madVR - high quality video renderer (GPU assisted)

[varekai]

how to figure out if the video card GPU is at the limit?
What video card do you have?
For only monitoring you can use GPU-Z
https://www.techpowerup.com/gpuz/
For tweaking such as changing memory clock and adjusting fans and such there are free software for that, both for AMD or NVIDIA.
I use EVGA Precision which is for NVIDIA GPU's.

[el Filou]

thought the processing happened on the GPU [...]
If I use NATIVE instead of COPYBACK the CPU use is halved suggesting some of MADVR's work is happening on the CPU, is this nornal?Some features run on the CPU: film detection & reverse pulldown; black bars detection; and possibly hdr "measure each frame's peak luminance" but I'm not sure about that last one.
(Edit: and when using native decoding, those features cannot work obviously)

[aufkrawall]

even with DXVA native i get easily 12 x more CPU usages than EVR. in my eye it is should be investigated.
12 times more? How high would that be?

I only get around 2-2.5% with DXVA2n/D3D11VAn, which isn't much more than EVR or the crappy Windows 10 video app.

Copyback modes weirdly show higher CPU usage with madVR than with EVR, especially D3D11VA-CB.

[huhn]

with EVR about 2% at ~1.5 GHZ with madVR about 10-12 @ 3.4 ghz

[HillieSan]

NGU is very slow on polaris compared to pascal/maxwell or even GCN 1-2.

Very slow? Perhaps a bit slower. Madvr runs fast here on polaris RX 480. Lot faster than in the beginning.

[huhn]

can you please use number...

just to give you an idea a 1060 can do NGU very high for 1080p->UHD with normal chroma.
that's easily 2x of that what a RX 480 can do. that'S has nothing todo with "a bit" slower.

[mclingo]

Some features run on the CPU: film detection & reverse pulldown; black bars detection; and possibly hdr "measure each frame's peak luminance" but I'm not sure about that last one.
(Edit: and when using native decoding, those features cannot work obviously)


Thanks

[aufkrawall]

with EVR about 2% at ~1.5 GHZ with madVR about 10-12 @ 3.4 ghz
I don't think that is normal or your CPU is really slow.
6700k stays at 800MHz here when using native decoding and balanced energy profile.
With modern CPUs like Skylake or Ryzen it's btw. better to set power profile to maximum performance and let the CPU itself decide the clocking (tools read out full clock then, but that's not true). I have to set C-States to auto for this, checked it with a wattmeter.

[HillieSan]

can you please use number...

just to give you an idea a 1060 can do NGU very high for 1080p->UHD with normal chroma.
that's easily 2x of that what a RX 480 can do. that'S has nothing todo with "a bit" slower.

I don't have numbers, sorry. There are lots of numbers in game benchmarks. But these do not apply here.

I confirm that the RX 480 cannot do NGU very high.

[huhn]

create a folder with ShowRenderSteps in the madVR folder and you can get numbers.

[dvd1]

the GTX 1050 for 4k is not good, would it be too small?

[huhn]

it can be used for 4k

[mclingo]

I don't have numbers, sorry. There are lots of numbers in game benchmarks. But these do not apply here.

I confirm that the RX 480 cannot do NGU very high.

it works ok if you are just using that for upscalling but it really struggled with it on CHROMA for me on nearly all 4k and 3D stuff. i just ended up using NGU for upscaling and laczos for CHROMA

[Manni]

EVR-CP doesn't support the special XySubfilter interface, so that comparison is quite mood, since it doesn't enter the picture anyway.

Anyway, it plays perfectly for me, and XySubfilter was never installed on my system. madshi could only reproduce it with XySubfilter installed, once he removed it, it was fine. So either you didn't really remove it, or your setup is screwed up even further then that.

My set-up isn't screwed. I have tried uninstalling XYsubfilter (using the uninstall batch file ran as an administrator), then I rebooted. XYsubfilter appeared as "uninstalled" in MPC-BE. That didn't make any difference, the issue was still there.

In any case, I thought XYSyubfilter was the recommended subtitle plugin to use with MadVR, as Madshi was involved in its development.

IF there is a problem with XYSubfilter, then what I am supposed to use with MPC-BE/LAV/MadVR to render subtitles in SDR/3D/HDR/10bits?

He uses MPC-BE, which does not, as far as I know. Additionally, LAV Video only uses this interface with madVR, so with any other renderer there would be no conflicts.

MPC-BE support XySubFilter long time ago.

I've NEVER had any issue with MPC-BE/LAV/MadVR/XYSubfilter except recently with this DVD title, so XYsubfilter is clearly supported indeed.

No idea why this can't be fixed if this is a MadVR/XYSubfilter issue (which doesn't seem obvious here), but I guess I'll sacrifice a few DVDs rather than stopping using XYSubfilter, unless there is a recommended alternative subtitles option for MPC-BE that works as well with MadVR.

I have no intention to switch to MPC-HC, I prefer MPC-BE by far :).

[clsid]

XYsubfilter development has been dead for years. Madshi isn't involved at all, except for making a few fixes for it in the past.

It is recommended to use the internal subtitles renderers of MPC-HC/MPC-BE. Those work fine and are reliable.

