Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Hardware & Software > Software players

Reply
 
Thread Tools Search this Thread Display Modes
Old 15th July 2016, 18:50   #38741  |  Link
Chyrka
Registered User
 
Join Date: Oct 2011
Location: Kyiv, Ukraine
Posts: 24
Hi.
I have something strange with playback.
My main GPU is Intel HD3000. I also have Nvidia 210 in the PC but it is not connected to the monitor.
I am using MPC-HC x86 1.7.10.
The choma scaling algorithm is set to default SoftCubic but for some reason it does not work as well as other scaling options.
When I am watching 8-bit video (i.e. NV12 colorspace is used), I always get DXVA chroma scaling in the stats instead of SoftCubic or whatever set in settings. This DXVA scaling gives ugly aliasing and moreover it changes matrices 601-709.
I got this
http://grab.by/Rufs
and this
http://grab.by/RufG
In my old PotPlayer everything is OK.
My previous madvr 0.87.10 acts well.
Chyrka is offline   Reply With Quote
Old 15th July 2016, 19:17   #38742  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by Chyrka View Post
Hi.
I have something strange with playback.
My main GPU is Intel HD3000. I also have Nvidia 210 in the PC but it is not connected to the monitor.
I am using MPC-HC x86 1.7.10.
The choma scaling algorithm is set to default SoftCubic but for some reason it does not work as well as other scaling options.
When I am watching 8-bit video (i.e. NV12 colorspace is used), I always get DXVA chroma scaling in the stats instead of SoftCubic or whatever set in settings. This DXVA scaling gives ugly aliasing and moreover it changes matrices 601-709.
I got this
http://grab.by/Rufs
and this
http://grab.by/RufG
In my old PotPlayer everything is OK.
My previous madvr 0.87.10 acts well.
you are using DXVA native. on default settings DXVA native is used with DXVA chroma scaling.

go to rendering -> trade quality for performance - disable "use DXVA chroma upscaling when..."

edit: or use DXVA copyback

Last edited by huhn; 15th July 2016 at 19:28.
huhn is offline   Reply With Quote
Old 15th July 2016, 19:24   #38743  |  Link
Uoppi
Registered User
 
Join Date: Oct 2015
Posts: 99
Rendering queue is sometimes hitting a ceiling of 8 max, even though I have set it at 20. Is this a known issue?

I remember there's been reports along the way about rendering queues not filling but this probably only started happening to me with the latest madvr version (v.22).

If I either
1) stop and restart the video or
2) skip forward (and probably backwards too),
the rendering queue is filled properly to max again.

I have GTX 960 with latest drivers + Kodi DSplayer.
Uoppi is offline   Reply With Quote
Old 15th July 2016, 21:31   #38744  |  Link
Chyrka
Registered User
 
Join Date: Oct 2011
Location: Kyiv, Ukraine
Posts: 24
Quote:
Originally Posted by huhn View Post
you are using DXVA native. on default settings DXVA native is used with DXVA chroma scaling.

go to rendering -> trade quality for performance - disable "use DXVA chroma upscaling when..."

edit: or use DXVA copyback
Thank you for your help. With DXVA copyback works great.
Chyrka is offline   Reply With Quote
Old 16th July 2016, 17:30   #38745  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by zoyd View Post
I do think the desaturation of high luminance red/orange is too aggressive when using preserve hue. Take this for example:

[...]

Without hue preserve the madVR image luminance is too low and appears oversaturated - with hue preserve you lose all of the amber light and a good portion of the red tent.

The CalMAN LUT appears to mildly desaturate and both white and amber headlights look correct. The red tent brightens without loosing too much red definition.
I've managed to reproduce something similar to the Calman result by keeping saturation the same and removing all excess luminance from each pixel. Which is pretty much the opposite of what madVR 0.99.22 does. The next madVR build will compromise by walking middle ground: I'm going to move such out-of-gamut pixels in-gamut by reducing both (perceived) luminance and saturation in a similar amount. It's not directly a dE formula I'm using, but it should be a decent approximation.

