madVR - high quality video renderer (GPU assisted)

[Asmodian]

Should I calibrate my monitor using a blu-ray player. and read these values ​​and insert them into ycms. Or can I calibrate using mpc-hc> madvr and paste values ​​from hcfr

Calibrate your monitor however you want to get a good calibration. I seem to get best results with dispcalGUI (Argyllcms) doing a curves+matrix calibration, I have been unable to get a good LUT calibration with dispcalGUI.

I then read the primaries & gray scale of my calibrated monitor using HCFR's own generator but I get the same values if I read "Free Measure" with red, green, blue, and white videos through MadVR when MadVR is set to "uncalibrated".

I then setup yCMS in MadVR with the values from HCFR and can check the output with more "Free Measures".

I just tried it again and got this after calibration:
Red Yxy 43.191177 0.676602 0.311908
Green Yxy 107.544548 0.202671 0.695760
Blue Yxy 12.016151 0.150056 0.059922
White Yxy 161.210911 0.313837 0.330033
Yellow Yxy 150.107807 0.426857 0.515279
Cyan Yxy 119.190946 0.172719 0.338045
Magenta Yxy 54.203559 0.363243 0.161500

And this after setting up yCMS in MadVR using the values above:
Red Yxy 33.705097 0.631845 0.339083
Green Yxy 110.922828 0.311980 0.594759
Blue Yxy 13.627455 0.154915 0.069454
White Yxy 157.175545 0.312975 0.329333
Yellow Yxy 142.449967 0.422370 0.506264
Cyan Yxy 122.954295 0.231975 0.327312
Magenta Yxy 46.949548 0.314173 0.160600

Gray dots are what I get using yCMS in MadVR
http://i1222.photobucket.com/albums/dd496/asmodian3/W3000H_MadVRyCMS_2.png

So it is not perfect but so much better than without it on this wide gamut display. No banding or anything like that. Though nasty things still do happen if I use <20% gray scale values.

[6233638]

This is what I typically see on displays as well. yCMS does pretty good guesswork at reigning in wide gamut displays, but could be so much more accurate if we were offered the control to fine-tune the RGB points to be correct. I do wonder if there is a bug with how green is handled, as that seems to be a common error.

I think you should be able to get much more accurate results if yCMS had some kind of control to adjust HSL (or xyY position) for each primary.

It is managing to get secondaries near perfect for your display from the look of things, which is always stupidly impressive to me for some reason. I guess because I am used to seeing color management systems that are broken and "twist" when you adjust RGB and don’t place the secondaries correctly.

Note: correct placement is to be opposite the primary, through the measured white point. It is not "on spec" for BT.709 unless your primaries are "on spec".

[Asmodian]

Thank you 6233638, I am liking the look better after taking your advice to not use yCMS for gray scale correction. I copied your method and use a 2.4 gamma on my calibrated to 2.2 display.

Is there a benefit to using D65 for the color values instead of 100% white as measured? Somehow it makes more sense to me to use the same color white for both the gamut and gamma but I don't really know why I would think that. They are pretty close in my case and I cannot tell the difference as long as I set gray scale to all the same color.

[6233638]

Is there a benefit to using D65 for the color values instead of 100% white as measured? Somehow it makes more sense to me to use the same color white for both the gamut and gamma but I don't really know why I would think that. They are pretty close in my case and I cannot tell the difference as long as I set gray scale to all the same color.If you aren’t seeing problems, use the measured value. I had a near-perfect D65 and once I let yCMS do its correction, greyscale was noticeably tinted after, which is why I just used D65 for the xy values.

Using 100% white for all values if you are using gamut correction is something I did intend to try, but I haven’t had as much time to spend on calibration stuff as I'd like recently.

Edit: now that you have switched to a uv chart rather than xy, it looks like there is probably nothing wrong with how yCMS is handling green, it's just that xy exaggerates the error. It's also interesting to see that your display actually has roughly the same over saturation with red as it does with green, when the xy chart made it look like the error was primarily green.

This is why I wish programs would move over to uv charts by default, as it's much more perceptually uniform.
Now if only HCFR would use the same scale for the horizontal and vertical axes, rather than stretching the charts out to fill your display… (I think there is an option to set the dimensions when exporting the chart at least)

[Xaurus]

madshi,

I know you wrote this in March 2012 in the eac3to thread:

eac3to is not dead, it's just kinda "on hold" for the moment, because I'm busy working on madVR. Once madVR has reached v1.0, I'll get back to eac3to development.
There are big problems with certain (7.1) TrueHD tracks from new BD releases. I don't know what is needed, an update of the libav decoder perhaps. Just thought I'd let you know.