[madshi]

Excruciatingly I could make three. #16-19 (https://drive.google.com/open?id=0ByfdfPvnoDuzWDUySEFpdUQ3aVk) (17 and 18 are from the same hang but the latter is after the alert sound, when unusually the player was still hung)
Thanks, will have a look at these later.

How would I go about creating a 3dlut for 4K HDR videos, that would take advantage of my SDR television's (panny st60) native color gamut? I have a colorimeter/spectro/calman and have played around with it some, but i can't figure out what the correct way to do this would be. Any ideas?
I don't think Calman can do an HDR -> SDR conversion 3dlut. So basically what you would do is create a normal SDR Calmut 3dlut for an extended color space like DCI-P3 and store it in the appropriate slot in madVR. And then simply activate HDR -> SDR conversion by using pixel shader math in madVR. That way madVR will first convert HDR to SDR, and then send it through the DCI-P3 SDR 3dlut.

there is a crash report here :D

v0.92.7 & v0.92.6 crash

v0.92.5&v0.92.4&v0.92.3 unknow，not test
Unfortunately attachments in this forum never ever get approved. So please upload these attachments elsewhere and then link to them here.

I've been struggling with several combinations the last week, since (all of a sudden, probably a TV FW upgrade) switching to HDR resulted in a purple video image. I can work-around this problem by toggling the TV OFF & ON again.
Could be a GPU driver or TV firmware problem. Hard to say. I hope you've configured your GPU to output Full Range (0-255) RGB?

As a sidenote: After installing the 1709 update, my desktop colors were really dark. I got this solved by running Display Color Calibration.
That doesn't sound good. It sounds like adding more processing on top of something that is not working correctly.

HTPC running Kodi 17.4 DSPlayer + madVR

Connected to Denon AVR X1200
Beamer Acer M550 4K HDR connected to Denon
Used HDMI cables are fine and support all i need

I can set 4k resolution in NVCC
60Hz but only RGB Full @8Bit

HEVC Playback works fine but HDR seems very dark
Beamer is switching to 4K 60Hz and recognize HDR automatically
I can stop a video file and start another one without issues.

I can also set 1080p 60Hz RGB Full with 12Bit in NVCC
I can play one HEVC 4K HDR file
Beamer is switching to 4K 60Hz and recognize HDR automatically
I can stop the file but i can't play another one

If i start another file Beamer is Switching to 4K 60Hz HDR, but i get back on the desktop and there is just a black square about 25% over my desktop. Have to kill Kodi with Taskmanager.

It seems there is a problem while switching the resolution after playing one file. I also played around with exclusive full screen, but that did not help.

Is there a resolution for dark HDR image quality if Desktop is set to 4k resolution?

Is there a way to fix the second playback issue for 4K files if desktop resolution is set to 1080p?
I'm not sure why you get dark HDR images. It sounds as if the black/white levels are getting screwed up somewhere, for some reason. Is it possible that your TV has different settings for HDR vs SDR sources? If so, you may have to edit the input levels settings (PC vs TV levels, 0-255 vs 16-235) for HDR vs SDR in your TV. It's also possible that your TV in HDR mode always expects levels to be TV levels, and only supports PC levels for SDR content. I don't really know, though. It's very hard to analyze these kind of problems remotely.

The other problem: Is it limited to Kodi? Or does it also happen if you use a different media player (e.g. MPC-HC/BE), just for testing?

These new algorithms pretty darn good for image processing:

http://screenshotcomparison.com/comparison/121039

http://screenshotcomparison.com/comparison/121038

http://screenshotcomparison.com/comparison/121040

I think you should release a tool or even better a paint.net plugin for these, they're worthy of it. I couldn't get a palatable result result for these images with paint.net filters ( even including some third party plugins)

Ironically I don't find them very practical for video, because even these photos took several minutes of trial and error, videos would take even longer and I'd have to optimize for ever video (or at least ones that are not related to each other). Also my GPU is not beefy enough to run these too in real time.
I may consider creating a Photoshop plugin at some point in the future. There might be different algo speed levels available in the future. But yes, these algos need to be manually adjusted per video, which does reduce their real life suitability.

Someone? :(
I've replied to your posts in some detail, but I don't think you've replied to my latest suggestions and comments. I'm not going to add any more hints until you reply to all my suggestions and comments.

MADVR is swapping resolutions and refresh rates to match the video, i'm using win 10 latest build ATi RX 550 latest driver.

it also happended when I switch my desktop from 1080p to 2160p manually so it may in fact be a windows 10 or ATI problem, I might roll back some GFX drivers this weekend.
It could be a GPU (driver) problem, or also a display problem. I have a similar problem sometimes when switching HDR on/off with my Sony 4K TV. I'm not actually sure if it's the TV's fault or the GPU's fault. I don't think there's too much I can do in madVR, unfortunately. madVR doesn't actually do much in terms of GPU <-> display communication. madVR just asks for a specific display mode to be used. And if that fails, it's probably either a problem with the GPU (driver) or the display. Or maybe madVR changes modes too quickly for the TV to handle, or something like that. I don't really know.