Quote:
Originally Posted by zaemon View Post
Dumb question but is it possible to create a 3DLUT for [...] HDR playback?
If you let madVR do the tone mapping ("enable HDR processing") then your SDR 3dlut should work just fine for HDR content, too. No need to create a separate 3dlut in that case.

Quote:
Originally Posted by Sunset1982 View Post
That's exactly what I found out when testing the RX480. Greenscreen when using madVR and LAV with DXVE Copyback. Any ideas why this is happening?
No. If software decoding works fine the greenscreen could be a LAV bug, because it shouldn't make a difference to madVR whether LAV uses software or DXVA copyback decoding. Native DXVA is a different topic, though. Can you upload a zipped debug log with Native DXVA decoding and the black screen somewhere?

Quote:
Originally Posted by DragonQ View Post
What are my options if I want to use MadVR without administrator rights? Presumably there's no way to use it with MPC-HC but are there any other players that include it without having to manually register the filter file?
I think J.River MC uses madVR without installing it. I'm not sure if MC itself might need installation, though.

Quote:
Originally Posted by Uoppi View Post
After updating from v0.90.20 to v0.90.21 (haven't got to try v0.90.22 yet), I started getting worse rendering times but only when scaling.
Can you please post a screenshot of the Ctrl+J OSD for this situation with both v0.90.20 and v0.90.21?

Quote:
Originally Posted by Uoppi View Post
With the .22 version (I can't remember seeing this before), the madvr OSD is reporting Bicubic150 AR downscaling in addition to SSIM 1D that is chosen for the profile.

Like this:
image x > super-xbr > super-xbr < SSIM 1D100AR
image y > super-xbr > super-xbr < SSIM 1D100AR < Bicubic150 AR
Interesting. Doesn't seem to happen on my PC. Again: Can you please post a screenshot of the Ctrl+J OSD, so I can see which exact source resolution and zoom factor etc you're using?

Quote:
Originally Posted by Uoppi View Post
Rendering queue is sometimes hitting a ceiling of 8 max, even though I have set it at 20. Is this a known issue?
Is the present queue near empty? Really, in all these situations a screenshot of the Ctrl+J OSD helps.
madshi is offline   Reply With Quote
Old 16th July 2016, 18:59   #38746  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by ShiftyFella View Post
After resetting my setting and nuking all of my profiles, I figured out the culprit. It was SuperRes turned on for chroma upscaling. When both SuperRes and HDR procesing turned on, it produces weird color shift and it changes depending on the SR level.
Just tried to reproduce this, but it seems to work just fine on my PC. Can you please make a screenshot of your Ctrl+J OSD, when the problem occurs?
madshi is offline   Reply With Quote
Old 16th July 2016, 19:28   #38747  |  Link
har3inger
Registered User
 
Join Date: Feb 2014
Posts: 139
Quote:
Originally Posted by madshi View Post

Three things:

1) Bilateral chroma upscaling is slower now (when using the new algo).
2) HDR handling is slower now.
3) Fixed a couple of bugs in deciding when image doubling and upscaling refinement becomes activated.
I've also noticed big (32ms->45ms) performance drops from .20->.22, but only today. Perhaps a windows update happened in the meantime? I didn't touch my GPU drivers in the meantime, and windows update didn't either. I've reverted for now, because the drop is huge.

What I've tested (not much):

1. At 1:1 scaling factor, render times are the same between the two versions.
2. At any scaling factor, the effect of the chroma scaler shows no differences between the two versions.
3. ED dithering isn't accidentally on. Neither are any of the image enhancement options.
3. Something in the image upscale chain (sxbr > bicubic 150 down > Superres) seems to cost way more with this version.