[MSL_DK]

I achieve this with ycms.

before ycms

0.622909 0.301711 0.150618
0.335525 0.645762 0.050345
18.527370 62.119041 3.016925

https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/test.JPG

after ycms

0.617490 0.305857 0.155411
0.331776 0.598824 0.076203
19.131506 65.204896 7.170791

https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/result.JPG

Blue is washed out and I see bars from 219 to 251 on blue and only blue. 4-Color Clipping (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/4-Color%20Clipping.mp4_snapshot_00.17_%5B2012.11.14_04.29.19%5D.jpg) and Clip Pattern (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/WRGBCYM%20Clip%20Pattern.mp4_snapshot_00.53_%5B2012.11.14_04.26.50%5D.jpg) The problem does not occur when using shader in mpc-hc.

[annovif]

With "lsfmod" (an avisynth sharpener script) i set supersampling to 1.4 so the script can apply the sharpness effect with more precision over a 2688x1512 resolution. Then i set the final dest_x and y (really a spline36) to resize the image to 4k resolution .
With Madvr i read in input a 3840x2160 resolution , so i can use madvr to downscale to full hd with bicubic 100 setting. Than i can joke with chroma parameters (another bicubic 100) and exclusive mode setted all flushed. Can i know your opinion? Do you think i have to use the new jinc algoritm for chroma and downscaling?The power of the pc or vga is not a problem for me , i only want the best image possible. If you are curios here there is my avisynth post http://forum.doom9.org/showthread.php?t=164628

Thank you very much , Fabio:helpful:

[Pra7ul]

i am getting problem playing file in MPC-HC with madVR..

i asked on animepassion.net about it then i was forwarded here..

Please take a look at this : -|HERE|- ("http://www.animepassion.net/index.php?/topic/348-cccp-vs-coreavc-vs-madvr/#entry19709)

My PC Specs if required:

MoBo : Intel DG31PR
CPU : Intel Core2Quad Q6600
No graphic card except internal 256MB (pixelshader 2.0)


help much appreciated , i'v been looking for this answer for months now.

[italospain]

With "lsfmod" (an avisynth sharpener script) i set supersampling to 1.4 so the script can apply the sharpness effect with more precision over a 2688x1512 resolution. Then i set the final dest_x and y (really a spline36) to resize the image to 4k resolution .
With Madvr i read in input a 3840x2160 resolution , so i can use madvr to downscale to full hd with bicubic 100 setting. Than i can joke with chroma parameters (another bicubic 100) and exclusive mode setted all flushed. Can i know your opinion? Do you think i have to use the new jinc algoritm for chroma and downscaling?The power of the pc or vga is not a problem for me , i only want the best image possible. If you are curios here there is my avisynth post http://forum.doom9.org/showthread.php?t=164628

Thank you very much , Fabio:helpful:

The Master of video postprocessing has answered your question in your avisynth post. A second opinion is pointless.:sly:

[annovif]

italospain wrote "The Master of video postprocessing has answered your question in your avisynth post. A second opinion is pointless"

...i wrote this post to improve the madvr settings .

Thank you, Fabio

[6233638]

I don't know anything about LSF, but optimal downsampling in madVR is achieved by using Catmull-Rom with the anti-ringing filter and linear light scaling for luma. You may not get the results that you want with a sharpened image though.

I achieve this with ycms.

before ycms

https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/test.JPG

https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/result.JPG

Blue is washed out and I see bars from 219 to 251 on blue and only blue. 4-Color Clipping (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/4-Color%20Clipping.mp4_snapshot_00.17_%5B2012.11.14_04.29.19%5D.jpg) and Clip Pattern (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/WRGBCYM%20Clip%20Pattern.mp4_snapshot_00.53_%5B2012.11.14_04.26.50%5D.jpg) The problem does not occur when using shader in mpc-hc.The clipping will be occurring because the final yCMS gamut is undersaturated.

I don't know why yCMS is giving you such under-saturated results for blue though. Are the values for xyY correct for your display? Make sure you are taking a measurement of madVR playing back 100% saturation test patterns with the calibration controls and gamma processing disabled.