Ctrl+J OSD when playing 3D.ISO MVC content as follows:

http://s4.toldacuccot.hu/letoltes?sid=0e20b221e3241825b337104bd2786541&file=madVR_screenshot2_3D.png
FWIW, your quotes are often missing the link to our discussion history, so I have to manually search back through old posts to find the history of what we were talking about exactly. Please in the future leave the quote mechanic fully intact, so I can click back through the related past comments.

Your screenshot suggests that madVR doesn't even know that you're playing a 3D movie. Do you have LAV installed with the optional 3D Intel Media SDK 3D decoder? Obviously, if madVR doesn't know that the movie is 3D, then it can't activate the 3D profile.

RCA is less effective on blocking artifacts than on mosquito noise etc., so a touch of RRN is useful for sources with blocking.
Really? In my tests it works reasonably well on blocking artifacts.

I assume that dedicated deblocking algorithm would be better and could preserve some more detail
The RCA algorithm tries to interpret the image structures correctly. Sometimes image edges can lie on exact macroblock borders, so madVR can't simply remove all image edges that happen to lie on macroblock borders. A "simple" dedicated deblocking algorithm might tend to remove too many edges.

Did you get a chance to look at the DVD playback issue with the Klute sample?
Yes. The issue is that madVR is currently limited to only support one provider of the "new" subtitle renderer interface. Currently there's XySubFilter, AssFilterMod and LAV Video Decoder which support/use this interface, so if 2 (or more) of them try to connect to madVR, only one gets a valid connection. When testing your Klute sample, I found that XySubFilter got the connection and LAV Video Decoder was refused, that's why the menu didn't work. When I changed the active subtitle renderer, the problem went away.

I've read in your other comments that you've already uninstalled XySubFilter. Did you also remove (or rename) the whole XySubFilter folder, just to be safe? FWIW, just switching to a different subtitle renderer in the MPC-HC/BE settings should suffice. Do you maybe also have AssFilterMod installed? That might be another explanation for the problem.

It's on my to do list to support more than one subtitle renderers using the "new" subtitle interface.

[madshi]

AdaptiveSharpen:

Understandable, if it has to go, it has to go. I have no shortage of sharpening options and artefact reduction compromises so it's fine. Yes, i've tried, i perfer no linear light, it helps just a bit.
However i'm sincerely seeing AS being different from from one another (where the line is drawn and attention to fine detail):
original - https://postimg.org/image/yq0j0scpr/
newASnoLL 2.0 - https://postimg.org/image/dt4aw6c4v/
oldAS 2.0 - https://postimg.org/image/9786nut6n/
I dont see that for example in LumaSharpen and the newAS (just a lot more ringing in LumaSharpen):
AS 1.0 - https://postimg.org/image/uosmmoz7j/
LS 4.0 - https://postimg.org/image/hxegg92b3/
Yeah I saw that earlier too, the old AS darkens lines wonderfully, really helps with lower res anime. Since I'm using Avisynth on the lower res stuff I can just use Hysteria to darken lines a touch, but maybe madshi can add a line darkener to madVR himself sometime to help the situation.
Ok, thanks for the test image, it's always helpful to see an example that demonstrates what you mean. I'll have a deeper look at this later.

For film sources, I think AS linear light can be removed. From my ISF calibrated display, it boosts the contrast and could cause eye strain on prolong staring at the screen.
AS Feedback) So I've spent some time with all these new algorithms. Lets start off with AS. I prefer the current AS for real life content. The old one was unusable. I don't like a lot of it, I'm running with a .2 strength on my 720p profile. I do not like AS with LL, softens/blurs the image ever so slightly.

The old AS was better for anime as it helped enhance dark lines and edges, however it was horrible for non anime. I never used it in the past as I don't watch much anime so, my preference is the new one.
It seems to me that the majority of users (but not all) seem to prefer linear light *off*.

---------------------------

Reduce compression artifacts:

its difficult choice,for Anime content, some times i forcd to use RCA*7 or *10 to clean the image, while "don't reduce texture detail" left little artifacts, and adding some RRN to fix it cost extra gpu power, and blur the image slightly, i really do not know, but i think get rid of "don't reduce texture detail" will be good for me, thanks for your effort @madshi.
Looking again the levels 1-2 with DRTD it actually does very little, I'm happy on sticking with original setting of 1 or 2. Surprisingly 2 with DRTD almost looks softer than without it? I'm not fussed what happens here now.
I did some tests and for my part I prefer strength 1-2 without "Don't reduce texture detail" activated because I can't see better details preservation with these low strenghts but I definitely see more compression artifacts.

For strength 3-10, I prefer when this option is activated, it does help to retain some details.

Strength 3 is quite debatable though... It depends if you value more perserving details or reducing compression artifacts. I switched on and off a lot of times but each time, I've found that I have a slight preference for "Don't reduce texture detail" activated in this case.
So it seems you guys agree on *not* using "don't reduce texture detail" on strength levels 1-2 or 1-3. But ABDO doesn't like the option at all, while ryrynz and Werewolfy prefer it on for higher strengths.

@ryrynz and @Werewolfy, do you happen to have a few screenshots which demonstrate the usefulness of the "don't reduce texture detail" option with higher strengths?

