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 17th February 2014, 11:38   #23341  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by James Freeman View Post
Subjective contrast can mean Enhanced Contrast and that's not a good thing for video purists or what MadVR is trying to achieve.
I think we should go for accuracy instead of subjective perception.

What do you think?
Accuracy as compared to what? The more it goes the more ED's come with a lower noise floor and consequently an improved sense of depth as dynamics are increased. Don't shoot the messenger, A2/A3 look foggy to me as moving objects seem too blended-in and edges aren't clearly defined anymore. I'll even go for RD over A2/A3, edges are far more discernable than in that big syrupy noise feast.

Please be so kind as to define "video purist", I'm new to this hobby.

Quote:
Originally Posted by James Freeman View Post
Hmm....without enhancements?
I quit doing weed ages ago if that's what you meant, it gave me the munches

Quote:
Originally Posted by James Freeman View Post
You seem to be the only one...
Mind you, many ppl see have previously seen differences between those algorithms, but apparently only one person bothered sitting down and scrutinize changes between the A builds so far:
Quote:
Originally Posted by mithra66 View Post
I prefer A1 > A3 >= A2.
Quote:
Originally Posted by James Freeman View Post
On what hardware do you compare the builds?
1GHz HD7850, W7SP1, NNEDI 64x/ED all the way, dead-on REC709 gamut mapped 3500:1 32" A-MVA panel LED TV with BFI turned on from a 80 cm distance in a pitch black room.
leeperry is offline   Reply With Quote
Old 17th February 2014, 11:53   #23342  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
Quote:
Originally Posted by leeperry View Post
Please be so kind as to define "video purist", I'm new to this hobby.
Audiophiles, Videophiles...

What comes in, comes out, as the mastering engineers intended.
Properly calibrated equipment (to a standard) without enhancements.
Quote:
I quit doing weed ages ago if that's what you meant, it gave me the munches
I never did.
Quote:
Mind you, many ppl see have previously seen differences between those algorithms, but apparently only one person bothered sitting down and scrutinize changes between the A builds so far:
What am I not a factor? I gave my best shot.
I voted A2 after a comprehensive test.
Quote:
1GHz HD7850, W7SP1, NNEDI 64x/ED all the way, dead-on REC709 gamut mapped 3500:1 32" A-MVA panel LED TV with BFI turned on from a 80 cm distance in a pitch black room.
Looks good.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 11:53   #23343  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by James Freeman View Post
Comparison of all the NL1-8 and AD1-4, Grey & Color: ED Builds Test.zip
Thank you.

Color 250, AD4 wins by a mile, that center bit just get fixed man.
The graduation is much better than the rest IMO. I'm all over AD4.
Only reason I could possibly see passing it up is if it's brighter than it should be..

Quote:
Originally Posted by leeperry View Post
FWIW I can DBT them anytime you want.
Anyone wanna put together a pot for this? XD

Mean signature. "I'll DBT anytime you want"

Last edited by ryrynz; 17th February 2014 at 11:56.
ryrynz is offline   Reply With Quote
Old 17th February 2014, 12:02   #23344  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
Quote:
Originally Posted by ryrynz View Post
Thank you.

Color 250, AD4 wins by a mile, that center bit just get fixed man.
The graduation is much better than the rest IMO. I'm all over AD4.
Only reason I could possibly see passing it up is if it's brighter than it should be..

The 250 Color undergone a slightly different procedure than 88 & 150, to see the difference clearer.

Yes, A4 is better than A1.
I wouldn't mind if A4 is selected as the final build, but NOT A1.
I still like A2 the most, but A4 will also do.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.

Last edited by James Freeman; 17th February 2014 at 15:03.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 12:07   #23345  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by leeperry View Post
1GHz HD7850, W7SP1, NNEDI 64x/ED all the way, dead-on REC709 gamut mapped 3500:1 32" A-MVA panel LED TV with BFI turned on from a 80 cm distance in a pitch black room.
on a 4:2:2 display yeah sounds reasonable.
huhn is offline   Reply With Quote
Old 17th February 2014, 12:19   #23346  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
@Ver Greeneyes

Can you please make an, 8-bit yuv444p video of Color & Grey, to compare the dithered 16-bit vs undithered 8-bit at the same frame?
True 8-bit video should not be dithered at all, right?
The purpose of this is to test the Gamma or any other deviations between the builds.

Thanks.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.