[annovif]

6233638 wrote "optimal downsampling in madVR is achieved by using Catmull-Rom with the anti-ringing filter and linear light scaling for luma. You may not get the results that you want with a sharpened image though"

thank you, i will try it for downsampling. And for chroma and an eventually upsampling from a 720p what settings can i try? Jinc is not useful ?

Thank you, Fabio

[MSL_DK]

I don't know anything about LSF, but optimal downsampling in madVR is achieved by using Catmull-Rom with the anti-ringing filter and linear light scaling for luma. You may not get the results that you want with a sharpened image though.

The clipping will be occurring because the final yCMS gamut is undersaturated.

I don't know why yCMS is giving you such under-saturated results for blue though. Are the values for xyY correct for your display? Make sure you are taking a measurement of madVR playing back 100% saturation test patterns with the calibration controls and gamma processing disabled.

That's exactly how I do it ... I suspected my Spyder4 to measure incorrectly, therefore, I tried with the shader in MPC-HC, there is the correct result.

Can I get you to make a eks.txt for use in either ycms or madvr> ycms from my .chc file from HCFR EKS.CHC (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/eks.chc)

It may be I overlook a trivial error. I'm about to give up. I do not understand it ... RGB Levels, gamma curve and luminance looks absolutely perfect.

[sexus]

6233638 wrote "optimal downsampling in madVR is achieved by using Catmull-Rom with the anti-ringing filter and linear light scaling for luma. You may not get the results that you want with a sharpened image though"

thank you, i will try it for downsampling. And for chroma and an eventually upsampling from a 720p what settings can i try? Jinc is not useful ?

Thank you, Fabio

woah thats alot of aliasing id rather go for lanczos 3 with antiringing and linear light scaling, same for image upscaling but no liniear light scaling for that one, except for chroma there you go for mitchell netravali and antiringing filter, and you got the best of the best , unless someone would like to update me ;)

p.s: btw what is this new thing going on with people using lav filters , madvr with this thing is it worth it !!! >>>> http://forum.doom9.org/showthread.php?t=166082&page=3 ???????

or should i stay with madvr , lavfilters pack and ffdshow raw filter in mpchc

[Asmodian]

If you aren’t seeing problems, use the measured value. I had a near-perfect D65 and once I let yCMS do its correction, greyscale was noticeably tinted after, which is why I just used D65 for the xy values.

Using 100% white for all values if you are using gamut correction is something I did intend to try, but I haven’t had as much time to spend on calibration stuff as I'd like recently.

Great, thanks.

Edit: now that you have switched to a uv chart rather than xy, it looks like there is probably nothing wrong with how yCMS is handling green, it's just that xy exaggerates the error. It's also interesting to see that your display actually has roughly the same over saturation with red as it does with green, when the xy chart made it look like the error was primarily green.

This is why I wish programs would move over to uv charts by default, as it's much more perceptually uniform.
Now if only HCFR would use the same scale for the horizontal and vertical axes, rather than stretching the charts out to fill your display… (I think there is an option to set the dimensions when exporting the chart at least)

I noticed your comment about uv mode being more perceptually relevant so I swapped the charts. I see what you mean too, it is much easier to see that yCMS is matching the hue to spec and matching all saturations to each other. I did make a special effort to not have the chart stretched in either dimension as it can get quite distorted. I am not sure that HCFR is using same scale for the horizontal vs. vertical axis but it will not scale either axis down anymore no matter what the window size.

I assume the saturation of the primaries has to match if you want the secondaries to be on spec for hue?

[6233638]

That's exactly how I do it ... I suspected my Spyder4 to measure incorrectly, therefore, I tried with the shader in MPC-HC, there is the correct result.

Can I get you to make a eks.txt for use in either ycms or madvr> ycms from my .chc file from HCFR EKS.CHC (https://dl.dropbox.com/u/111324524/Images%20for%20Sharing/madVR/eks.chc)

It may be I overlook a trivial error. I'm about to give up. I do not understand it ... RGB Levels, gamma curve and luminance looks absolutely perfect.Are you creating the 3DLUT file externally, or entering your data into madVR? i.e.

http://i.imgur.com/oZoNJ.png

I don't have HCFR installed right now. (I use CalMAN so I'd rather not install it to be honest)

woah thats alot of aliasing id rather go for lanczos 3 with antiringing and linear light scaling, same for image upscaling but no liniear light scaling for that one, except for chroma there you go for mitchell netravali and antiringing filter, and you got the best of the best , unless someone would like to update me ;)Are you just looking at the madVR charts, or are you actually seeing aliasing? Please post a sample if you are.