Generally, of course I've love for the option to only preserve detail but not artifacts. In real life sometimes it's hard to differ between one and the other, so obviously activating "don't reduce texture detail" will preserve some detail *and* some artifacts. So it's a valid question if it's a good option at all. I suppose the key question is if it preservers more details or more artifacts?

It seems I was late for the discussion, but anyway: I prefer "don't reduce texture detail" for RCA strength 1-5 maybe 1-4. Not sure about higher strength RCA, I'll check later.
Hmmmm... That's exactly the opposite of what Werewolfy suggests. Werewolfy suggests to turn "don't reduce texture detail" *off* for lower strengths and on for higher strengths.

Uh, this will be tough to decide... :(

with RCA@1 DRTD doesn't look nearly as soft but also removes less artifacts indeed so it's a mixed bag, I guess you'd need to work your magic again and refine it please :p
I don't think I can refine it any further right now. We will just have to come to a decision on whether to use DRTD or not, at each strength level.

[Razoola]

madshi, I though xysubfilter was the preferred subtitle filter for use with madVR. Is it now the case its better to use MPC-HC internal subtitle filter? Does that support moving the subtitles into the black bars etc?

[Jtacdf]

The issue is that madVR is currently limited to only support one provider of the "new" subtitle renderer interface. Currently there's XySubFilter, AssFilterMod and LAV Video Decoder which support/use this interface, so if 2 (or more) of them try to connect to madVR, only one gets a valid connection. When testing your Klute sample, I found that XySubFilter got the connection and LAV Video Decoder was refused, that's why the menu didn't work. When I changed the active subtitle renderer, the problem went away.


So both MPC-HC and MPC-BE ISR does not use the subtitle renderer interface? That might explain why adjusting color and gamma settings in madvr resulted in subtitles not doing the same.
I did a minor test of setting saturation to -65 in madvr and the results are below.

Saturation -65, MPC-HC/BE ISR https://www.mediafire.com/file/d99gurail580qij/Saturation%20-65%2C%20MPC-HC%2C%20MPC-BE%20ISR.png

Saturation -65, Assfilter, XySub https://www.mediafire.com/view/1m7bazzsqo16ygm/Saturation%20-65%2C%20Assfilter%2C%20XySub.png

Original, Saturation 0 http://www.mediafire.com/view/739584cvf8h5e6c/MPC-BE-ISR.png

[YxP]

We will just have to come to a decision on whether to use DRTD or not, at each strength level.

Maybe you just named the new option wrong and people subconsciously think it's better since it doesn't remove texture detail ;)

I agree with someone who said it earlier (can't remember who), but when artifacts are bad enough to enable the filter, I use it without the option.

[YxP]

Does that support moving the subtitles into the black bars etc?

You can change the place of the subtitles whereever you want in the player options, but no automatic moving.

[mparade]

FWIW, your quotes are often missing the link to our discussion history, so I have to manually search back through old posts to find the history of what we were talking about exactly. Please in the future leave the quote mechanic fully intact, so I can click back through the related past comments.

Your screenshot suggests that madVR doesn't even know that you're playing a 3D movie. Do you have LAV installed with the optional 3D Intel Media SDK 3D decoder? Obviously, if madVR doesn't know that the movie is 3D, then it can't activate the 3D profile.

I have JRiver and the latest Nvidia driver package installed for my GTX 1080 Ti special edition. I am pretty sure that JRiver is based on LAV filters thanks to Nevcairiel (JRivers' plugin directory includes LAV filters :cool:). On top of it all, "3D Blu-ray and mkv 3D decoding" is enabled in JRiver.

P.S. Sorry for the missing link, I will take care of it next time.

[Neo-XP]

AdaptiveSharpen:

After testing with and without LL, I get the best result with NGU Sharp without any other image enhancement (just some added grain).

However, NGU Sharp is still not at good as it was in v0.91.1 with little details in the image and aliasing.
Do you plan to improve it before v1.0 ?

If so, you may take a look at this if you need test images :

https://forum.doom9.org/showthread.php?p=1804708#post1804708
https://forum.doom9.org/showthread.php?p=1804766#post1804766

The aliasing is quite distracting sometimes while watching a movie.
The improvement I see in current NGU is less dark halos comparing to v.91.1.

Reduce compression artifacts:


I agree for not using "don't reduce texture detail" at lower strengths (1-3), because at best it does nothing and leaves compressions artifacts in the image.
Is there a way to detect if a source material needs RCA so we will not have to activate it manually depending on the content ?

[Werewolfy]

Check this (https://mega.nz/#!vNonVDBR!iEjopF1VnUKhQmBM3J1lPQr2Tv_NOQGk0Nt2mUHS62E) out, cut from a particularly low quality release of '94 cartoon, The Tick.

Personally I feel 11 and 12 are so heavy that DRTD should be enabled, I can't imagine any scenario where having it off would be beneficial. Even with the above video being almost as bad as they come I still think the absolute limit I would ever use on that is 10, above that as resemblance of lines/objects pretty much disappear. 3 and above with DRTD has my vote.

Thanks for the video! Really helpful.