Last edited by James Freeman; 17th February 2014 at 13:18.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 12:28   #23347  |  Link
fairchild
Registered User
 
Join Date: Sep 2010
Posts: 321
Quote:
Originally Posted by Asmodian View Post
IMO dithering should never be disabled in madVR, especially if you can use the great OpenCL (really DirectCompute now) error diffusion dithering.
Not entirely true. In my case any added dithering through MadVR raises my black level floor slightly where it doesn't match with the Brightness setting which was calibrated to through a different device such as a pattern generator and/or blu-ray player. So if someone out there has a plasma or any other device that generates it's video through dithering in the panel itself, and they plan to calibrate through MadVR for say an HTPC, then it may be beneficial to disable dithering entirely so that all sources in the display chain match. (OTA TV broadcasts, cable/dish broadcasts, blu-ray player, gaming systems, etc...)

But whenever you are going to watch any video content through MadVR, then of course enable dithering as the banding that is produced without it can be visible in some content. I have tried to tell the difference between all the new ED builds, but frankly I guess my eyes aren't keen enough to see the difference. I'm currently using EDAD4 and have seen no problems with anything I've watched through it. Same goes for all the previous builds which I have tried with regular content.
__________________
MPC-HC/MPC-BE, Lav Filters, MadVR
CPU: AMD Ryzen 5 1600, Video: AMD Radeon RX Vega 56 -> TCL S405 55", Audio: Audio-Technica M50S
fairchild is offline   Reply With Quote
Old 17th February 2014, 13:04   #23348  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
The clearest Comparison: Color 110 AD1-4.zip

EDIT:
The last comparison had an error, this zip (fixed) should accurately reflect whats going on.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.

Last edited by James Freeman; 17th February 2014 at 14:58.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 14:04   #23349  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by James Freeman View Post
What am I not a factor? I gave my best shot.
I voted A2 after a comprehensive test.
My bad, I thought you were only nitpicking about screenshots.

I should have also stated that I run Reclock & have "low" debanding forced in mVR.

Quote:
Originally Posted by James Freeman View Post
I wouldn't mind if A4 is selected as the final build, but NOT A1.
A1 is obsolete compared to A4, both technically and visually IMHO.

Anyway, this quote might save the day:
Quote:
Originally Posted by madshi View Post
Whether A3 or A4 is better might be a matter of taste. I think A3 has a slightly more random look. Which could be positive or negative, depending on taste.
madshi will not try to convince either of us that strawberry tastes better than banana, maybe A4 + A2(or 3) might be the final ED's

Quote:
Originally Posted by ryrynz View Post
Anyone wanna put together a pot for this? XD

Mean signature. "I'll DBT anytime you want"
It's rather exhausting to compare them as it takes a lot of concentration but make it interesting and we got a deal

Quote:
Originally Posted by huhn View Post
on a 4:2:2 display yeah sounds reasonable.
Yep, chroma is such a big deal when it comes to sharpness. I'll take 4:2:2 BFI over blurry 60Hz-only 4:4:4 LCD anytime of the day. Especially as the TV runs a 10/12bit video engine and I feed it J3AR chroma to begin with. Necessary evil - perfect world, you know the drill.

Last edited by leeperry; 17th February 2014 at 14:06.
leeperry is offline   Reply With Quote
Old 17th February 2014, 14:09   #23350  |  Link
cyberbeing
Broadband Junkie
 
Join Date: Oct 2005
Posts: 1,859
Quote:
Originally Posted by James Freeman View Post
Using my special method.
...
A1 & A4 are a tad brighter than A2 & A3.
That isn't just a tad brighter, but rather multiple levels brighter. I'd question your special brightening method for causing this discrepancy. You must have brightened A1 & A4 more than A2 & A3 because of noise level differences. In untouched form, all these adaptive routines should have identical overall brightness with 1/100th of a level step accuracy.
cyberbeing is offline   Reply With Quote
Old 17th February 2014, 14:19   #23351  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
Quote:
Originally Posted by cyberbeing View Post
That isn't just a tad brighter, but rather multiple levels brighter. I'd question your special brightening method for causing this discrepancy. You must have brightened A1 & A4 more than A2 & A3 because of noise level differences. In untouched form, all these adaptive routines should have identical overall brightness with 1/100th of a level step accuracy.
EDIT:

You are correct.
There is one step error with the last zip file, I'll fix that right away!

Fixed: This post.

Instead of changing the Input Levels, I just move the middle slider of the Output Levels closer to 255 to give that result.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.