Last edited by har3inger; 16th July 2016 at 19:33.
har3inger is offline   Reply With Quote
Old 16th July 2016, 19:32   #38748  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by leeperry View Post
OK, here goes:

http://screenshotcomparison.com/comparison/178293

http://screenshotcomparison.com/comparison/178295

http://screenshotcomparison.com/comparison/178298

files on mega

I got a good bunch of other features enabled in the upscaling refinement window, my point was in my real-world use AB75 does wonders. Of course SR looks sharper than without but I believe the top left of the picture in my first link with SR1 looks better with AB25 than without. The whole point of AB to me is that motion looks smoother because edges aren't nearly as noisy anymore. Anyway, please bring it back
I've downloaded both your mega files, and looked at all your screenshot comparisons. To be honest, I'm not fully sure which image was made with which settings. The files are not named with the full settings in the file name. E.g. they should have been named "SR1 no AB" or "SR2 AB75". But instead they have names such as "1.png", "2.png", "3 orignal 4X.png" etc. How am I supposed to understand which SR level and which AB percent those were made with?

Because all the names are not clear, it's hard to draw final conclusions. But e.g. if I compare your "3 SR1 no AB 4X.png" image with "3 AB75 4X.png", then I think the first one (without AB) looks better with less noise around the edges, especially at the bottom left part of the image.
madshi is offline   Reply With Quote
Old 16th July 2016, 19:34   #38749  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by har3inger View Post
I've also noticed big (32ms->45ms) performance drops from .20->.22, but only today. Perhaps a windows update happened in the meantime? I didn't touch my GPU drivers in the meantime, and windows update didn't either. I've reverted for now, because the drop is huge.
So far nobody has provided even screenshots of the Ctrl+J OSD. Pretty sad, like 4 reports now, but nobody bothered to post any screenshots or anything else for me to look at.
madshi is offline   Reply With Quote
Old 16th July 2016, 19:48   #38750  |  Link
jerryleungwh
Registered User
 
Join Date: Jun 2016
Posts: 39
Quote:
Originally Posted by madshi View Post
According to the log madVR *tries* to enter fullscreen exclusive mode, but DXGI/Direct3D reports "DXGI_STATUS_OCCLUDED". Which according to Microsoft means:

> DXGI_STATUS_OCCLUDED if you request full-screen mode
> and it is unavailable

Whatever that means. I don't think there's much I can do about it.
I have managed to solve the fullscreen exclusive mode problem. Turns out the problem is that I set madVR to present 16 frames in advance while in exclusive mode but apparently my GPU doesn't support that. I change it back to any value lower than 16 then it works perfectly. Thanks!
jerryleungwh is offline   Reply With Quote
Old 16th July 2016, 21:08   #38751  |  Link
zoyd
Registered User
 
Join Date: Sep 2009
Posts: 43
Quote:
Originally Posted by madshi View Post
I've managed to reproduce something similar to the Calman result by keeping saturation the same and removing all excess luminance from each pixel. Which is pretty much the opposite of what madVR 0.99.22 does. The next madVR build will compromise by walking middle ground: I'm going to move such out-of-gamut pixels in-gamut by reducing both (perceived) luminance and saturation in a similar amount. It's not directly a dE formula I'm using, but it should be a decent approximation.
That should improve things, will test when ready.

I forgot to ask, what does madVR do with BT.2020 input when you specify

1. "this display is already calibrated" for BT.709 primaries
vs.
2. "this display is already calibrated" for BT.2020 primaries

and is it the same whether HDR processing is on or off?
zoyd is offline   Reply With Quote
Old 16th July 2016, 21:55   #38752  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by madshi View Post
I've downloaded both your mega files, and looked at all your screenshot comparisons. To be honest, I'm not fully sure which image was made with which settings. The files are not named with the full settings in the file name. E.g. they should have been named "SR1 no AB" or "SR2 AB75". But instead they have names such as "1.png", "2.png", "3 orignal 4X.png" etc. How am I supposed to understand which SR level and which AB percent those were made with?