I've tested strength 11 and 12 on this sample but "don't reduce texture detail" doesn't seem to make a difference. It's maybe so high that this option can't make the big texture detail reduction unavoidable.


So it seems you guys agree on *not* using "don't reduce texture detail" on strength levels 1-2 or 1-3. But ABDO doesn't like the option at all, while ryrynz and Werewolfy prefer it on for higher strengths.

@ryrynz and @Werewolfy, do you happen to have a few screenshots which demonstrate the usefulness of the "don't reduce texture detail" option with higher strengths?

Generally, of course I've love for the option to only preserve detail but not artifacts. In real life sometimes it's hard to differ between one and the other, so obviously activating "don't reduce texture detail" will preserve some detail *and* some artifacts. So it's a valid question if it's a good option at all. I suppose the key question is if it preservers more details or more artifacts?


1) Just to demonstrate my point about RCA 1 : If you activate "don't reduce texture detail" on strength 1, this strength becomes totally useless. There is very little to no difference.

Original VS RCA 1 DRTD (http://screenshotcomparison.com/comparison/121286)

If you activate RCA1 without this option, you can see that artifacts are a little bit reduced.

Original VS RCA 1 (http://screenshotcomparison.com/comparison/121287)

2) "don't reduce texture detail" is useful with high strengths... but sometimes it doesn't actually work very well... I've had more time to test it and I'm not sure if I still really like this option.

First, here is an example where I find it useful : This is an extreme exemple but even if there are more artifacts around the "Judges" letters you can clearly see that a lot of details are totally blurred without this option.

RCA 10 VS RCA 10 DRTD (http://screenshotcomparison.com/comparison/121281)

But here's another scenario : RCA 4 VS RCA 4 DRTD (http://screenshotcomparison.com/comparison/121288)
Here I prefer the look of RCA 4 without "don't reduce texture detail".

And then I've found this example :

RCA 6 VS RCA 6 DRTD (http://screenshotcomparison.com/comparison/121289)

There are mot artifacts around the lips and nose of the character than with RCA 4 DRTD actually :

RCA 4 DRTD VS RCA 6 DRTD (http://screenshotcomparison.com/comparison/121290)

Is it a bug?
Here what happens at 10 : RCA 10 VS RCA 10 DRTD (http://screenshotcomparison.com/comparison/121291)

Here's the video I used to do this comparison : Original video (https://mega.nz/#!ovB1VTJC!rWMhReF8ziGnJv0U0GJ2-K3CK421M0RpV4_MMJjhO8Y)
I know it's a very compressed video but a less compressed version of this video has the same issues. Video (https://mega.nz/#!FvRjQaQB!NqannBKVUv935qWxBtAgNjFFrIVlnX8F_ZV9PFC62W0)
You can find the other video I used in a previous post by ryrynz :o

[madshi]

madshi, I though xysubfilter was the preferred subtitle filter for use with madVR. Is it now the case its better to use MPC-HC internal subtitle filter? Does that support moving the subtitles into the black bars etc?
They all have advantages and disadvantages. When and if XySubFilter works, it has some technical advantages, because madVR has more control over everything. But it crashes with some files, has problems with external SRT files when trying to "move" subtitles, and the developer has gone MIA.

So both MPC-HC and MPC-BE ISR does not use the subtitle renderer interface? That might explain why adjusting color and gamma settings in madvr resulted in subtitles not doing the same.
Whether or not subtitles are affected by color and gamma settings depends on in which order subtitles are drawn and color & gamma settings are applied. Usually I'm doing color & gamma stuff relatively early, because it's faster to do it before upscaling (less pixels to work on). Subtitles are usually drawn after scaling - except when talking about bitmap subtitles. So it's sometimes this way and sometimes that way.

I have JRiver and the latest Nvidia driver package installed for my GTX 1080 Ti special edition. I am pretty sure that JRiver is based on LAV filters thanks to Nevcairiel (JRivers' plugin directory includes LAV filters :cool:). On top of it all, "3D Blu-ray and mkv 3D decoding" is enabled in JRiver.
Well, all I can say is that if the decoder delivers 3D to madVR then madVR's OSD says: "NV12, 8bit, 4:2:0 (3D)", which it doesn't in your case, so for some reason the decoder isn't delivering 3D to madVR. I don't know the exact reason why. It's unlikely to be madVR's fault, though. Maybe it would make sense to ask in the LAV thread for help how to find out the reason why LAV doesn't send 3D to madVR? I don't know the exact circumstances when it might fail, except when the optional 3D decoder is not downloaded/installed.

NGU Sharp is still not at good as it was in v0.91.1 with little details in the image and aliasing.
Do you plan to improve it before v1.0 ?

If so, you may take a look at this if you need test images :

https://forum.doom9.org/showthread.php?p=1804708#post1804708
https://forum.doom9.org/showthread.php?p=1804766#post1804766

The aliasing is quite distracting sometimes while watching a movie.
The improvement I see in current NGU is less dark halos comparing to v.91.1.
It's on my to do list.

Is there a way to detect if a source material needs RCA so we will not have to activate it manually depending on the content ?
I wish. Right now I don't know a good way how to detect that, but maybe I'll have an idea in the future.