Last edited by James Freeman; 17th February 2014 at 15:04.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 15:07   #23352  |  Link
Ver Greeneyes
Registered User
 
Join Date: May 2012
Posts: 447
Quote:
Originally Posted by James Freeman View Post
@Ver Greeneyes

Can you please make an, 8-bit yuv444p video of Color & Grey, to compare the dithered 16-bit vs undithered 8-bit at the same frame?
True 8-bit video should not be dithered at all, right?
The purpose of this is to test the Gamma or any other deviations between the builds.
Okay, link in my signature.

I still need to update the colored version.

Edit: 8-bit version re-encoded in "bgr0" color space (should be the same result as rgb24), and updated the colored versions.

Last edited by Ver Greeneyes; 17th February 2014 at 15:56.
Ver Greeneyes is offline   Reply With Quote
Old 17th February 2014, 16:07   #23353  |  Link
NicolasRobidoux
Nicolas Robidoux
 
NicolasRobidoux's Avatar
 
Join Date: Mar 2011
Location: Montreal Canada
Posts: 269
Quote:
Originally Posted by 6233638 View Post
Honestly, pixel art does not look good when you use video-style filters on it. It's just not how it was intended to look...
Yes: What I pointed to was a general purpose image resampling method... that does reasonably well with pixel art.
Really (and I believe madshi has pointed that out already, and you are basically reiterating here) pixel art calls for specialized methods.
This being said, if you are willing to use NNEDI on pixel art, you may as well give a try to a tuned Jinc.
NicolasRobidoux is offline   Reply With Quote
Old 17th February 2014, 16:08   #23354  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
Quote:
Originally Posted by Ver Greeneyes View Post
Okay, link in my signature.

I still need to update the colored version.

Edit: 8-bit version re-encoded in "bgr0" color space (should be the same result as rgb24), and updated the colored versions.
Thank you very much.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.
James Freeman is offline   Reply With Quote
Old 17th February 2014, 17:33   #23355  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by James Freeman View Post
@Ver Greeneyes

Can you please make an, 8-bit yuv444p video of Color & Grey, to compare the dithered 16-bit vs undithered 8-bit at the same frame?
True 8-bit video should not be dithered at all, right?
The purpose of this is to test the Gamma or any other deviations between the builds.

Thanks.
This can be confusing at first, but even 8bit needs to be dithered, because madVR upsamples at the very beginning and the dithering we are testing atm is the dithering that is applied at the very end, before madVR sends it to the display itself.

Of course it is very valid to test with 8bit sources though, because thatīs what most of the current content comes in. Thatīs why I asked some pages ago if someone had access to some real world high-bitdepth footage (like TimeScapes), since we could (I know leeperry would be able to, his display is also well equipped, because he has a way higher dynamic range to begin with) more effectively see the differences, because our eyes are trained to perceive nature and compare it at the same time with reality, instead of artificial patterns.

Last edited by iSunrise; 17th February 2014 at 17:39.
iSunrise is offline   Reply With Quote
Old 17th February 2014, 18:15   #23356  |  Link
Ver Greeneyes
Registered User
 
Join Date: May 2012
Posts: 447
Quote:
Originally Posted by iSunrise View Post
This can be confusing at first, but even 8bit needs to be dithered, because madVR upsamples at the very beginning and the dithering we are testing atm is the dithering that is applied at the very end, before madVR sends it to the display itself.
There shouldn't need to be any upsampling if you test this at 100% zoom on a monitor that's big enough (that's one reason I used 1920 for the width - I could make the height smaller for 1920x1080 monitors though, so you can open it in a maximized window instead of being forced to use full screen). The 8-bit video is encoded with 24-bit RGB, so there's no need for conversion, and the 16-bit video is encoded with YUV444, which also doesn't require chroma upscaling (but does require conversion into RGB). Of course, any other processing that madVR applies (such as a 3DLUT) is going to require dithering.
Ver Greeneyes is offline   Reply With Quote
Old 17th February 2014, 18:24   #23357  |  Link
MistahBonzai
Registered User
 
Join Date: Mar 2013
Posts: 101
Quote:
Originally Posted by James Freeman View Post
MistahBonzai,

Everything is fine, ED is changing.
The difference is so refined between the latest builds, that its almost invisible with the naked eye (without software intervention)..
Thanks..it's just that leeperry seems to be able to make subjective impressions quite easily. I have an accurately calibrated 10bit 40" display which serves as my desktop monitor, along with a practiced eye, and figured I could define differences by viewing actual content in a normal viewing environment. Apparently not..and lord knows I tried...
MistahBonzai is offline   Reply With Quote
Old 17th February 2014, 18:32   #23358  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by MistahBonzai View Post
I have an accurately calibrated 10bit 40" display which serves as my desktop monitor, along with a practiced eye, and figured I could define differences by viewing actual content in a normal viewing environment. Apparently not..and lord knows I tried...
When you compare NL6 and A4, don't you see the major decrease in noise? Are you running Reclock in 24/60Hz?

