can reproduce, you'd Bring that data and playback as the highest frequency the speaker CAN produce..

Everything should just be truncated.. if it's too dark or too bright , just cut it off at 0 or 255..

Asmodian · 12th October 2017, 01:17

Highlights in HDR are not “too bright” they are the change in brightness inside very bright objects. The ability to keep detail in deep shadows and very bright lights is one of the benefits of HDR and I prefer to try and keep as much of that detail as possible.

Flat white for everything bright looks like a bad camera work where the highlights have been blown out. Just watch the SDR master if you don’t want the extra detail in deep shadows or highlights, it will look better.

The frequency analogy doesn’t work, would you want all sounds from higher frequencies played at your max frequency? Clamping a video signal does not have the same visual impact as discarding high frequencies. Volume would be a better analogy and dynamic range compression sounds better with sigmoidal compression, not simply clamping everything above or below a certain volume level.

tp4tissue · 12th October 2017, 01:41

Quote:

Originally Posted by Asmodian

Highlights in HDR are not “too bright” they are the change in brightness inside very bright objects. The ability to keep detail in deep shadows and very bright lights is one of the benefits of HDR and I prefer to try and keep as much of that detail as possible.

Flat white for everything bright looks like a bad camera work where the highlights have been blown out. Just watch the SDR master if you don’t want the extra detail in deep shadows or highlights, it will look better.

The frequency analogy doesn’t work, would you want all sounds from higher frequencies played at your max frequency? Clamping a video signal does not have the same visual impact as discarding high frequencies. Volume would be a better analogy and dynamic range compression sounds better with sigmoidal compression, not simply clamping everything above or below a certain volume level.

Ur right.. I don't know what I'm talking about..

Without compression, it just blows up

ryrynz · 12th October 2017, 01:54

Quote:

Originally Posted by leeperry

hopefully I can leave "compression artifacts" at 1 and call it a day?

Yeah, we're all pretty keen on that, once the levels get tweaked I think we'll be golden, it'll hopefully it'll end up being one of those boxes you tick first.

leeperry · 12th October 2017, 04:34

Quote:

Originally Posted by ryrynz

Yeah, we're all pretty keen on that, once the levels get tweaked I think we'll be golden

I beg to differ as after more thorough testing even 1 appears to smooth out heavy cinema grain, all this said more granularity to go 0.x / 1.x and the ability to mess with it in profiles would be full of win

ryrynz · 12th October 2017, 04:37

That's what I said and something madshi is already working on as I also think 1 is a bit high for general purpose use.
Thoughts on the new AA sharpener?

mzso · 12th October 2017, 10:02

Hi!

I always get madvr's "an error occured in the application" when bookmarking in potplayer. Did anyone experience this? (update: it seems to only happen after frame jump back or forward.)

I also can't send the report because it requires an e-mail client which I don't have (I use gmail in the browser) and it doesn't give me other options, such as saving it into a text file or something.

It seems like some of the reports somehow was copied to the clipboard and caught by the clipboard monitor (It took a long while for me to figure out that I have to press ctrl+c):

https://drive.google.com/open?id=0By...m1abTE4U3R3M3M

I actually get three errors. One when bookmarking, one when stopping playback after, and another right after that. (uploaded one occasion of three reports after numbering them in sequence)

d3rd3vil · 12th October 2017, 10:17

Only madshi-clause or santa madshi can save us regarding Dolby Vision. There wont be any other way to play the stuff via software except MPC. Lets pray together

leeperry · 12th October 2017, 11:21

Quote:

Originally Posted by ryrynz

Thoughts on the new AA sharpener?

I prefer "enhance detail"@0.1 to AA@0.1, still super impressed by RCA@1 on noisy upscales huh

ashlar42 · 12th October 2017, 11:26

Quote:

Originally Posted by leeperry

-because those new settings are upfront in the "processing" tree we can't quite make profiles for them apparently, hopefully I can leave "compression artifacts" at 1 and call it a day?

I haven't tried to create a profile for "processing", as so far I've always used them only for the scaling algorithms but... why do you say "we can't quite make profiles for them"? I don't understand. The option is there.

Edit: did a quick test, profiles work perfectly.

madshi · 12th October 2017, 12:16

Quote:

Originally Posted by Sarlaith

This error is the new error when you can't save because the NVIDIA driver bugged up? doesn't look like the/a GPU/driver error.