Because all the names are not clear, it's hard to draw final conclusions. But e.g. if I compare your "3 SR1 no AB 4X.png" image with "3 AB75 4X.png", then I think the first one (without AB) looks better with less noise around the edges, especially at the bottom left part of the image.
1 is the full size 1080p picture, 2 the 300x300 isolated bit and 3 the 4X magnified version of the 300x300 bit. I initially used SR2 before you asked for SR1. Here are the fully renamed versions of the 1080p files on mega.

I don't really understand how you could disagree that the top left corner on this picture with SR@1 doesn't look edgier without AB25: http://screenshotcomparison.com/comparison/178293

Ultimately, I guess the previous posts with the guitar make a better point than my pictures as SR without AB in mVR looks way too sharp and edgy to me and I might not be the only one who thinks so for that matter.

I use SSIM 2D 100% for downscale and I was refining SR after every 2X upscale. I fully realize that SR is still a work in progress but I have never found any use to its AR subsetting when OTOH the AB one is vital to me.

But I must admit that Shiandow's result with his own version of SR looks very impressive as in not overly sharp and very natural looking, I'd love to give it a shot if any possible please

Last edited by leeperry; 16th July 2016 at 22:27.
leeperry is offline   Reply With Quote
Old 16th July 2016, 22:01   #38753  |  Link
Iznogūd
Registered User
 
Iznogūd's Avatar
 
Join Date: Sep 2002
Location: Barcelona
Posts: 20
Requested screenshots

Hi, Madshi, my apologies for the delay in posting these:

madVR v09.90.20
madVR v09.90.21

Now, the settings are the same in both cases:

Chroma upscaling: super-xbr 150, AR on, SR 3.
Image downscaling: SSIM 1D 50%, AR on, linear light on, anti-bloating on 100%.
Image doubling: always if needed, super-xbr anti-bloating 100
Image upscaling: Jinc, AR on
Upscaling refinement: sharpen edges 1, thin edges 1, enhance detail 1, anti-bloating 100%, AR on, SR 1 linear light AR on.


I hope this will be useful to you. Please let me know if you need more info from me.
Iznogūd is offline   Reply With Quote
Old 16th July 2016, 23:17   #38754  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by zoyd View Post
I forgot to ask, what does madVR do with BT.2020 input when you specify

1. "this display is already calibrated" for BT.709 primaries
vs.
2. "this display is already calibrated" for BT.2020 primaries

and is it the same whether HDR processing is on or off?
With SDR BT.2020 content and "this display is already calibrated to BT.709", madVR converts the gamut from BT.2020 to BT.709, but colors are simply clipped in the most simple way to BT.709, which is not optimal. With SDR content and "this display is already calibrated to BT.2020", no processing is necessary/done.

With HDR BT.2020 content, it's almost the same. However, the BT.2020 -> BT.709 gamut mapping is performed in much higher quality.

It's on my to do list to perform the higher quality processing also for SDR gamut down-conversions.

Quote:
Originally Posted by leeperry View Post
I don't really understand how you could disagree that the top left corner on this picture with SR@1 doesn't look edgier without AB25: http://screenshotcomparison.com/comparison/178293
Comparing SR1 with or without AB was not the intention of this whole operation. I've asked for SR1 screenshots because my suspicion is that AB for SR does nothing but simply decrease SR strength. If my suspicion is right than the images to compare would be 1) SR2 with AB and 2) SR1 without AB (or with less AB). If both images are nearly the same then that would indicate that AB's only benefit is in reducing SR strength.

So here we go:

SR2 AB75
SR1 no AB

Although the images don't have exactly the same look, they appear reasonably close, with the SR1 image having less noisy edges. Based on this one test image, I'd prefer SR1 over SR2 AB75. Wouldn't you?

Quote:
Originally Posted by leeperry View Post
Ultimately, I guess the previous posts with the guitar make a better point than my pictures as SR without AB in mVR looks way too sharp and edgy to me
Who's to judge which image is better there? We don't have a groundtruth for that one.

