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. |
11th December 2016, 04:05 | #41361 | Link | |
Registered User
Join Date: Apr 2002
Location: Germany
Posts: 4,926
|
Quote:
Neo-XP got some very nice results already even @ NGU-Med which should save some processing and he used much more inferior source Lower Resolution 2nd Generation Blu-Ray Transcodes instead of a 1st Generation 1080p H.264 Source This is where Profiles actually could really become handy todo source based setup decisions in a fast switching way Nice would be these Presets then would become easy accessible through the Player Interface itself (not the MadVR Icon only) Your result even from the Screen Capture looks nice especially with Samsungs Cinema Magic Bright PP (which unfortunately has its own sharpening filter ontop of it's Cinema color tone mapping which though doesn't create that much over sharpening based on your Nikons camera focus and jpg result at all).
__________________
all my compares are riddles so please try to decipher them yourselves :) It is about Time Join the Revolution NOW before it is to Late ! http://forum.doom9.org/showthread.php?t=168004 Last edited by CruNcher; 11th December 2016 at 05:01. |
|
11th December 2016, 06:18 | #41362 | Link | |
Registered User
Join Date: Aug 2006
Posts: 2,229
|
Quote:
Now there are two schools of thought regarding GPU capability. he first is the max out all the settings including NGU etc, so when using a side by side comparison with your eyes 2 inches from a 55 inch screen you may be able to tell a slight quality difference between those settings and lower settings, in the right scene. The second is to use more realistic settings and let the GPU peacefully tick over, but without compromising perceptive quality. The last point is important! Basically there is no point (in my opinion) to use higher settings when there is no perceptible difference (and not psychological) between actually watching the video as you would (not doing zoomed in analysis or looking at still images closely), especially when you may be using 3 times the GPU power to do so. Download the latest GPU-z from here: https://www.techpowerup.com/downloads/SysInfo/GPU-Z/ Once running (you don't need to install, just select 'No' or 'Not Now'), click on the sensors tab. This will show you the GPU clocks, temperature, load, and GPU power draw (on the newer cards). Assuming you have dual screens, play back on the screen you wish to play back on, and have GPU-z on the second screen. Watch the GPU use and adjust settings as necessary. I use NGU Medium (or high) for Luma, Jinc AR for upscaling after doubling (only used if requried), and for downscaling 'use "image downscaling" settings'. For those SSIM is the sharpest but introduces aliasing, 2D 100% is the sharpest but the most GPU unfriendly. I think a good compromise is Jinc with linear light and anti-ringing enabled. For Chroma upscaling having Jinc with anti-ringing and superres (1 or 2) is good. With upscaling refinement I have 'add grain', and for dithering I have 'Random dithering' enabled. Error Diffusion may be 'better', but have others have pointed out unnecessary when outputting to a 8 or 10 bit display. So, with the latest drivers, try: Devices (set to highest display depth the display can handle) Artifact removal: Debanding Low/high or medium/high, reduce ringing artifacts Image enhancements: thin edges (1.0), enhance detail (1.0), adaptivesharpen (0.3, for example), anti-rining. Chroma upscaling: Jinc, anti-ringing, SuperRes (1 or 2) Image downscaling: Jinc, linear ligh, anti-rining relaxed Image upscaling: NGU medium (or high IF you can actually see the diffence when watching a video), after doubling algorithm Jinc AR, use image downscaling settings for downscaling. Upscaling refinement: add grain (maybe 1 or 2 depending on your taste). A tiny bit of grain added can perceptively bring out detaisl whilst hiding artifacts. Rendering. General settings. Delay playback until queue is full, overlay mode (not sure if that does anything in Win 10), Fullscreen exclsuive mode, separate device both enabled. Large CPU and GPU queue size seems to work fine (say, 64 and 8 or whatever). The windowed and exclusive mode settings are fine as default (present 8 in advance etc), random dithering, and no trade quality for performance settings. Now on drivers before 16.12.1 these settings wouldn't work, mainly because of DX 11 Exclusive. It would also not be smooth etc. On 16.12.1 with these settings the GPU just ticks over nicely. You will see the GPU only power draw being low as a result of the GPU clock remaining low, which means the card runs cool, and remains under 60C (for me). It may be so that the fans don't even have to run! If you need to zoom in to see if something is better, use a ridiculously low resolution (320x240 for example) and blow it up to 4k resolution on a 60 inch screen and then look closely at a still image to see the difference, or any variation of not actually watching it normally, to be able to tell the difference, then the difference is not worth quibbling over if the settings differences require much more processing power. Back to the new driver, I can even have the computer turning off the display etc when in paused windows mode, and now it will just wake up and play perfectly again! . Actually, all faults I reported earlier seem to be fixed, so it probably was a driver issue. Just strange Nvidia users were affected as well, it's probably related to updated API support (such as SM 6.0, DirectX 11.4, WDDM 2.1 etc and other changes) in the driver no being perfected, particularly since for both Nvidia and AMD you need the new cards (GTX 1000 series/RX 400 series) to make use of it. When starting playback after monitor turn off/resume you do need to go back to full screen mode, or windowed mode then fullscreen if you paused in fullscreen, to reset the output. CLSID mentioned this earlier, it's not a driver or madVR thing, it's just a thing with Windows. There haven't been any associated issues with this (such as MPC-HC freezing etc) with the new drivers. Last edited by burfadel; 11th December 2016 at 06:24. |
|
11th December 2016, 06:43 | #41363 | Link | |
Registered User
Join Date: Mar 2009
Posts: 3,646
|
Claiming that these time spent and provided to end-user options are next to pointless is the height of rudeness.