After clicking "Ok" it reverts back to the previous resolution

Edit:
If I stop the timer, change the resolution in the NVIDIA control panel I get a very different message. (logical probably because I change resolutions in between)

Edit 2:
It just appears to be still in "test mode" in the NVIDIA API and thus not allowing to save it

Edit 3:
Checked with MadVR 0.9.2.0.3, it is 99% sure a MadVR bug.
On that version I can set pretty much any resolution and it will save it. Going back to 9.2.0.6 gives me the same behavior/error again.
So can you please fix that?

Looks to not exit/still be in "test mode" when you click save, thus giving the error it cannot save the active mode.

Quote:

Originally Posted by Razoola

I'm trying to add a custom resolution in 95.6 using the custon resolution but it is impossible (nvidia).

Does this one fix the problem?

http://madshi.net/madHcCtrl.rar

Quote:

Originally Posted by bitterman

No, it didn't. I just tried it again and it stays on "D3D9 exclusive (new path)" when the freeze image issue happens.

What you see in this video I recorded a couple of days earlier is what I get every time. https://www.youtube.com/watch?v=Ljlp67uHH_0

Quote:

Originally Posted by nevcairiel

Hybrid decoding (like HEVC on Kepler cards) is generally best to stay away from. Its usually slow and often presents problems if the GPU is under high load (ie. with high madVR settings).

^

I suppose as said as it might seem you may have to use software decoding. I'm not completely sure why it's only a problem with madVR and not with VMR9/EVR, maybe it's the higher GPU load, as nevcairiel suggests. Probably not something we can do anything about.

Quote:

Originally Posted by mrmarioman

Yes, that toggle.
I'm not using any beta build.
I just updated to the latest nvidia drivers (387.92) (Also using latest madvr and LAV nightly).
https://www.youtube.com/watch?v=XcVMMBUVTmE
It looks like it's using both, nvidia and windows hdr, at the same time.

Sometimes it works, sometimes colours look washed out.
I've unistalled everything and installed it back again. Not sure what else to do.

Edit- Rolled back to the previous nvidia driver. It's now working. I still need to open and close the player a few times before it displays hdr colours correctly though. It' s a hit and miss for me.

Thanks for the video. I'll have to double check with the latest Nvidia driver. Maybe it got broken again. If it did, I have a good contact to complain to. But it might take a couple of driver revisions before it's fixed again. For now please stick to an older driver.

Not sure, though, why you have to start & close multiple times, that doesn't seem to be the case for the other users. It may help to get back to very old drivers, ideally one directly before the (spring) Creator's Update was officially released. Those driver builds don't support the "HDR and Advanced Color" OS switch yet, and they were the most reliable drivers for HDR for Windows 10.

Alternatively you can do what I've been recommending for many many months now, and use Windows 8.1 instead of Windows 10. But nobody listens to me, obviously.

Quote:

Originally Posted by Werewolfy

I see your point and it was what I have suspected... I've never liked single-frame algo for denoising. If I compare to what I remember then your algo is the best I've seen in this category, definitely not bad but not to my taste.

Could you at least give the choice to select a lower strength?

Yes, lower strength random noise reduction coming in the next build.

Quote:

Originally Posted by Werewolfy

But on the other hand, I'm still amazed of what you have achieved with your RCA

Each time I've tested this type of algo, I've ended up using a very low strength that made little to no difference. Your algo is definitely different, I think I will use it on all my DVDs!

Glad to hear that!

Quote:

Originally Posted by MS-DOS

Is it me, or RRN str1 is actually stronger than str2 ?

The next build will have all RRN strength settings revisited.

Quote:

Originally Posted by el Filou

Here's a comparison on overly compressed SD MPEG2 cable source:
Off
Reduce Compression Artifacts @ 3
Reduce Random Noise @ 1
RCA has to be pushed a bit to get results on really bad source, but when watching the content in motion I prefer RRN at 1 even though it smothes a bit too much. I concur that a lower RRN setting would be best.

In those screenshots it seems to me that RCA removes more artifacts and less texture than RRN, so it seems better in this case at least?

Quote:

Originally Posted by egandt

Looking for where to try "reduce compression artifacts" or "reduce random noise", I have 0.92.6 installed (I have profiles 7 to be exact), but can not find these options on any screen. I hate to delete profiles, as it takes so much time to recreate them from scratch.

Maybe you're running an old version of the MadHcCtrl.exe (settings dialog)? The new settings are supposed to be in the "processing -> artifact removal" settings page.