It's possible that things have changed since my last testing, but Lanczos still had ringing problems even with the anti-ringing filter enabled when downsampling in linear light the last time I looked at it.

My preference is Jinc 3 anti-ring for both Chroma and Luma upsampling now. (not in linear light) This can be quite demanding on hardware though, and doesn't completely eliminate ringing and artefacts the way that SoftCubic 80 did, but I think the sharpness improvement may be worth it.

I haven't had time to go really in-depth with my testing, but there haven't been any problems that have stood out with this combination.

EDIT: I think you posted this as I replied:I noticed your comment about uv mode being more perceptually relevant so I swapped the charts. I see what you mean too, it is much easier to see that yCMS is matching the hue to spec and matching all saturations to each other. I did make a special effort to not have the chart stretched in either dimension as it can get quite distorted. I am not sure that HCFR is using same scale for the horizontal vs. vertical axis but it will not scale either axis down anymore no matter what the window size.I think HCFR has a specific export function (that changes the background on the charts to white by default) Here, you can specify dimensions and I think using the same for width and height will give you the same scale on both axes. (or close enough)

I assume the saturation of the primaries has to match if you want the secondaries to be on spec for hue?For hue, what you do is draw a path through wherever the measured primary point is, through your white point, and wherever that lands up is your target hue.
For example green hue (not saturation) and greyscale affects the position of magenta hue.
Even if green saturation is out of spec, if green hue and white balance are on spec, magenta hue will be on spec.

CalMAN can automate this by creating a new calibration target once you have calibrated Greyscale and RGB as best as you can. It then takes that data and calculates new secondary hue positions and target luminance values for RGBCMY. (if your saturations are not on spec, luminance target changes as well)

I'm not sure that HCFR has this function. There was a stand-alone program that can also calculate these new values (I found it on AVS Forum a number of years ago) but I do not recall its name.

As for saturations, I don't think it is necessary to match them, but rather to get each primary as accurate to BT.709 as possible. If blue is under-saturated, but red and green are able to hit the BT.709 target, I think that you should get them as accurate as possible, rather than constrain the entire gamut to the level of saturation that blue has.

It's possible that this is what yCMS is trying to do though, I'm not sure. If it looks that way, it may just be coincidence.

I believe that secondaries should then be placed along the line between each primary. E.g. Magenta saturation should fit on the line between red and blue. This is actually something you might want to bring up over at the AVS Forum display calibration section, as I have not seem much discussion on the topic.

It's possible that it may be better to try and match secondary saturation (but not hue) as close to BT.709 spec, if possible, rather than limit it based on the two primaries it lies between.

It will look odd in programs which only draw gamut as a triangle, rather than a 6-sided shape, but may not necessarily be a bad thing.

[sexus]

ok thanks as said im open to suggestions , btw yes ive just looked at the madvr charts lols xD, btw i guess lanczos3 it is then for downscaling ey or Catmull-Rom with the anti-ringing filter and linear light scaling as suggested? or what you really think catmull-rom is best? never seen it in real but those charts look scary , lol

what about this finesharpen mumbo jumbo going on is it worth anything or what as ive linked too

[leeperry]

So I'm on the 306.81 drivers/XPSP3 and using the new rendering path, presentation glitches keep increasing every few secs just as they did before when OTOH the old rendering path works like a charm...so please be so kind as to not toss it :o

:thanks:

My preference is Jinc 3 anti-ring for both Chroma and Luma upsampling now. (not in linear light)
Yes, it does look impressive! I wish I could use softcubic50 AR for SD chroma and Jinc3 AR for HD chroma.....I believe madshi said that adding automatic profiles for different resolutions was on his TODO list :)

I don't have HCFR installed right now. (I use CalMAN so I'd rather not install it to be honest)
HCFR3 comes unpacked in an archive file, there's nothing to install.