Well if it's not for luma then it's "pointless and amounts to nothing/placebo" apparently. Saying that he can't see any difference on his outdated setup is all he needed to do. Not the following drivel.. Quote:
*facepalm* I'm done. Last edited by ryrynz; 11th December 2016 at 06:47. |
|
11th December 2016, 07:04 | #41364 | Link |
Registered User
Join Date: Aug 2006
Posts: 2,229
|
The argument is between differences in still images where you have to look closely to tell the difference, or zoomed in output, or low resolution images upsampled, versus actually watching a video and being able to tell the difference. You seem to be in the former category, I'm in the latter. By the sounds of it, Madshi is also in the latter category. If the differences when playing a video and sitting back and not 10 cm from the screen are not perceptible, then putting an extra 100 percent load+ on the GPU for high end Chroma or Luma settings is pointless. Sorry, that's reality. If you can achieve those fine improvements that are only visible when zooming/looking closely 10 cm from the screen on still image without sacrificing GPU performance (or extremely little) then the argument gains for validity.
Spend more time using it for playing videos rather than nitpicking quality settings that have no perceptible difference when ACTUALLY USING it to watch videos normally, then complaining using those settings puts too much load on the GPU. Duh! lols. |
11th December 2016, 07:19 | #41365 | Link |
X Cinema Projectionist NZ
Join Date: Feb 2006
Location: Auckland NZ
Posts: 310
|
Guys I wonder if someone can offer some help please...
I have thee 1080 card and the JVC X9000 projector. Ive always had horrible banding and none of the banding settings on madvr make any difference at all for me. Im still only using 1080p for all my ripped BDs and NEEDI3, haven't tested NGU yet. As I play everything at 60Hz RGB 8bit to avoid the horrible slow sync of the JVC for mixed content, is it possible that 60Hz introduces the banding? Any help would be greatly appreciated. |
11th December 2016, 09:06 | #41366 | Link | |
Registered User
Join Date: May 2015
Posts: 106
|
Quote:
However the question is what's actually better? People tend to prefer sharper and brighter images but that doesn't mean it's closer to intended reproduction. And with the tone changes caused be different chroma algorithms I'm unsure how much sense stuff like color calibration even makes. In other words I think that the differences are so big that we'll never come close to correct (whatever that may be). So we're very much back to personal taste. |
|
11th December 2016, 09:55 | #41367 | Link |
Registered User
Join Date: Mar 2016
Location: Switzerland
Posts: 140
|
I have just noticed a weird behavior with the dark halos removal :
http://screenshotcomparison.com/comparison/193397 The processed lines are not straight anymore as it creates artifacts all around them. It is very distracting, even more in motion. Is there room for improvement there @madshi ? Maybe a less aggressive version ? There are some artifacts too with just the ringing removal, but they are less noticeable : http://screenshotcomparison.com/comparison/193398 Last edited by Neo-XP; 11th December 2016 at 10:01. |
11th December 2016, 10:09 | #41369 | Link | |
Registered User
Join Date: Jan 2014
Posts: 216
|
Quote:
It is not worth it to me to use 2-3x more gpu so that a couple blocks of red are slightly sharper on some scenes, and nothing that I would be able to see in motion from a 6-7' seating distance. I don't know why you can't get that through your thick skull. I will probably switch to superxbr 150ar or NGU low, since render times are only 2-4 ms higher than bicubic 60ar even if it's just placebo in motion. You have a horrible attitude problem, and are just trolling at this point. Take a step back from the keyboard. It will be ok, I promise. |
|
11th December 2016, 10:20 | #41370 | Link |
ゼカマッハ
Join Date: Dec 2016
Posts: 4
|
So I've noticed an odd behavior in MadVR. It was there in 0.89, and it's still there in 0.91.4. For files that are encoded (perhaps incorrectly?) with RGB color space, MadVR turns the whole video a deep magenta tint. Switching MPC-HC to EVR Custom Presenter (while changing no other settings) resolves the issue, as does playing the files with VLC, so it does appear to be a MadVR issue. I've been poring over options in MadVR but I can't find anything that fixes it. Is this a known issue? I couldn't find anyone mention it with the search function, or on Google for that matter.