Maybe you just named the new option wrong and people subconsciously think it's better since it doesn't remove texture detail ;)
Haha, fair enough!

I've tested strength 11 and 12 on this sample but "don't reduce texture detail" doesn't seem to make a difference. It's maybe so high that this option can't make the big texture detail reduction unavoidable.
I've done some tests with that video at strength 9 and found several frames where the DRTD (don't reduce texture detail) option resulted in visible artifacts being left in the image, although 99% of the image were smoothed, which looked rather weird and distracting.

1) Just to demonstrate my point about RCA 1 : If you activate "don't reduce texture detail" on strength 1, this strength becomes totally useless. There is very little to no difference.

Original VS RCA 1 DRTD (http://screenshotcomparison.com/comparison/121286)

If you activate RCA1 without this option, you can see that artifacts are a little bit reduced.

Original VS RCA 1 (http://screenshotcomparison.com/comparison/121287)
Yes, that's a good example.

2) "don't reduce texture detail" is useful with high strengths... but sometimes it doesn't actually work very well... I've had more time to test it and I'm not sure if I still really like this option.

First, here is an example where I find it useful : This is an extreme exemple but even if there are more artifacts around the "Judges" letters you can clearly see that a lot of details are totally blurred without this option.

RCA 10 VS RCA 10 DRTD (http://screenshotcomparison.com/comparison/121281)

But here's another scenario : RCA 4 VS RCA 4 DRTD (http://screenshotcomparison.com/comparison/121288)
Here I prefer the look of RCA 4 without "don't reduce texture detail".

And then I've found this example :

RCA 6 VS RCA 6 DRTD (http://screenshotcomparison.com/comparison/121289)

There are mot artifacts around the lips and nose of the character than with RCA 4 DRTD actually :

RCA 4 DRTD VS RCA 6 DRTD (http://screenshotcomparison.com/comparison/121290)

Is it a bug?
Here what happens at 10 : RCA 10 VS RCA 10 DRTD (http://screenshotcomparison.com/comparison/121291)

Here's the video I used to do this comparison : Original video (https://mega.nz/#!ovB1VTJC!rWMhReF8ziGnJv0U0GJ2-K3CK421M0RpV4_MMJjhO8Y)
I know it's a very compressed video but a less compressed version of this video has the same issues.
Atm I think I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image, even at very high algo strength, which doesn't make too much sense.

Any strong objections, anyone?

[veggav]

Quick question:

I have a PC with an i5 3570k and gtx 970, with 16gb of ram and I have a 4k TV that can't play HDR.
So I'm trying to use MadVR to do HDR to SDR.

I tried every performance option and selecting DXVA2 for everything but it's still stuttering.

Do I have any chance playing this file here:

MediaInfo:
General
ID : 0 (0x0)
Complete name : \BDMV\STREAM\00000.m2ts
Format : BDAV
Format/Info : Blu-ray Video
File size : 88.4 GiB
Overall bit rate mode : Variable
Maximum Overall bit rate : 109 Mbps

Video
ID : 4113 (0x1011)
Menu ID : 1 (0x1)
Format : HEVC
Format/Info : High Efficiency Video Coding
Format profile : Main 10@L5.1
Codec ID : 36
Width : 3 840 pixels
Height : 2 160 pixels
Display aspect ratio : 16:9
Frame rate : 23.976 fps
Color space : YUV
Chroma subsampling : 4:2:0
Bit depth : 10 bits
Writing library : ATEME Titan File 3.7.9 (4.7.9.0)
Color primaries : BT.2020
Matrix coefficients : BT.2020 non-constant
Color range : Limited

or I need to replace my 970 for a 1060?

[Manni]

Yes. The issue is that madVR is currently limited to only support one provider of the "new" subtitle renderer interface. Currently there's XySubFilter, AssFilterMod and LAV Video Decoder which support/use this interface, so if 2 (or more) of them try to connect to madVR, only one gets a valid connection. When testing your Klute sample, I found that XySubFilter got the connection and LAV Video Decoder was refused, that's why the menu didn't work. When I changed the active subtitle renderer, the problem went away.

I've read in your other comments that you've already uninstalled XySubFilter. Did you also remove (or rename) the whole XySubFilter folder, just to be safe? FWIW, just switching to a different subtitle renderer in the MPC-HC/BE settings should suffice. Do you maybe also have AssFilterMod installed? That might be another explanation for the problem.

It's on my to do list to support more than one subtitle renderers using the "new" subtitle interface.

Thanks Madshi, that makes a lot more sense.

In fact, it has allowed me to resolve the problem, which doesn't seem come from XySubFilter... but from LAV!

I tried to play the same file from my Win10 VM on the Mac and the menu played fine, whether XySubFilter was selected or not.

So I did some debugging on the HTPC and realized that in order to NOT have the issue, I had to select anything but D3D11 in LAV Video configuration (on the MacBook Pro VM it works fine with D3D11 too).

If I select any other hardware acceleration option on the HTPC (I tried none and DXVA2 Copy-back), the menu displays perfectly both with XySubFilter and with another subtitle option.