Quote:

Originally Posted by ryrynz

Always liked Adaptive Sharpen and I just took your word it was nicer than the old version. It's stronger isn't it? I did find myself reducing the level to 0.2/0.3 when running with NGU Sharp. I also like sharpen edges and enhance detail enabled together.. When I was tinkering I realized I now had to decide whether I use NGU Sharp and lower sharpening levels or NGU standard and higher levels. Then whether edge sharpening and enhancing details were worth keeping enabled with an improved Adaptive Sharpen and how they compared to it. I'm tempted to look at all this over the next few days with various content.

Quote:

Originally Posted by ryrynz

Looks like the new AS sharpens more edges, I'd say the new version's sharpness of 0.7-0.8 matches the old 1.0, some areas are much sharper others not so much.

This new AS sharpens the edges of shadowing and lots of inner details which the old one didn't, as it focused more on just the stronger lines which made it very nice for anime.

Comparing new AS 0.7 to old 1.0 I'm finding is that I'm not getting the line sharpness I did (would have to boost to 1.0 to achieve this) but minor shadowing edges are quite visually oversharpened.
I'll look at it on some standard content but you'd probably hate me saying this but..

Look here for a quick comparison of the old vs new AdaptiveSharpen on a grainy image, using much too high sharpening values:

old adaptive sharpen - - | - - new adaptive sharpen

You can see that the old algo produced very artificial catoonish results, while the new algo looks much better. The old algo did sharpen edges more than anything else while the new one algo sharpens texture to some extent, but the end results look much better to my eyes.

Also try with this well known ringing test image:

http://madshi.net/small.png

You'll see that the old algo produced visible ringing while the new doesn't. So overall I think the new algo is a noticeable improvements.

Why not both? Because there are already 6 different sharpening type algorithms in madVR and we don't really need any more. We need less.

Quote:

Originally Posted by cyberscott

Here are a couple of screen shots. These are up-scaled from 480p to 4k, with the "reduce random noise /process chroma channels, too" both ticked and set to strength 1. It looks real good from my viewing distance of around six feet (188.66 cm) from a 55 inch (139.7 cm) OLED 4k TV.

Before and after.

https://photos.app.goo.gl/Q4FzchlC17DeJQ5v2

Thanks! FWIW, in the first image the algo also softens over the white shape on the girl's shoulder, which is not a good thing. Maybe a slightly lower strength of the algo would be better?

Quote:

Originally Posted by Drakko01

I think than the best I can do is chroma upscaling= ngu soft medium and image upscalling NGU Sharp medium without entering in frame drops , or what you saying is do even less on chroma , if that so, which scaler you choose.

The most important thing is image upscaling. So if you can go to NGU Sharp high there, it might be worth using a very cheap chroma algo, e.g. Bicubic75.

Quote:

Originally Posted by Neo-XP

@madshi Do you plan to improve the RRA and RDH algos ?

I am still using aviSynth with FineDehalo because RRA + RDH is not good with NGU (Source / MadVR RRA + RDH / FineDehalo) :

FineDehalo works very well with NGU, but madVR RRA + RDH creates undesired artifacts around the edges.

Haha, you guys with your abbreviations! Took me a bit to figure out what you mean. Anyway, revisiting RRA is on my to do list, but it doesn't have highest priority right now. It might help if you can collect a few short images where the artifacts you're seeing are most obvious, so I have something to look at when I find the time to revisit the algo.

The images you just posted don't really convince me. I see that RRA produces slightly different results than FineDehalo, but I don't really notice any ugly artifacts. Do you have some better images maybe? And can you mark the positions where the artifacts are most obvious?

Quote:

Originally Posted by ashlar42

Tried the new filters on old anime, very low res, crappy compression. They're nothing short of miraculous, as far as my taste goes. Thank you madshi, these are very welcome additions.

madshi · 12th October 2017, 12:21

Quote:

Originally Posted by x7007

Why madvr doesn't switch to the automatic 1080p25p from 4K ?

I set it

1080p23, 1080p24, 1080p25, 2160p23, 2160p24, 2160p25, 2160p30, 2160p60

But it even go to 4K60 Hz instead staying 4K59 Hz which what I use default for everything.

As huhn suggested, this may happen because madVR is not sure if the 25fps movie might need deinterlacing or not. If deinterlacing is needed, 50Hz is the best match, but there's no 50Hz option in your list of display modes. Try adding 1080p50.

Quote:

Originally Posted by huhn

adding 1080p23 and 2160p23 makes 1080p23 obsolete BTW.

