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. |
1st March 2014, 12:10 | #24101 | Link |
Registered User
Join Date: Nov 2012
Posts: 99
|
One question beside the dithering hype:
I now replaced my old AMD HD4550 by a R7 250. Formerly I used Powerstrip for custom resolutions to get judder free playback of 23,976 / 24.000 / 25,000 / 29,970 fps movies. Powerstrip doesnīt support the newer GPUs so I switched to the EDID override method used by CRU utility. Works like a charm, so I have the four different refresh rates for my new GPU (refresh rate = fps x 2 each because Iīm feeding a projector which accepts refresh rates in the 30 - 60 Hz range). Now as a bonus i can use the madvrīs display switcher, works great too. No Powerstrip needed anymore, automatic switching. The custom resolutions are accurate enough for a judder free playback > 3h. But there is one problem: In Windows the decimal numbers are truncated /rounded, so 48,000 and 47,952 Hz canīt be distinguished. It is not possible to keep both refresh rates side by side. I want to work with the display switcher and the accurate refresh rates, NOT with smoothmotion as workaround. Does anybody know a solution for this problem? At the moment Iīm working with 720p48 /720p50 /720p60 in the display switcher (720p projector!) by using the 47,952 setting for 720p48, so I canīt use the 48,000 setting, I hope you know what I mean... There must be a solution, is there no place where you can fool Windows with the names, e.g. in the registry? A possible workaround would be to use 24,000 or 72,000, but my projector doesnīt accept refresh rates so low / so high. Hope somebody has worked around this problem and can give me some assistance. Hannes. |
1st March 2014, 12:18 | #24102 | Link | ||||||||||
Registered Developer
Join Date: Sep 2006
Posts: 9,140
|
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Quote:
Are you using the official v0.87.4 build or one of the recent test builds? The official v0.87.4 build still uses OpenCL to do error diffusion. The latest test builds use DirectCompute instead, which seems to work better for both AMD and NVidia GPUs. About glitches: Have you tried increasing the number of pre-presented frames, as I suggested earlier? If that doesn't work, then I don't know what else to do. As I said before, glitches are a problem between Direct3D and the GPU driver, so it's mostly out of my control. |
||||||||||
1st March 2014, 13:25 | #24105 | Link | |
Registered User
Join Date: Apr 2009
Posts: 1,019
|
Quote:
Mono Ordered Dither displays black correctly, Colored Ordered Dither shows some noise on black. When using 3DLUTs, Mono dither does not show any noise with the Error Diffusion options, but does show noise on black with Colored noise. I thought it might simply be due to the fact that there are so few colors, but I wonder if there might be a solution for this? |
|
1st March 2014, 13:57 | #24106 | Link | ||
Registered Developer
Join Date: Sep 2006
Posts: 9,140
|
Quote:
Quote:
Please feel free to report this as a bug in the bug tracker. It's not really a bug, but I should be able to add a limiter which stops pixels from getting noise if they're truely black. |
||
1st March 2014, 14:02 | #24107 | Link | |
Registered User
Join Date: Sep 2013
Posts: 919
|
Quote:
I also heard you got some funky 2-bit version... When using Static dithering at low bit depth the image appears brighter, but in dynamic mode the gamma appears correct. You can pause the video in dynamic mode (too look like Static) and press the Up/Down arrow and keep holding (volume) to update the picture and look like Dynamic but paused. You can remove the OSD (Ctrl+J) to stop the refreshing every second that the OSD is doing.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410. Last edited by James Freeman; 1st March 2014 at 14:31. |
|
1st March 2014, 14:05 | #24108 | Link |
Registered User
Join Date: Oct 2013
Location: Sweden
Posts: 128
|
One thing that is noticable is that the image is getting brighter the lower bit depth you dither to.
Linear light would help here as stated before, so i made some graphs to took at how much brighter or darker the image gets when converting a bt709 gamma encoded image to linear, using power functions. Curves! If you are using a power function to decode gamma i think ~1.85 would be the best approximation, it gives an integral between 0(black) and 1(white) of ~1.01. Perfect linear gamma gives an integral of 1. And the previous discussion about linear light was a bit misleading because the discussed images was near black and almost invisible if not enhanced. Last edited by bacondither; 1st March 2014 at 14:17. |
1st March 2014, 14:27 | #24109 | Link | |
Registered User
Join Date: Apr 2009
Posts: 1,019
|
Quote:
Because it's a video file, it seems that one image showed noise on black and the other did not. It just coincided with me toggling the "use colored noise" option. Normally, there is some amount of noise on black, but not below-black, and with a 3DLUT I'm seeing noise on below-black values (which should be clipped) too. Looks like it's back to Error Diffusion and the "This display is already calibrated" options for me, which at least eliminates the noise on black for now. |
|
1st March 2014, 14:35 | #24110 | Link | |
Registered User
Join Date: Sep 2013
Posts: 919
|
Quote:
Try to turn up Contrast (and/or turning down Brightness) in MPC-HC to see it clearer. Are you the one with the plasma that does not dim the black area because of this?
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410. Last edited by James Freeman; 1st March 2014 at 14:40. |
|
1st March 2014, 15:11 | #24113 | Link | |
Registered User
Join Date: Dec 2008
Posts: 496
|
Quote:
Still, we have to keep in mind that itīs a "half-hack" (like madshi already stated), because we would only need it for the final conversion anyway, not further processing. Also, if we had access to a 2bit/1bit mode, too, we would be able to analyse this even better. When we analysed LL last time, we didnīt even have something like the 4bit mode "hack". Last edited by iSunrise; 1st March 2014 at 15:47. |
|
1st March 2014, 16:16 | #24115 | Link |
Registered User
Join Date: Dec 2013
Posts: 753
|
The 'blinking' seems to be caused by performing smooth motion in linear light, and dithering the result in gamma light. If you either stop smooth motion or perform the blending in gamma light then the blinking stops. So if you want to solve the 'blinking' then dithering and smooth motion should both use the same linear light conversion (or none at all).
Last edited by Shiandow; 1st March 2014 at 16:19. |
1st March 2014, 16:29 | #24116 | Link |
Registered Developer
Join Date: Sep 2006
Posts: 9,140
|
FWIW, I'm getting blinking here without smooth motion frc, too, if the framerates don't match.
Anyway, linear light error diffusion has serious algorithm problems. Without a limiter it should work, but it would increase noise quite noticeably. I don't plan to try linear light error diffusion again at moment. |
1st March 2014, 17:41 | #24118 | Link |
Registered User
Join Date: Oct 2013
Location: Sweden
Posts: 128
|
@madshi
Why is reference white dithered with error diffusion?. There should be no error to spread to neighbour pixels. All testing is done in 8-bit output in madvr. No shaders, debanding or 3dlut is active. Disabling colored noise or "dynamic dither" made no difference. All images are contrast enhanced. White clipping - ED1 White clipping - ED2 White clipping - Random I created a pure white image with RGB 255 on all pixels and with random dithering after contrast enhancment the image is white. With both ED modes it shows dither snow after enhancement, it should show all white! RGB 255 - Random dithering. RGB 255 - Error diffusion. It is a minor issue but i dont understand why ED is dithering pure white. |
1st March 2014, 17:47 | #24119 | Link | |
Registered User
Join Date: Sep 2013
Posts: 919
|
Quote:
madshi, Its already in the Bug Tracker by 6233638. If you think its appropriate, do the same Limiting procedure to the whites.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410. |
|
Tags |
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling |
|
|