Does that make any sense to you/NevCairiel? You said the issue disappears if you don't select XySubFilter, did you try without D3D11 in LAV?

That would explain why I didn't have the issue the last time I played this title, because D3D11 wasn't available in LAV then.

PS: just to answer your question, I had not installed any other subtitle renderer and I had renamed the XySubFilter folder after uninstalling it, that didn't help. I also tried to simply select the internal MPC-BE subtitle renderer, that didn't help either, even when XySubFilter was uninstalled (and reported as uninstalled by MPC-BE).

[el Filou]

@veggav : the 970 only has hybrid HEVC decoding. I doubt it can handle the bitrate of an UHD Blu-ray.
Try to pass this file through DXVA Checker's playback benchmark and compare different decode methods (software/CUVID/DXVA).

Note: I don't know if it's a must-have for you, but FYI the 1060 can't decode VP9 10-bit (YouTube HDR) in hardware.

[Werewolfy]

Atm I think I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image, even at very high algo strength, which doesn't make too much sense.

Any strong objections, anyone?

No, I think it's safer to disable this option. The results are too unpredictable and when it doesn't work well it's really bad.

[ABDO]

Atm I think I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image, even at very high algo strength, which doesn't make too much sense.

Any strong objections, anyone?
No, tottly agree with that, RCA is the most powerfull and Effective algorithm i have ever use, and Unfortunately DRTD seem to be weakness point for it.

[tyrindor]

After updating to Windows 10 Fall update, exclusive fullscreen mode no longers works just black video with audio. I am using MPC-BE. Hardware acceleration is set to DXVA2 (copy-back).

Updated both MadVR and MPC-BE with no results.

[veggav]

@veggav : the 970 only has hybrid HEVC decoding. I doubt it can handle the bitrate of an UHD Blu-ray.
Try to pass this file through DXVA Checker's playback benchmark and compare different decode methods (software/CUVID/DXVA).

Note: I don't know if it's a must-have for you, but FYI the 1060 can't decode VP9 10-bit (YouTube HDR) in hardware.

Never heard of a DXVA checker. Going to try it.

Does the 1070 have VP9 10bit hardware decoder?

EDIT: Just tested it and I got 42 fps with 98% GPU use and 40% CPU use. Played with 64bit.
So.. why it's stuttering? Maybe because of HDR to SDR conversion?

DOUBLE EDIT: Tried the 32bit playback benhchmark, it was down to 28fps.
So I guess I need 64bit version of MadVR?

[asmo42]

Atm I think I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image, even at very high algo strength, which doesn't make too much sense.

Any strong objections, anyone?
No my findings pretty much echo what you describe. I had one bad video where without DRTD it started weakening artifacts around some letters from strength 1 and they were completely gone at strength 4.

However with DRTD on nothing happened with these particular artifacts until strength 11 where obviously the rest of the picture was ruined at that point.

[austonrush]

Atm I think I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image, even at very high algo strength, which doesn't make too much sense.

Any strong objections, anyone?

nope I agree, remove it!

[Magik Mark]

After updating to Windows 10 Fall update, exclusive fullscreen mode no longers works just black video with audio. I am using MPC-BE. Hardware acceleration is set to DXVA2 (copy-back).

Updated both MadVR and MPC-BE with no results.



Confirmed. MPC BE bug. DsPlayer working just fine


Sent from my iPhone using Tapatalk

[Aleksoid1978]

Confirmed. MPC BE bug. DsPlayer working just fine


Sent from my iPhone using Tapatalk

It's not a MPC-BE bug. Internal EVR-CP perfect working in FSE mode.

[wushantao]

hi @madshi

there is a crash report here :D

v0.92.7 & v0.92.6 crash

v0.92.5&v0.92.4&v0.92.3 unknow，not test


crash step：
1、play a movie file
2、full screen
3、pause play back（don’t touch anything，just stop）
4、auto close screen（by power management setting）
5、wait 1 minute more
6、open screen（move mouse or touch any keyboard）
7、you can see crash now

other info：
win10 x64 15063、mpc be x64 v1.5.2、lav x64 v0.70.2、16G ram、i7 4760HQ、iris 5200

there is a crash report and settings.bin

16093
16094
i hope this information is useful for bug fix:thanks:

update：

win10 x64 1709（10.0.16299.19）

not crash，but black screen，can‘t see any image，audio is ok


Unfortunately attachments in this forum never ever get approved. So please upload these attachments elsewhere and then link to them here.



@madshi

https://www.dropjiffy.com/f/0XGTIKYO
http://www.solidfiles.com/d/dkNGYNV38NGAY
http://katfile.com/xfn10nq0r9xp

3 url，same file

thx for your reply，wish this bug to be fixed

[nsnhd]

Originally Posted by madshi

I don't think Calman can do an HDR -> SDR conversion 3dlut. So basically what you would do is create a normal SDR Calmut 3dlut for an extended color space like DCI-P3 and store it in the appropriate slot in madVR. And then simply activate HDR -> SDR conversion by using pixel shader math in madVR. That way madVR will first convert HDR to SDR, and then send it through the DCI-P3 SDR 3dlut.
Can you hint more about how to combine 3dluts and HDR -> SDR conversion ? Where is the slot to store a 3dlut ?
Right now I only have 2 choices, either to use HDR -> SDR conversion by using pixel shader math OR by using an external 3DLUT, but not combining both of them.