[6233638]

Unfortunately, it seems the tool I was thinking of was the AccuPel Display Calibration Calculator (http://www.abload.de/img/e09a34d3_vbattach1136f6shq.jpg). AccuPel has been acquired by ChromaPure, and the software no longer seems to be available for download. You might want to try contacting them, if HCFR does not have the ability to set a custom calibration target based on your measured primaries.

They do have some spreadsheets available to download which may calculate everything you need, however: http://www.chromapure.com/colorscience-calculators.asp

As mentioned though, until yCMS offers more control, you're pretty much stuck with whatever it calculates as being "best" for gamut, at least if you are creating the LUT using its tools.
I don't know if converting an ArgyllCMS file to a yCMS 3DLUT (http://forum.doom9.org/showthread.php?t=162285) may work better for gamut correction. I did not have good results (http://forum.doom9.org/showpost.php?p=1540847&postcount=230) when I tried it.

Similar to the "trick" I use of adjusting the Y values to change gamma at specific points (this is one of the reasons I use xyY rather than XYZ) I tried adjusting my input primary color values, but was unable to push the output primaries/secondaries into a desirable result.

But still, if you have large gamut errors, and no CMS on your display, even if yCMS does not manage to get things perfect, it is often a big improvement.

HCFR3 comes unpacked in an archive file, there's nothing to install.Thanks, I was not aware of this.

ok thanks as said im open to suggestions , btw yes ive just looked at the madvr charts lols xD, btw i guess lanczos3 it is then for downscaling ey or Catmull-Rom with the anti-ringing filter and linear light scaling as suggested? or what you really think catmull-rom is best? never seen it in real but those charts look scary , lolWith linear light scaling enabled, everything changes. Catmull-Rom with the anti-ringing filter was not only the best option, but it was the only option which produced satisfactory results in my tests. I would only use the graphs in the madVR as a rough guide, as they may not be accurate for all combinations of settings.

If you want to use something else for downsampling, I would not recommend having the linear light scaling option enabled. (even though linear light scaling can be important when downsampling)

what about this finesharpen mumbo jumbo going on is it worth anything or what as ive linked tooOne of the results posted looked interesting, but generally sharpening is detrimental to image quality in my experience, and I would certainly not want to be sharpening while upsampling the image. I think that discussion belongs there, rather than in this topic though.

[sexus]

thanks for the replies make sense , ill go with catmull then , thanks , and yes i guess your right about the sharpening filter thread

[dansrfe]

So, what are the recommended image upscaling/downscaling and chroma resize algorithms and options?

[ryrynz]

*Brings out the dead horse*

[sexus]

lmfao

[pirlouy]

So, what are the recommended image upscaling/downscaling and chroma resize algorithms and options?
As you guess, people won't agree on the good answer.

But it seems most users agree on upscaling with Jinc 3 and Madshi Anti-Ringing option enabled.
There's a good chance it works with ATI or Nvidia GPU from 2-3 years and output like 30fps 1080p.
If your GPU cannot handle this, try Lanczos 3 (taps) with AR. If the GPU still can't handle this, nobody agrees on what is the best upscaling after these 2.

For downscaling algorithm, I think most of us don't need it. But you know, there always be (strange) people with 60fps source, or 1440p output ! :D

[nevcairiel]

But you know, there always be (strange) people with 60fps source, or 1440p output ! :D

My Desktop has a 2560x1600 screen, but i don't usually use it for watching stuff. :p
Hardly strange, just the envy of people not being able to afford a proper screen. ;)

Its still somewhat sad that my 30" screen on the desktop has a higher resolution then the 50" screen of my tv..

[dansrfe]

Jinc 3 + Anti-Ringing for chroma too? What about scaling in linear light?

[nevcairiel]

What about scaling in linear light?

Not for upscaling. Some people recommend it for downscaling, but since the majority of people probably don't downscale, its hard to say.

[cremor]

Found a small bug in v0.84.7: If you open the downscaling settings screen once the Jinc RadioButton stays disabled even if you switch to upscaling pages.

[Prinz]

Could this be good news about directx 11.1 stereoscopic support for win7?

https://blogs.msdn.com/b/chuckw/archive/2012/11/14/directx-11-1-and-windows-7.aspx?Redirected=true

[mark0077]

Unfortunately I think this is a Windows 8 or driver bug.

If I disable FSE and the display mode switcher in madVR, and use the MPC-HC switcher instead, the same thing happens.