Here's the mediainfo output for one of the affected files: http://pastebin.com/21bp4Bct System Specs: MPC-HC 1.7.9 madVR 0.91.4 LAVFilters 0.66.0 Windows 7 Ultimate SP1 x64 i7-4790K GeForce 760GTX |
11th December 2016, 10:24 | #41371 | Link | |
Registered User
Join Date: Sep 2016
Posts: 176
|
Quote:
Important: Turn off the dynamic contrast in de nvidia drivers. This causes banding! Use PC 0-255 full color space in driver setting. Set rgb 4:2:2 (not 4:4:4) in driver. Leave X5000 on auto color space. I leave most options in the X5000 as default except for blur reduction setting which I have turned off. They introduce artifacts. I play everything at 60 Hz. Even 24hz movies. This works best. I hope this helps. |
|
11th December 2016, 10:28 | #41372 | Link |
Registered User
Join Date: Aug 2006
Posts: 2,229
|
You LAVFilters is a really old version! 0.68.1-53 is the latest interim build (from 10/12/2016). Also your MPC-HC is out of date (1.7.10.269 is the latest). In LAVFilters you can select to output as whatever you like. If the input was RGB, you can force output in YV12, RGB32 or whatever else and see if that helps. If you go MPC-HC BE, do note that it has a different numbering system, such that 1.5.0 is the latest and is 'newer' than the MPC-HC build.
Are you using MPC-HC 32-bit or 64-bit? |
11th December 2016, 10:35 | #41373 | Link | |
ゼカマッハ
Join Date: Dec 2016
Posts: 4
|
Quote:
Also, I -did- try changing the output settings in LAV, but that did not affect the horrible magenta color output by madVR. For the sake of argument, however, I'll go ahead and update my MPC-HC and LAV. I highly doubt it will change anything, but better safe than sorry, right? EDIT: Am now on MPC-HC 1.7.10x86 and LAVFilters 0.68.1 Stable. No change, as expected. Last edited by nomakewan; 11th December 2016 at 10:44. Reason: Added information |
|
11th December 2016, 10:57 | #41374 | Link | |
Registered User
Join Date: Sep 2016
Posts: 176
|
Quote:
|
|
11th December 2016, 11:05 | #41375 | Link |
ゼカマッハ
Join Date: Dec 2016
Posts: 4
|
My settings are correct (E248WFP; PC levels, 8-bit, no 3D, no calibration, passthrough HDR) for my monitor and had never been an issue until I ran into this odd files that had RGB color space. Every single other file I have ever played plays through perfectly fine. So it's not exactly killing me--I can just play these few files in VLC if I have to--but I figured it was worth reporting since it appears to be a legitimate bug in how madVR is handling them.
|
11th December 2016, 11:30 | #41376 | Link | |
Registered Developer
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,344
|
Quote:
With "no calibration" all videos are passed to the screen without changes, which would result in wrong output on anything that doesn't natively match your screen. Otherwise, madshi will likely need a sample video.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders Last edited by nevcairiel; 11th December 2016 at 11:33. |
|
11th December 2016, 11:37 | #41377 | Link |
Registered User
Join Date: Dec 2014
Posts: 666
|
Guys,
Did somebody posted a test image for Nedi3 64 neuron vs NGU all variants? I don't know if it was just my eyes. Nedi3 64 produces cleaner and more natural video. It manages noise very well as depicted in video such as TWD. TWD is very grainy and Nedi3 doesn't sharpen this. Can anybody concur? The only downside is that it hugs too much resources but if you have something to spare then this algo seems to be the best
__________________
Asus ProArt Z790 - 13th Gen Intel i9 - RTX 3080 - DDR5 64GB Predator - LG OLED C9 - Yamaha A3030 - Windows 11 x64 - PotPlayerr - Lav - MadVR |
11th December 2016, 11:49 | #41378 | Link | |
Registered User
Join Date: Feb 2012
Posts: 54
|
Quote:
And how much rise up the quality of the image? |
|
11th December 2016, 12:36 | #41379 | Link | |
Registered User
Join Date: Jul 2010
Posts: 22
|
Quote:
|
|
11th December 2016, 13:03 | #41380 | Link | |
ゼカマッハ
Join Date: Dec 2016
Posts: 4
|
Quote:
I have samples I can provide if truly necessary, that is, if the mediainfo output alone is not sufficient to pinpoint the issue at play. As this is my first time posting on Doom9 (I had an account years ago but it's apparently been pruned), what method is generally accepted for submitting samples nowadays? |
|
Tags |
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling |
Thread Tools | Search this Thread |
Display Modes | |
|
|