What's the calibrated contrast figure on that monitor? Because I guess things would get tough with IPS/TN and a thick antiglare pearly/grainy layer, possibly with PWM on top of it.

Also, are you wearing glasses? Are they made of organic glass? Mineral glass is the way to go when you need optical resolution, nothing matches their constringence AFAIK. Ages ago I bought some super light polycarbonate glasses but they only offered like 20 Abbe, I was sitting 3 meters away from a 2m large projection screen and it was so frigging blurry......I finally went for 59 Abbe mineral glass et voilā ))
leeperry is offline   Reply With Quote
Old 17th February 2014, 19:15   #23359  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by Ver Greeneyes View Post
There shouldn't need to be any upsampling if you test this at 100% zoom on a monitor that's big enough (that's one reason I used 1920 for the width - I could make the height smaller for 1920x1080 monitors though, so you can open it in a maximized window instead of being forced to use full screen). The 8-bit video is encoded with 24-bit RGB, so there's no need for conversion, and the 16-bit video is encoded with YUV444, which also doesn't require chroma upscaling (but does require conversion into RGB). Of course, any other processing that madVR applies (such as a 3DLUT) is going to require dithering.
My understanding is that madshi implemented the "donīt use dithering" option in the "trade quality for performance" tab, because otherwise, there would always be dithering, even if you wouldnīt need upsampling/scaling at all. Is that a wrong assumption from my side? Otherwise why would he implement the "donīt use dithering" option? Only for lower bit-depth sources?

Also, it would require to always select "8bit (or higher)" under the "devices - properties" tab.

Quote:
Originally Posted by leeperry View Post
When you compare NL6 and A4, don't you see the major decrease in noise? Are you running Reclock in 24/60Hz?
You mean in actually movie watching or with our level-reduced dithering comparisons? Because it depends where the noise is and what it is. Is it "hidden" encoding-noise (could mean bad calibration) or is it the cameraīs source-noise and if itīs the camera source-noise, does that just get more visible or does it get less visible? Could also be artificially added noise, which got added in post, to make the image appear more lifelike (or an artistic choice, in the case of Michael Mann movies). Thatīs where the difficulties begin.

Do you have a sample where you see this noise, because Iīm interested to compare it on my hardware-calibrated Eizo (that only has about 700:1 left after calibration to REC709).

Now that we have A4, I can see a lot more noise in general. And it think thatīs a good thing, because that doesnīt seem to be the cause of the new dithering-algorithms at all, but it finally shows us a higher effective bit-depth of the original source, whereas the smearing/extreme noise of random dithering made that less visible. With random dithering, there was just so much noise all around, it added a whole layer of unnecessary thick noise, that even was colored noise, so that there was no breathing room for fine details at all. Now, with the newer algorithms, actual source noise has more room to breathe. And especially the blacks became a lot more refined and differentiation seems to have improved. At least for higher dynamic range display, this will probably make a world of a difference and I am sure thatīs why some also see differences than others.

Iīm still not entirely sure about the differences between the adaptive variants on my display, it takes so much concentration in actual movies/clips that itīs very time consuming and cannot be done at day times, because my eyes (all senses) work better at night time (more sensitive). All I can say is that, with A4, we seem to have arrived at a level that satisfies me a lot. Even though Iīm not sure what 6233638 and cyberbeing think about it and I want to hear their opinions.

Last edited by iSunrise; 17th February 2014 at 20:13.
iSunrise is offline   Reply With Quote
Old 17th February 2014, 19:57   #23360  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by iSunrise View Post
My understanding is that madshi implemented the "donīt use dithering" option in the "trade quality for performance" tab, because otherwise, there would always be dithering, even if you wouldnīt need upsampling/scaling at all. Is that a wrong assumption from my side? Otherwise why would he implement the "donīt use dithering" option? Only for lower bit-depth sources?
FWIW if all pixel values are whole numbers between 0 and 255 then even if you perform dithering this won't change the output. So in a sense not performing any dithering doesn't ever make the result better. Maybe we should check that the way error diffusion is currently implemented also doesn't change the output, but at the very least the limited builds should be 'safe' since they always output the value rounded down or rounded up which leaves whole numbers unchanged.
Shiandow 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 12:32.


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