Quote:
Originally Posted by leeperry View Post
But I must admit that Shiandow's result with his own version of SR looks very impressive as in not overly sharp and very natural looking, I'd love to give it a shot if any possible please
No. Every time I compare it, it looks worse to my eyes than what madVR produces. But you're welcome to try yourself. If you do, please use an image where you have a ground truth, and be careful that you pick similar algorithms everywhere, so that you really only test the different SuperRes algos and not differences in other algos.

Quote:
Originally Posted by Iznogūd View Post
Hi, Madshi, my apologies for the delay in posting these:

madVR v09.90.20
madVR v09.90.21

Now, the settings are the same in both cases:

Chroma upscaling: super-xbr 150, AR on, SR 3.
Image downscaling: SSIM 1D 50%, AR on, linear light on, anti-bloating on 100%.
Image doubling: always if needed, super-xbr anti-bloating 100
Image upscaling: Jinc, AR on
Upscaling refinement: sharpen edges 1, thin edges 1, enhance detail 1, anti-bloating 100%, AR on, SR 1 linear light AR on.

I hope this will be useful to you. Please let me know if you need more info from me.
Thanks. That's looks like quite a strong difference. Is it possible that with 0.90.20 the upscaling refinement was not applied? Please try unchecking those options and see if it makes a difference ot the render times.

There was a user who complained about upscaling refinement settings not becoming active when upscaling by a relatively small scaling factor. I've changed that, which will of course slow things down if upscaling refinement suddenly becomes active when it didn't before.
madshi is offline   Reply With Quote
Old 16th July 2016, 23:44   #38755  |  Link
zoyd
Registered User
 
Join Date: Sep 2009
Posts: 43
Quote:
Originally Posted by madshi View Post
With SDR BT.2020 content and "this display is already calibrated to BT.709", madVR converts the gamut from BT.2020 to BT.709, but colors are simply clipped in the most simple way to BT.709, which is not optimal. With SDR content and "this display is already calibrated to BT.2020", no processing is necessary/done.

With HDR BT.2020 content, it's almost the same. However, the BT.2020 -> BT.709 gamut mapping is performed in much higher quality.
Thanks! So if I use instead "calibrate with 3dlut" and load a BT.2020->BT.709 LUT in the BT.2020 slot and enable HDR processing, madVR will use the LUT for gamut mapping and HDR algos for tone mapping?
zoyd is offline   Reply With Quote
Old 16th July 2016, 23:59   #38756  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by madshi View Post
Comparing SR1 with or without AB was not the intention of this whole operation. I've asked for SR1 screenshots because my suspicion is that AB for SR does nothing but simply decrease SR strength. If my suspicion is right than the images to compare would be 1) SR2 with AB and 2) SR1 without AB (or with less AB). If both images are nearly the same then that would indicate that AB's only benefit is in reducing SR strength.

So here we go:

SR2 AB75
SR1 no AB

Although the images don't have exactly the same look, they appear reasonably close, with the SR1 image having less noisy edges. Based on this one test image, I'd prefer SR1 over SR2 AB75. Wouldn't you?
I understood your point that AB might just decrease SR's strength but my point was to compare SR1 with/without AB and AB does make the top left edge smoother and less noisy.

Well, screenshots are one thing, how AB fares in combination with FRC in 60Hz motion another, basically SR1 doesn't look sharp enough and SR2 too sharp. AB gives SR an analog look and does exactly what you initially said: "The new anti-bloating tries to concentrate sharpening on higher frequencies and to remove lower frequencies. I like the look that produces, but your mileage may vary".

I -and others- like the look that it produces, we like how it sharpens higher frequencies and removes lower ones

BUT if you really wanna compare SR1 and SR2 AB75, only 4X zooms make sense to me as I run a 32" TV from a 1 meter distance: http://screenshotcomparison.com/comparison/178799

Apart from the top left corner I far prefer the SR2 AB75, it even makes the shadow in the bottom left corner lighter & closer to the ground truth, very obviously less bloated™:

http://screenshotcomparison.com/comparison/178801

http://screenshotcomparison.com/comparison/178802

Quote:
Originally Posted by madshi View Post
Who's to judge which image is better there? We don't have a groundtruth for that one.
Oh so now a ground truth has become mandatory? So we went from "I like the look that produces, but your mileage may vary" to "got ground truth?"