[Asmodian]

The output of the SDR conversion is run through the standard SDR 3DLUT set for your display. Once it is SDR DCI-P3 the normal DCI-P3 3DLUT is appropriate.

[huhn]

Never heard of a DXVA checker. Going to try it.

Does the 1070 have VP9 10bit hardware decoder?
no the 1030, 1050/ti and maybe the 1080 ti are supposed to have a vp9 10 bit decoder.
EDIT: Just tested it and I got 42 fps with 98% GPU use and 40% CPU use. Played with 64bit.
So.. why it's stuttering? Maybe because of HDR to SDR conversion?
if your gpus is maxed from decoding how should it use madVR?
DOUBLE EDIT: Tried the 32bit playback benhchmark, it was down to 28fps.
So I guess I need 64bit version of MadVR?

only depands on your player. but you are pretty much wasting your time hybrid decoding is in general bad but hybrid decoding and madVR is unlikely to work and the higher the bit deep the slower this decoding is getting.

[leeperry]

I'll probably remove (always disable) the "don't reduce texture detail" option, because it sometimes leaves ugly artifacts in the image

Well, vanilla RCA@1 looks pretty soft with AB25% SSIM2D and I like what AB25% does to SSIM2D.....if refining is no option, could you please lower the actual strength of @1 then? Some videos call for deblocking but because it can't quite differentiate artifacts and details it definitely oversmooths things on non-hopeless footage IME :o

[ryrynz]

Drop 12 add a new 1 and shuffle em up xD

[HomeY_]

I assume you are using Nvidia Control Panel to change colors right?

You change color in Resolution tab?

Yes for both.

Could be a GPU driver or TV firmware problem. Hard to say. I hope you've configured your GPU to output Full Range (0-255) RGB?
Yeps, full RGB, 8bit is set in Nvidia CP. I suspect the TV FW, since the colors & switching to/from HDR was working fine before.
When i start playing HDR content, it results in purple video, after turning the TV OFF & ON, it displays proper colors & HDR (see attachments).

That doesn't sound good. It sounds like adding more processing on top of something that is not working correctly.
Actually i didn't change any settings. I 'clicked' through and when comparing with previous settings, the new settings were a lot brighter.

[Neo-XP]

Any strong objections, anyone?

No, the image is cleaner with it off.

Here is what RCA can do at 4 with grain added to a pretty bad source :

http://screenshotcomparison.com/comparison/121322

I really like the result in motion, no more disturbing artifacts :)

[mzso]

Thanks, will have a look at these later.
Hopefully something positive will come out of it.

I may consider creating a Photoshop plugin at some point in the future. There might be different algo speed levels available in the future. But yes, these algos need to be manually adjusted per video, which does reduce their real life suitability.

Cool. Are you planning to sell it for money?
If you're not a plugin for a free and simple image editor would be far more accessible for people like me who only use these casually.

(I mentioned paint.net, but there might be others. Not sure if there's anything is more widespread or appropriate. I'm certainly not familiar with any.)

[Rexian]

I started having this problem a few weeks ago with no change I could point to. Since then I upgraded to the latest versions of pretty much everything -
madVR 0.92.5 then 0.92.7 today
mpc-hc 1.17.11 then 1.17.13 today
Lav 0.70.2
nvidia 287.92 (updated from 285.xx today)

Both mpc-hc and mpc-be work fine in a window (TV recognizes HDR videos) but the screen goes dark as soon as I press alt+enter (audio continues) and then when the picture comes back after few seconds, it crashes.

Here are the settings -
nvidia
http://i63.tinypic.com/24b3huc.jpg

LAV
http://i64.tinypic.com/2hnbbck.png

madVR prop
http://i66.tinypic.com/110g2ro.jpg

madVR reder
http://i63.tinypic.com/xnf2q9.jpg

Also tried disabling FSE. Screen doesn't go dark but crashes right away -
http://i67.tinypic.com/n1oynb.jpg

Updating to W10 Build 1709 now... anything else should I try? Thanks for any pointer.

[fedpul]

Hi, quick question. Can I create profiles based on bitrate? Thanks in advance.

[Q-the-STORM]

Hi, quick question. Can I create profiles based on bitrate? Thanks in advance.

Quick Answer: No.

[Anima123]

Madshi,

As we know, there are much more blocky when video is in motion than static views, especially if the video stream is poorly encoded, like constant bitrate for both scenes. Is it possible for some kind of adaptive strength or algorithm been applied to RCA in the future?

[HillieSan]

After updating to Windows 10 Fall update, exclusive fullscreen mode no longers works just black video with audio. I am using MPC-BE. Hardware acceleration is set to DXVA2 (copy-back).

Updated both MadVR and MPC-BE with no results.

Confirmed. MPC BE bug. DsPlayer working just fine

I had the same problem. It appeared that the D3D11 setting in madvr (rendering->general settings->Use Direct 3D 11...) was disabled. I enabled it and restarted MPC BE and problem was solved.