It doesn’t switch to 59Hz when playing back 30p at 60Hz like madVR does (the display is already at 60Hz by default, but madVR still switches upon entering FSE) but if I start with my display at 50Hz, it will switch to 59Hz rather than 60Hz for playback.

It also switches to 23Hz rather than 24Hz.

However, unlike madVR, it does manage to restore my display to 60Hz after playback. (madVR "restores" to 59Hz)


If I had to guess, Windows 8 must be doing something different with how it's handling the requests to switch resolution/refresh rate, if MPC-HC is somehow managing to restore resolution correctly, but not switch correctly. (I guess restoring reads the current value and restores it, rather than sending a "switch to 60Hz" command)

Yeah its so annoying, and 59hz on my nvidia card is actually exactly 59.90hz, not 59.94hz so I was seeing dropped frames every few seconds. I tried lots of workarounds like using the OverrideEdidFlags0 in the nv_disp.inf and adding custom resolutions of 60.001 etc, but I found only one that worked around it for me, which was adding a custom resolution of 59hz, and manually setting its rate to 59.95hz (don't ask me why, these figures are always a mystery). Any higher than 59.95 and it would still switch to 59.90 hz on FSE, any lower and nvidia said the custom resolution failed... I hate stupid windows / nvidia bugs like this.. Now I can switch into FSE mode without the display rate changing from what I want, 60.0 hz.

[frozenice]

Thanks mark0077 for your "Workaround", i try it later on my System.

I hope Nvidia/Microsoft already know about this Bug, so it can be fixed verry soon. Don't want downgrade to Windows 7 :-D

[aufkrawall]

I hope Nvidia/Microsoft already know about this Bug, so it can be fixed verry soon. Don't want downgrade to Windows 7 :-D
Reporting it to Nvidia won't hurt:
https://forums.geforce.com/default/topic/523630/geforce-drivers/official-nvidia-r310-54-beta-display-driver-feedback-thread-released-11-12-12-/

I'd do it for you via a better way but unfortunately I can't since I don't own Windows 8 and won't ever.

[Prinz]

Could this be good news about directx 11.1 stereoscopic support for win7?

https://blogs.msdn.com/b/chuckw/archive/2012/11/14/directx-11-1-and-windows-7.aspx?Redirected=true

I read a bit more. Stereoscopic isn't supported for windows 7, because it needs wddm 1.2. :(

[6233638]

Yeah its so annoying, and 59hz on my nvidia card is actually exactly 59.90hz, not 59.94hz so I was seeing dropped frames every few seconds. I tried lots of workarounds like using the OverrideEdidFlags0 in the nv_disp.inf and adding custom resolutions of 60.001 etc, but I found only one that worked around it for me, which was adding a custom resolution of 59hz, and manually setting its rate to 59.95hz (don't ask me why, these figures are always a mystery). Any higher than 59.95 and it would still switch to 59.90 hz on FSE, any lower and nvidia said the custom resolution failed... I hate stupid windows / nvidia bugs like this.. Now I can switch into FSE mode without the display rate changing from what I want, 60.0 hz.Thanks for the suggestion. Unfortunately, my display only accepts CEA/EIA timings so that's not an option here.

[mark0077]

Thanks for the suggestion. Unfortunately, my display only accepts CEA/EIA timings so that's not an option here.

Yeah its a mess. I reported it at the nvidia forums but I'm sure fixes in games will always come before things like this which seem to go on for years.

[nevcairiel]

This is why you don't go and adopt a new OS for a system you rely on for anything. :p

[6233638]

This is why you don't go and adopt a new OS for a system you rely on for anything. :p24p running at 23Hz is not the end of the world, because surely it will be fixed in a reasonable amount of time, assuming it's an Nvidia driver bug and not something relating to madVR. (as Madshi isn't running Windows 8, it could be some time before it gets fixed if it's not an Nvidia problem)

I haven't seen any reports from Intel/AMD users on Windows 8 though, so it's difficult to know whether it's an Nvidia-specific issue, or Windows 8 in general.


There are a number of features and improvements that have made Windows 8 a worthwhile upgrade for me already though, and there have been very few issues so far.