Not necessarily. IIRC, if both 1080p23 and 2160p23 are listed then madVR will switch to the mode which matches the native video resolution. I'm not sure why a user would want that, though, because madVR is very likely better at upscaling than the TV.

Quote:

Originally Posted by clsid

Madshi, maybe you could add a heuristic "quality" value? That could be used as a parameter in the preset selection script. Would be very useful particularly for the various processing algos.

I was thinking a value from 1 (very low) to 5 (very high). This value could be determined based the video format (FourCC) and the bits per pixel. The latter value is not known directly yet. Perhaps LAV Filters could export the bitrate (if known). Otherwise a rough estimate could be made using file size and duration.

Hmmmm... Might make sense. FWIW, the profiles already support "MPEG2", "VC-1" and "h264" as boolean checks. So I think I don't need to include this in such a "quality" value. "bitDepth" is available, too. I don't know the bitrate, and I'm not sure that LAV can export this easily? Is the video bitrate information available to the splitter/decoder? I suppose I could make the file size and duration available, or an estimated bitrate based on the file size vs duration, but I'm not sure how reliable that would be because you could have e.g. a low bitrate MPEG2 video with a PCM audio track.

Thoughts?

Quote:

Originally Posted by edcrfv94

Some time when D3D11 windowed change to D3D11 fullscreen windowed or D3D11 fullscreen windowed change to D3D11 windowed, video still playing but no image I still hear sound 3-15 sec then video image back.
Also If playing time change pgs subtitle will get white screen few sec.

CPU: ryzen 1950x
GPU: MSI GTX 1080 Ti GAMING X (Drive: 385.69)
OS: Windows 10 home 1703(15063.632)

MPC-BE v1.5.0 x64 + LAV v0.70.2 x64 + madvr v0.92.6 x64 + XySubFilter 3.1.0.746 BETA3 x64

Does the OSD (Ctrl+J) still update in that situation, or is it stuck (or even invisible)? Which decoder type are you using (software, native dxva, copyback dxva, cuvid, d3d11) in LAV? Have you tried software decoding?

Quote:

Originally Posted by cork_OS

No, I definitely prefer RCA because a) RRN is texture detail killer; b) RCA better removes mosquito noise.

RCA5 - original - RRN5

RCA and especially RRN definitely needs strength levels lower 1.

Ah yes, agreed on texture detail killing!

Quote:

Originally Posted by mbze430

Anyone else having issue with playing HDR videos? Right now with Kodi/DSplayer+MadVR it just crashes for me. Potplayer+MadVR shows a static screen of the video but you can hear the audio playing. And the TV pops in to HDR mode

Windows 10 16299.15 build MadVR .v0.92.4 Nvidia drivers 387.92 (Titan Xp SLI)

Did it ever work? If so, what did you change? Seems latest Nvidia drivers are troublesome for HDR, so maybe try older drivers.

Quote:

Originally Posted by Ver Greeneyes

I retested those LP videos using the new build, and I think it looks best with "deblock=3 deblockChroma=5". So not a huge difference, but I think the chroma can take a bit more punishment before you start to lose any important details (at least on these videos). The deblocking filter definitely helps these videos a lot!

In general I think I'd like it if I could enable chroma deblocking without enabling luma deblocking at all, for videos where the luma channel is fine but the chroma channels need some help. Or do you think that would just be a waste of cycles for something imperceptible? I'd like to think that some deblocking might help the chroma upscaler do a better job.

I'm a bit torn on how to make this available to the user. Of course I could simply split luma and chroma up (for both deblocking and denoising) and make it possible to choose totally different strengths for both and even enable chroma without luma deblocking etc. However, doesn't this make the whole usability worse again? If users quickly want to just deblock the whole video (luma + chroma), they would have to always update both strengths to match. That's currently easier to do. So I'm not sure what I should do.

Thoughts?

Quote:

Originally Posted by Grimsdyke

@ Madshi. MPC-BE and 92_6.
When clicking stop during playback MPC-BE restarts playback from beginning !! When using EVR-CP playback stops as it should. I had the buggy v92_5 installed. Maybe that's some sideeffect ?

Is this a new problem with v0.92.5/6? Or was it always this way when using madVR? If it's a new problem can you please test with exact madVR version introduced this problem? You can download old builds for testing here:

http://www.videohelp.com/tools/madVR...sions#download

Quote:

Originally Posted by videobruce

46,500 posts???? Sheesh...............