Maybe the OP would be able to provide it, but it seems rather hard to deny that mVR's SR looks way oversharpened in his example IMHO. I'm certainly not interested in such a sharp image myself, and neither would be others as it would appear.

Quote:
Originally Posted by madshi View Post
No. Every time I compare it, it looks worse to my eyes than what madVR produces. But you're welcome to try yourself.
Apparently the only way to try it would be that MPDN thingie but I've never used and I have no interest in learning as it seems tied to MPC.

Last edited by leeperry; 17th July 2016 at 02:38.
leeperry is offline   Reply With Quote
Old 17th July 2016, 00:06   #38757  |  Link
pose
Registered User
 
Join Date: Jul 2014
Posts: 69
Any RX 480 users here? Id like to compare my rendering times on 7870 to the new card! Mine could do 1280x720 24FPS video to 1080p with luma doubling 64 NNEDI3, smooth motion, error diffusion dithering and low/high debanding with around 35ms rendering time. Trade quality for performance - all unchecked. Win 7 Catalyst 13.12 driver.
pose is offline   Reply With Quote
Old 17th July 2016, 00:15   #38758  |  Link
ShiftyFella
Registered User
 
Join Date: Apr 2016
Posts: 17
Quote:
Originally Posted by madshi View Post
Just tried to reproduce this, but it seems to work just fine on my PC. Can you please make a screenshot of your Ctrl+J OSD, when the problem occurs?
Ok, here it is:

SuperRes turned on:
http://i.imgur.com/3khJ82K.png
http://i.imgur.com/p6JGGyZ.png
SuperRes turned off:
http://i.imgur.com/57jtmWP.png
http://i.imgur.com/IwiMxG6.png

heck, even did a gif switching sr back and forth to show changes:
https://gfycat.com/FloweryIdolizedCaterpillar

In HDR processing everything turned on, preserve hue is on high and peak luminance is at 400 nits. Having different chroma upscaling, image downscaling, doubling, image upscaling algos has no effect on image color shift including turning on or off superres in upscaling refinement, only turning off superres for chroma gets rid of color shit.
ShiftyFella is offline   Reply With Quote
Old 17th July 2016, 01:14   #38759  |  Link
QBhd
QB the Slayer
 
QBhd's Avatar
 
Join Date: Feb 2011
Location: Toronto
Posts: 697
Sorry for the no screen shots madshi... but as Iznogūd posted, the screen shots wouldn't have made much of a difference since they are the same except for the performance hit... I plan on doing some testing of the upscaling refinements myself but work has not let me... and I am about to leave soon for my first Saturday night out in what seems like years! LOL... Hopefully tomorrow, I will have plenty of time to test this performance hit. I'm just glad I'm not the only one who sees it. Although I think I'm the first to report it since I always push the GPU to the max on my 720p content (which is what I mostly watch) and the slightest performance hit will always push my GPU over the edge.

QB

EDIT:

p.s. Is there anyway to create a debug OSD that will show every setting being used? Even if it covers the entire screen... something we could use to narrow things down for you, and provide more detailed SS's. It could also be useful for hardcore users to tweak their own settings
__________________

Last edited by QBhd; 17th July 2016 at 01:20.
QBhd is offline   Reply With Quote
Old 17th July 2016, 02:53   #38760  |  Link
har3inger
Registered User
 
Join Date: Feb 2014
Posts: 139
No screenshots for the OSD for now, but literally everything in it is the same, except render time (35ish -> 45ish) , and the render queue emptying slowly over time.

I have no upscale refinement algo's checked except SR (which is definitely on in both cases), so it shouldn't be that, unless one became forced on regardless of the UI setting.
har3inger is offline   Reply With Quote
Reply

Tags
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 00:21.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.