[turbojet]

For those that want the 'natural look' of Jinc but can't use it due to hardware requirements or those wishing Jinc was sharper or those just looking for another resizer to mess with there's the little known spline100/144 (http://www.wilbertdijkhof.com/SplineResize_v02.zip). To try it in ffdshow:

function Resizer(clip c,int newwidth) {
nhtemp=floor(eval(string(c.height)+". * "+string(newwidth)+". / "+string(c.width)+"."))
nh=(nhtemp+2)/4*4
spline144Resize(c,newwidth,nh)
}
Resizer(1920) #set screen width here

If you have multiple displays profiles can be setup to auto set resolution but it doesn't work on extended displays (ffdshow only recognizes primary display).

I mention it in this thread, again, along with a way to use it, because it's so little known and deserves a lot more attention. Also to see what madshi thinks about it, first post with comparisons was buried among a bunch of ycms talk.

[6233638]

For those that want the 'natural look' of Jinc but can't use it due to hardware requirements or those wishing Jinc was sharper or those just looking for another resizer to mess with there's the little known spline100/144 (http://www.wilbertdijkhof.com/SplineResize_v02.zip)How does this differ from Spline in madVR?

[cyberbeing]

How does this differ from Spline in madVR?

madVR uses Spline36 (3-tap 6x6) and Spline64 (4-tap 8x8)

I mention it in this thread, again, along with a way to use it, because it's so little known and deserves a lot more attention.

I know I suggested the sharper splines ~3 months ago, and I still think it would be interesting to test with madVR anti-ringing.

One thing I can say though is that Wilbert's spline implementation in that plugin is different from madshi's in madVR. The plugin's Spline100 (5-tap 10x10) has extremely soft results similar to madVR's Mitchell-Netravali, while the Spline144 (6-tap 12x12) results are extremely sharp surpassing even madVR's Lanczos8 sharpness by a large margin. It makes me question whether the sharper splines were implemented correctly, or if Wilbert is using a different type of spline compared to madVR. Even so, I'd be interested what a madVR Spline144 w/ anti-ringing would look like.

[secvensor]

Asked already, but can that that has changed.
Whether it is possible to use it in madVR?
http:// www.twcenter.net/forums/showthread.php? t=536904
http://mrhaandi.blogspot.com/p/injectsmaa.html

[ryrynz]

For those that want the 'natural look' of Jinc but can't use it due to hardware requirements or those wishing Jinc was sharper or those just looking for another resizer to mess with there's the little known spline100/144 (http://www.wilbertdijkhof.com/SplineResize_v02.zip).


Too sharp for my tastes, though I've only looked at 144 ATM. Wouldn't mind seeing it with MadVR's anti-ringing on it.

[Keiyakusha]

Just checked these spline resizers:
Spline144. It is sharper compared to spline36 but it adds like 40% sharpness, 60% artifacts. So don't see any use for it. By using bicubic with sharpening on top of it I can achieve same sharpness with almost no artifacts.
Spline100 to me looks like bicubic+lots of aliasing. Don't see any use for it either. Maybe it is not that surprising that they are unpopular?
I was upscaling anime vector-based image rendered at 720x480 resolution (non-square pixels) to 1920x1080. On less perfect sources artifacts may be amplified...

[cyberbeing]

When I first suggested this, it was under the expectation that Spline144 would have sharpness between Lanczos3 & Lanczos4 with less ringing, while Spline256 would have sharpness between Lanczos4 & Lanczos8 also with less ringing. If madshi implemented the sharper splines in madVR, I would expect them to be nice reduced ringing & artifact alternatives to the Lanczos4, and look nothing like the extremely soft Spline100 and the insanely sharp ringing & artifact heavy Spline144 from Wilbert's plugin. Unless I'm unaware of something that causes these higher-order splines to take on strange properties like this?

[06_taro]

madVR uses Spline36 (3-tap 6x6) and Spline64 (4-tap 8x8)



I know I suggested the sharper splines ~3 months ago, and I still think it would be interesting to test with madVR anti-ringing.

One thing I can say though is that Wilbert's spline implementation in that plugin is different from madshi's in madVR. The plugin's Spline100 (5-tap 10x10) has extremely soft results similar to madVR's Mitchell-Netravali, while the Spline144 (6-tap 12x12) results are extremely sharp surpassing even madVR's Lanczos8 sharpness by a large margin. It makes me question whether the sharper splines were implemented correctly, or if Wilbert is using a different type of spline compared to madVR. Even so, I'd be interested what a madVR Spline144 w/ anti-ringing would look like.

And if you compare SplineResize(taps=5) with Spline100Resize, or compare SplineResize(taps=6) with Spline144Resize, you will also find large differences :o

[nlnl]

madshi
My config:
Soft - W7, MPC HC 1.6.5.6187, Madvr 0.84.7, LAV audio (bitstreamig), LAV video 0.53.1
Video chain - ATI 5750 (passive), Onkyo 1007 AVR, Pioneer display

All scaling is set to bilinear. FSE mode and Aero on.

Mediainfo and madvr log report for the file I play:
https://dl.dropbox.com/u/43376972/Madlog%20%2B%20mediainfo%2018-11-12.zip

I start playing file using DTS HD MA audio track at 23 hz (film), then switch to 59 (v) and have audio stuttering.
When playing PCM track it does not stutters but cracks and hisses.
Mediainfo and madvr log report for the file I play:
https://dl.dropbox.com/u/43376972/Madlog%20%2B%20mediainfo%2018-11-12.zip

(1) With the very same madVR chroma/image upscaling settings, does the issue only occur if you start with 23 and then switch to 59? Or does it also occur
if you start playback already with 59?
It's very important to know whether the 23 -> 59 switch is responsible for the problem, or whether the 23 -> 59 switch has nothing to do with the problem.
I guess that 23 (f) -> 59 (v) switch is responsible (or 59 (v) -> 23 (f)). The proplem does not occur starting playback with 59 (v).
(2) You say you need a Windows reload to cure the stuttering. Do you mean, once it stutters, it also stutters in 23 playback mode? Or does it only stutter in 59 playback mode then?
Yes, it starts stuttering in 23 and 59 playback mode.
(3) Instead of a Windows reboot, maybe manually changing display modes from 59 -> 23 (and maybe back) could also help?
Yes, manually changing display modes using CCC from 59 -> 23 or back helps.
(4) Are we talking about windowed playback? With aero on/off? Or overlay playback? Or FSE mode?
I am using FSE mode and aero on.

[leeperry]

This is why you don't go and adopt a new OS for a system you rely on for anything. :p
XP all the way baby! DS is bit-perfect when set to 100%, the stock TSC/7.8ms timings suck majorly but when forcing the ACPI timer at 0.98ms they are really quite good...I'm not sure that HPET at 0.5ms would make much improvement IRL because the ACPI clock is serious business(and OOTB W7 is often set to TSC at 10ms). No silly GUI, no troubles rolling refresh rates, it just works: amazing huh :devil:

BTW, the more it goes the more I'm realizing that outputting to 8bit forces you to choose between posterizing and dithering noise.....atm, I'm getting sick and tired of all this parasite noise being added onto my movies tbh. All it really does is flatten the picture and dull the colors. I've been using "Kung Fu Panda Holiday Special" for running tests lately, and on this very video I tend to prefer having dithering disabled: PQ is snappier; less hazy and just more 3D.

All this leading to my ole request of providing the ability to finetune the dithering strength in mVR so one could pick his personal "acceptable" compromise between posterizing and parasite noise on real world footage, not just test patterns that hardly ever reflect real world content IMHO. It's also mandatory to take the display built-in dithering in account IMO, especially when most of them work in 10/12bit internally...and the noisier the input, the more it will be amplified afterwards :o

:thanks: in advance for your attention on that matter.

[mindz]

My setup is powerful enough to run Jinc 4 or 8 AR. But most of the people here use Jinc 3 AR - and i can imagine more people have an even faster PC then mine. So, what are the pros and cons of using, say, Jinc 8 AR except for more sharpness and more PC load? Isnt more taps better in this regard, if a PC can handle the load?

[secvensor]

I can use the:
Jink
croma upscaling x4
image upscaling x8
on my hd4770 ^_^ (!!! 100MHz@1024x768 !!!)

[6233638]

My setup is powerful enough to run Jinc 4 or 8 AR. But most of the people here use Jinc 3 AR - and i can imagine more people have an even faster PC then mine. So, what are the pros and cons of using, say, Jinc 8 AR except for more sharpness and more PC load? Isnt more taps better in this regard, if a PC can handle the load?Generally with scaling algorithms, as the number of taps increases, sharpness goes up, aliasing goes down, and ringing increases.

While madVR's anti-ringing filter is great, I wouldn't recommend anything more than 3-tap filters for optimal image quality. (maybe 4-tap at a push with some filters)