I did a search for "Light Alloy" in this thread and only found one hit. They (LA) claim it works with this software, but on madvr's home page, Light Alloy Player is not listed. There are at least two issues with it, but it may be LA's software. Anyone else use LA here?

Wasn't aware of the "Light Alloy" player. If the player devs want to discuss problems with me, or have their player listed, they're welcome to contact me.

Quote:

Originally Posted by tp4tissue

Just borrowed a gpu that can playback 4K remuxes.

Now.. The color conversion..

Is there a way to set this to a reference setting ?

No, because HDR videos ask for specific pixels to be rendered to a specific luminance value, and that is only possible for madVR if you tell it the exact peak luminance value of your display. Every display is different.

Quote:

Originally Posted by tp4tissue

Because what's happening is, I'm playing with the nits setting in hdr to sdr, along with Luminance compression On vs Off, and I'm really not sure how to eyeball this,

I can only compare it to the 1080p remuxes i have of the same movie, but the difference between mastering makes the task kind of a confounded.

How are you guys setting the hdr to sdr , just eyeballing it?

So far with Luminace compression OFF and nits set at 270, seems as close to the 1080p remux as I can get using these settings alone.

I've usually found that a good HDR movie looks better than the 1080p studio SDR encode, but it's not always the same nits value that gives the HDR movie an overall "comparable" brightness/contrast. It differs from movie to movie. My advice would be to try a couple of HDR movies and trust your eyes in picking a nits value that looks good to your eyes. It might also depends on the ambient light levels. A lower nits value makes images more contrasty/punchy, but if there's a lot of ambient light, shadow detail might get lost.

Quote:

Originally Posted by leeperry

been running tests with the new settings on 1/4 DVD and 1/4 1080p noisy footage @1080p and so far:
-random noise doesn't help at all
-compression artifacts does miracles but 1 isn't quite enough and 2 is too soft so I'd appreciate more granularity here please
-compression artifacts chroma doesn't improve picture clarity when using quad NGU high + both chroma & luma SR + SSIM 2D 100% LL AB25% on my 4:2:2 Sammy TV
-because those new settings are upfront in the "processing" tree we can't quite make profiles for them apparently, hopefully I can leave "compression artifacts" at 1 and call it a day?

Why can't you make profiles? Works just fine for me.

Quote:

Originally Posted by mzso

I always get madvr's "an error occured in the application" when bookmarking in potplayer.

The PotPlayer dev just fixed a probably related bug today, so this problem might be fixed by the next PotPlayer version.

Quote:

Originally Posted by d3rd3vil

Question though: is there a "faster" way to at least ehm show that Dolby Vision is running in the information overview?
Cause when I play a Dolby Vision TS file nothing indicates that this is even Dolby Vision. Then when I convert it via MKVToolNix Dolby Vision doesnt work anymore but the MPC info is more or less the same.

Is there a way to show if DV is active(even though not working via MPC)?

What do you mean with "Dolby Vision doesn't work, anymore"? Does it ever work? How?

Quote:

Originally Posted by nevcairiel

playing Dolby Vision properly with its dynamic metadata is likely entirely impossible for the forseeable future.

Actually, it might be possible. Dolby Vision consists of an additional video stream/layer, and additional metadata. What madVR could in theory do is:

1) madVR could make use of the additional video layer + additional metadata to move the video up from 10bit to 12bit+. This should produce a benefit for every display, regardless of whether it supports Dolby Vision or not, or whether it even supports HDR or not.

2) madVR could behave like any official Dolby Vision source device and send the video to a Dolby Vision compatible display, including all the dynamic metadata etc. This is theoretically possible, with any HDMI 1.4+ GPU, as long as it supports a perfect RGB 8bit "passthrough", where all bits stay untouched by the GPU driver (and Windows). The reason this would work (if perfect passthrough is available) is that the dynamic Dolby metadata is actually encoded into the image pixels, as a sort of watermark. So it doesn't require any specific HDMI version, and it even works if there's an AVR in the middle which doesn't support Dolby Vision (as long as the AVR keeps the watermarked pixels untouched).

The whole Dolby Vision processing is very complicated, but I would be willing to at least give 1) a try, maybe even 2), but it would require me to get access to the decoded Dolby Vision enhancement layer, plus to all the added Dolby metadata information contained in the base + enhancement layers.

So basically for any of this to work, first the splitter + decoder would have to be enhanced to deliver the Dolby Vision information to madVR. Then I could look into how to process it all.

leeperry · 12th October 2017, 12:51

Quote: