madVR - high quality video renderer (GPU assisted)

[*Touche*]

I have to say that I don't see any "pop", "3D", "it's like looking through a window", "wow, so lifelike"...differences between these builds on my calibrated >15k:1 CR 55" plasma. No offence, but it sounds alot like the infamous audiophile BS to me.

In fact, I've even accidentaly left one of the static builds active the other night and haven't even noticed until after the movie. It's only when I concentrate do I see the actual differences, and then in noise only, not esotheric properties.

[James Freeman]

I'd suggest you go try BFI somewhere because boasting about 4:4:4 being superior when IRL movies are 4:2:0 to begin with and LCD is utterly sluggish without BFI...so much that both the picture itself and the dithering noise will be smeared all over in a big pixely jelly mess.


Yes I know all about BFI, I thoroughly read the Blur Busters website, and tried it myself.
I have tried BFI with an Avisynth script (72Hz monitor in 24+BF+BF, 24+24+Bf).
I also tried rotating mechanical shutter in sync with the monitor refresh rate.

The scrolling in movies look Amazing, no doubt.

But the colors and gamma DID change.
You can't deny that fast change from black to any shade of grey, doesn't cause overshoot.
It clearly does, especially in AMVA (VA) panels.

BFI or Blacklight Strobing is a fix for you brain, it fixes "eye-tracking based motion blur".
When the object appears at a different place, but your brain expects it to be in another, you get "eye-tracking based motion blur".

BUT, when the object (dithered pixels in our case) are stationary noise, BFI means absolutely NOTHING (nothing for your brain/eyes to track), so no "Jelly Mess" as you call it.
Pixel Response Time of the panel & Overshoot DOES.
BFI creates Overshoot.

Hope you understand.


Yes, movies are 4:2:0, but the dithering is RGB (4:4:4), and your panel is 4:2:2.
Add the blur effect of a 4:2:2 panel with the BFI overshoot and you got yourself an inaccurate picture.


I have to say that I don't see any "pop", "3D", "it's like looking through a window", "wow, so lifelike"...differences between these builds on my calibrated >15k:1 CR 55" plasma. No offence, but it sounds alot like the infamous audiophile BS to me.

WOW, 15k:1... its not a mere 3.5k:1 & BFI.... and you didn't even noticed the Static build was on....

No denying that it is... BUT be careful *Touche*, LeePerry got a user banned for this kind of comment (although it was not him this time). ;)

[leeperry]

I have to say that I don't see any "pop", "3D", "it's like looking through a window", "wow, so lifelike"...differences between these builds on my calibrated >15k:1 CR 55" plasma. No offence, but it sounds alot like the infamous audiophile BS to me.

In fact, I've even accidentaly left one of the static builds active the other night and haven't even noticed until after the movie.
To some ppl everything looks and sounds the same,if anything you are extremely lucky..more time for less frivolous activities, winning :)

Of course 6233638 would say that plasma is 4bit to begin with but I couldn't confirm as I've got no interest in plasma...those things flicker like helll to my eyes, and not just to me:

http://www.avsforum.com/t/945089/
Unlike LCD TVs, Plasma TVs flicker. This is the end-product of a design choice to create shading of colors. To do this Plasma panels use a method of strobing the pixels on and off to create the perception of shading of colors to the view.
http://www.avsforum.com/t/1119016/
Now I'm considering a return due to this
http://www.tested.com/forums/home-theater/13335-plasma-flicker/
I looked around the store when I returned the set, and could see flicker on all their plasmas, all brands and models


But the colors and gamma DID change.
Lo and behold, I run my calibrations with BFI enabled and everything's fine...you might have tried a poor BFI implementation, generalizations are never a good thing.

BFI or Blacklight Strobing is a fix for you brain, it fixes "eye-tracking based motion blur".
Of course not, it's a backlight-driven fix for the sluggishness of the LCD panel. All roads lead to Rome, as the panel is too slow we'll hide the motion blurring artifacts.

when the object (dithered pixels in our case) is in the exact same place (stationary), BFI means absolutely NOTHING.
Breaking news, mVR is a Video Renderer...meaning that there is motion taking place.

be careful *Touche*, LeePerry got a user banned for this kind of comment.
I'm no mod, you are imagining things again I'm afraid.

[bacondither]

Well, DLPs and Plasmas are 1-bit native, and by using a high speed temporal(dynamic) dither(at several khz),gives an apparent bitdeph somewhere around 6-7 bits for DLP.
I am sure they are using some spatial dithering in conjunction with the latter.

[Asmodian]

FWIW, I plan on releasing one more test build set in 1-2 days, probably containing 3 (very different) new test builds. One of them is going to be "simple" ordered dithering (based on a modified/improved version of "void-and-cluster"). The 2nd one is going to be a combination of ordered dithering and error diffusion. The 3rd one is going to be an updated low-noise error diffusion algorithm. Straight ordered dithering should perform faster than error diffusion. The combination of ordered dithering and error diffusion could maybe bring another quality improvement to dither dot placement. The low-noise-alternative might be another option to select for users. These test builds will probably only be monoColor, to make testing simpler. This final test build set will decide which algorithms are going to be offered as options in madVR.

This has me very interested, I liked some of the lower noise options in the first set of error diffusion builds. I will definitely give them a shot on my 27" 1440p IPS. :D

Of course not, it's a backlight-driven fix for the sluggishness of the LCD panel. All roads lead to Rome, as the panel is too slow we'll hide the motion blurring artifacts.

Actually this isn't the main reason BFI is so nice. It is actually retention on your retina, the light receptors have a change time and flashing black helps with motion blur, like Film, CRT, Plasma, and backlight based BFI LCD. :)
edit: I should have said the only reason BFI is so nice.

[madshi]

I'd still love to try new luma builds based on 0.97 & the NL6 random numbers that would look better in 4bit then.

[...]

Hopefully you won't give up on 0.97 & the A4/NL6 numbers
A4 is already 0.97.

I'd love to see a low noise A4 if that's technically possible :)
That would be oppositeColor... :p

I finally got some time to put my computer back together to be able to test the Mono/Colored/Opposite builds myself. (I moved while working 60h weeks :( )

I did my testing with very clean anime in 8-bit mode. I went into testing expecting to not be able to notice the difference or to like opposite-static because of what I have read so far and the screenshots as I think I like low noise. (no offence leeperry) :)

To my surprise I really do think the mono builds look better while I barely notice the noise difference between dynamic and static. I like "madVR_monoColor_static.ax" the most with "madVR_monoColor_dynamic.ax" a close second. I like Colored more than Opposite too. Weather this is due to my HVS system liking a small amount of almost invisible noise or something else I am not sure but in my subjective tests I really did like monocolor dithering more when watching the same videos over. I tried switching madVR versions in different orders and watching a few different scenes as I wasn't expecting the results. :o
Thanks for your feedback. Some votes for monoColor coming in late in the game now.

Done. (and marked it as a "major" bug by mistake - sorry)

Turns out it's the combination of NNEDI3 chroma upscaling (not doubling) + video mode deinterlacing that is causing it.
Thx.

[sajara]

I have to say that I don't see any "pop", "3D", "it's like looking through a window", "wow, so lifelike"...differences between these builds on my calibrated >15k:1 CR 55" plasma. No offence, but it sounds alot like the infamous audiophile BS to me.

^ :goodpost: And I second this entirely.

@madshi,

I've been testing oppositeColor over previous builds and my subjective opinion is that is my favorite so far.

I've tested this time in all my TV's, a 55" LED, a 32" LED, as well as my 15.4" laptop screen at 768p

Tested with blueray; B&W DVD media, several transcoded sample clips in SD and 1080p.

Now, I can only see dithering in 4bit mode. Period. In 8 bit is pretty damn impossible. Even zooming in with the magnifier is indiscernible.
So, I think madVR is making a pretty good job at it and improving at each new build.

[James Freeman]

BFI or Blacklight Strobing is a fix for you brain, it fixes "eye-tracking based motion blur".
Of course not, it's a backlight-driven fix for the sluggishness of the LCD panel. All roads lead to Rome, as the panel is too slow we'll hide the motion blurring artifacts.

You don't seem to understand the difference between BFI and Backlight Strobing.

Backlight stobing strobest after the pixel transition has already happened, so it hides the overshoot.
BFI is not backlight strobing, it does not hide the pixel transition, on the contrary, it creates it.

The transistors that driving each of the pixel elements having no fun to got from 0 to 5v every 1/120 of a second.
Especially when they are overdriven to have faster response time, thus creating overshoot.

Breaking news, mVR is a Video Renderer...meaning that there is motion taking place.

Random noise (dithering) is not a moving object for your eyes.
BFI/Strobing does nothing in terms of eye-tacking motion blur removal for pixel/noise.

[*Touche*]

To some ppl everything looks and sounds the same,if anything you are extremely lucky..more time for less frivolous activities, winning :)


It's not that I don't notice when something is different, or appreciate subtle improvements in audio or video, but I tend to be more critical, sceptical, scientific, down to earth with it. I have to keep in mind that some things are real and some are placebo. That is why I renamed the builds for testing and later compared the hashes to see which ones I could honestly identify.

That is not to say that I don't believe that colored builds look worse to you, but I would like more hard facts and blind tests and less of the beforementioned hard to take seriously impressions.

For further testing I think madshi should, at first, provide non-descriptic names and 8bit only, and we should try to test with real material without artificial test patterns and image manipulation to blow up the differences. That kind of scrutinizing should come after we choose favourites objectively.

Oh, and it would be fun to insert a duplicate build here and there, just to see if there will be differences between them :)

[DragonQ]

It's not that I don't notice when something is different, or appreciate subtle improvements in audio or video, but I tend to be more critical, sceptical, scientific, down to earth with it. I have to keep in mind that some things are real and some are placebo. That is why I renamed the builds for testing and later compared the hashes to see which ones I could honestly identify.
Completely agree, which is why I've largely stayed out of this discussion. There is at least some *actual* difference though, which is evidenced by the reduced-bit-depth comparisons, but whether they can be objectively differentiated with normal viewing material is another thing.

By comparison, most audiophile nonsense doesn't even have a scientific basis. :sly:

[Asmodian]

Thanks for your feedback. Some votes for monoColor coming in late in the game now.

I have to say; oppositeColor_dynamic looks great too. Only if I had to choose. :p

[Ver Greeneyes]

Actually, that was my plan, yes. If I do add those options (and it seems I will have to) then they will apply to random dithering, too.Nice! I'll have to play around with it on my laptop (which can't do ED) if you do add the options :)

Yes, I'm quite sure that is going to happen... :(To be fair there's already a few options in madVR where we tell people "just use the default unless you see a clear advantage" - I don't think these will be different.

[*Touche*]

To be fair there's already a few options in madVR where we tell people "just use the default unless you see a clear advantage" - I don't think these will be different.

A "Do not touch unless you know what you're doing!" section could be added for these advanced options.

[cyberbeing]

FWIW, I plan on releasing one more test build set in 1-2 days, probably containing 3 (very different) new test builds.
Well just keep in mind, I was also the one few who was strongly bothered by the patterns produced by the initial low noise non-random error diffusion builds, when watching actual videos. If any of the new dithering ideas produces a discernible patterns, you could predict in advance that I personally won't like it. So if I'm hearing you correctly that you're now moving back to deciding on the non-random "low-noise" algorithm, I was always planning to sit out and not influence the decision about a dithering method I'd never use myself.

[madshi]

I've been testing oppositeColor over previous builds and my subjective opinion is that is my favorite so far.

I've tested this time in all my TV's, a 55" LED, a 32" LED, as well as my 15.4" laptop screen at 768p

Tested with blueray; B&W DVD media, several transcoded sample clips in SD and 1080p.

Now, I can only see dithering in 4bit mode. Period. In 8 bit is pretty damn impossible. Even zooming in with the magnifier is indiscernible.
Ok, thanks for your feedback!

I have to say; oppositeColor_dynamic looks great too. Only if I had to choose. :p
Ok, thx... :)

Well just keep in mind, I was also the one few who was strongly bothered by the patterns produced by the initial low noise non-random error diffusion builds, when watching actual videos. If any of the new dithering ideas produces a discernible patterns, you could predict in advance that I personally won't like it. So if I'm hearing you correctly that you're now moving back to deciding on the non-random "low-noise" algorithm, I was always planning to sit out and not influence the decision about a dithering method I'd never use myself.
Only one of the 3 new test builds will be aimed at low-noise.

[iSunrise]

Done. (and marked it as a "major" bug by mistake - sorry)

Turns out it's the combination of NNEDI3 chroma upscaling (not doubling) + video mode deinterlacing that is causing it.
Yes, that is the same bug I reported and madshi had confirmed to be reproducible.

Thanks for the bug report, it sounded like madshi already fixed that (http://forum.doom9.org/showthread.php?p=1665240#post1665240) in his sources. Apparently not.

[leeperry]

A4 is already 0.97.

That would be oppositeColor... :p
Yep I know, but how about a less noisy luma-only 0.97/NL6/A4 monster? Like with a more diffuse noise that would look better in 4bit? I'm sure you could pull this performance without resorting to the blinking RGB noise madness. I kinda got used to that pretty much daily game-changing build euphoria, enough is never enough :p

Actually this isn't the main reason BFI is so nice. It is actually retention on your retina, the light receptors have a change time and flashing black helps with motion blur, like Film, CRT, Plasma, and backlight based BFI LCD. :)
edit: I should have said the only reason BFI is so nice.
Fully agreed, BFI is on a necessary evil basis....and I'm willing to believe that 4:4:4 might somewhat matter for the fluctuating RGB noise mVR builds but +95% of TV's in the world are used to watch movies in 4:2:2.

mVR is conservative enough to still supports XP, it cannot be seriously expected for every mVR user to watch movies in 4:4:4 when pretty much all TV's turn down all their best features(24/50Hz support, REC.709 gamut, BFI, etc).

This said madshi said that 4:4:4 didn't matter much if at all, I guess he'd know :sly:

It's not that I don't notice when something is different, or appreciate subtle improvements in audio or video, but I tend to be more critical, sceptical, scientific, down to earth with it. I have to keep in mind that some things are real and some are placebo. That is why I renamed the builds for testing and later compared the hashes to see which ones I could honestly identify.
Every A/V reviewer is well aware that placebo & nocebo are very real, that the human brain is a very poor measuring tool and all that. The DBT game has been run and madshi already drew his own conclusions.

Well, DLPs and Plasmas are 1-bit native, and by using a high speed temporal(dynamic) dither(at several khz),gives an apparent bitdeph somewhere around 6-7 bits for DLP.
I am sure they are using some spatial dithering in conjunction with the latter.
Strangely I can stand fast enough DLP(many ppl claim that Mitsu are/were some of the best at mastering it) but yeah it's very obviously a dithering feast and I would presume that this is the foremost reason to the grainy appearance of those two technologies. DLP is one nasty trick(and no studies were made regarding its influence on a long time basis, that doesn't seem too healthy IMHO) but Plasma is really a flickering feast.

[Olivier C.]

Hi,

I think the calibration process madVR + dispcalGui is not very consistent because madVR 0.87.4 - madTPG does not work.
I had to replace with another release (november 2013) : MadVR Test Build with madTPG fixes (http://madshi.net/madVRanotherTestBuild1.rar) wich does not contain last (very nice) work on Error Diffusion.

When i extract and overwrite the contents of MadVR Test Build with madTPG fixes (http://madshi.net/madVRanotherTestBuild1.rar) to madVR folder, new features of 0.87.4 are not available (especially Error diffusion, but also Image doubling, etc).
For profiling / calibration, i have to use an old madVR release which does not contain Error Diffusion great and for displaying movies, i want to use the last release 0.87.4 for Error Diffusion great stuff.

May be i am wrong

Thanks a lot

PS : Thanks for all your great stuff madshi :)
And sorry for my poor english (i'm french)

[leeperry]

Hi
BTW, what is it that you meant on HCFR regarding the "ITC Processing" option in the AMD GPU drivers and its effect on mVR exactly?

[YxP]

I may be too late for the party since new test builds are coming, but after a close battle my favourite is opposite colour static. Earlier in this thread someone said the dynamic was "energetic", and I have come to the same conclusion. Static somehow feels more peaceful and relaxing to the eye.

[Olivier C.]

BTW, what is it that you meant on HCFR regarding the "ITC Processing" option in the AMD GPU drivers and its effect on mVR exactly?

Hi leepery ;-)

I realized that disabling "ITC processing" in AMD GPU drivers has a very noticeable effect on first % of luminance (maybe between 0% and 5%). It seems that disabling "ITC processing" crushes blacks but does not lower black level, according to measurements.
In my opinion, the effect of ITC Processing is more general, and does not only affect madVR (picture display, games, etc).

[omarank]

Madshi, would it be too much work to implement Directcompute support for DX10 cards (only for ED algorithms)? I find it too difficult resisting myself from testing these ED builds. I understand though that performance might not be great, but I would be happy using ED even if it works only for SD content.

[magus]

Hey guys I am benchmarking some videos with madvr using a new vbios of my GPU and was wondering what a good way to test the limits of the GPU related to video rendering(and more specifically madvr) would be? Would downloading a very high-bit video and checking Cntrl+J to see framerate and dropped frames be good?


Also I notice some tearing on videos when Aero is disabled when using madvr but other renderers don't seem to have as much a problem. Does anyone else see this or does everyone just leave Aero on?

[cyberbeing]

Yes, that is the same bug I reported and madshi had confirmed to be reproducible.

Thanks for the bug report, it sounded like madshi already fixed that (http://forum.doom9.org/showthread.php?p=1665240#post1665240) in his sources. Apparently not.
Yeah, I remembering mentioning that issue shortly after you did (http://forum.doom9.org/showpost.php?p=1665051&postcount=22200) as well:
NNEDI3 4:2:0->4:4:4 chroma upscaling is broken on my GTX 770 with Video Mode deinterlacing. The screen turns a greenish yellow. Film Mode is unaffected

Just goes to show, madshi forgets things when we don't add entries to his bug tracker.

[iSunrise]

Just goes to show, madshi forgets things when we don't add entries to his bug tracker.
Just to be safe, I recently added a bug report for a similiar DCI-P3 issue, which is probably related to the one you already reported, that triggers every time for me when I enable DCI-P3, whether or not I use a 3DLUT (it´s also not resolution dependent).

[cyberbeing]

It's probably about time I create a bug tracker entry for that uncompressed v210 in avi issue as well.

Edit: Done (http://bugs.madshi.net/view.php?id=160)

[leeperry]

I am benchmarking some videos with madvr using a new vbios of my GPU and was wondering what the best way to test the limits of the GPU related to video rendering(and more specifically madvr) would be?
OpenCL benchmarks would make sense if you care about NNEDI, otherwise the best throughputs in any simple D3D benchmark the better basically.

I notice some tearing on videos when Aero is disabled when using madvr but other renderers don't seem to have as much a problem. Does anyone else see this or does everyone just leave Aero on?
I also get tearing when I disable Aero on W7SP1, FSE takes care of it :)

I realized that disabling "ITC processing" in AMD GPU drivers has a very noticeable effect on first % of luminance (maybe between 0% and 5%). It seems that disabling "ITC processing" crushes blacks but does not lower black level, according to measurements.
Humm interesting, thanks for the info.

Were you able to reproduce this behavior with several displays of different brands? What's the easiest way to find out? Running ArgyllCMS on the first 5%?

[James Freeman]

Why not keep the whole pipeline (Windows and everything) in 16-bit (or better: Floating point) and dither everything at the output of the video card?
Originally Posted by madshi
Well, I'd like that. Not sure why Microsoft didn't introduce something like this when they implemented desktop composition. Would have been a good opportunity.

Anyone know how can we directly contact the manufacturers to clearly convey this message to them?

It'd be VERY nice too see "Dither Output" in the Nvidia/Catalyst Control Panel/Center.
Gone will be the problem of banding after calibration, and 10-bit Pro Monitors & Video Cards ($$$$) would be less important (almost obsolete) even for critical editing like Photoshop.

I am really grateful that MadVR does just that,
Thank you madshi.

[The 8472]

FWIW, I plan on releasing one more test build set in 1-2 days, probably containing 3 (very different) new test builds. One of them is going to be "simple" ordered dithering (based on a modified/improved version of "void-and-cluster"). The 2nd one is going to be a combination of ordered dithering and error diffusion. The 3rd one is going to be an updated low-noise error diffusion algorithm. Straight ordered dithering should perform faster than error diffusion. The combination of ordered dithering and error diffusion could maybe bring another quality improvement to dither dot placement. The low-noise-alternative might be another option to select for users. These test builds will probably only be monoColor, to make testing simpler. This final test build set will decide which algorithms are going to be offered as options in madVR.

Nice. For day-to-day video playback I don't care much for the finer variants of the error diffusion builds, apart from an interest in tinkering with algorithms.

But a possibly faster alternative that allows me to use smooth motion together with dithering and without turning my gfx card into a heating unit is certainly welcome for practical use.

[nevcairiel]

But a possibly faster alternative that allows me to use smooth motion together with dithering and without turning my gfx card into a heating unit is certainly welcome for practical use.

You mean like random dithering? =)

[The 8472]

You mean like random dithering? =)

Something between random dithering and error diffusion. I suggested void and cluster previously since that might provide just that tradeoff.

[madshi]

Thanks for the bug report, it sounded like madshi already fixed that (http://forum.doom9.org/showthread.php?p=1665240#post1665240) in his sources. Apparently not.
Maybe I have, I don't remember.

Yep I know, but how about a less noisy luma-only 0.97/NL6/A4 monster? Like with a more diffuse noise that would look better in 4bit?
A4/NL6 is the way it is. There's no way to make the noise look more diffuse, except by using a totally different algorithm.

I think the calibration process madVR + dispcalGui is not very consistent because madVR 0.87.4 - madTPG does not work.
It does not? Then please create a bug entry in the madVR bug tracker. Link see first post in this thread. Thanks.

I may be too late for the party since new test builds are coming, but after a close battle my favourite is opposite colour static. Earlier in this thread someone said the dynamic was "energetic", and I have come to the same conclusion. Static somehow feels more peaceful and relaxing to the eye.
Ok, thanks for your feedback.

Madshi, would it be too much work to implement Directcompute support for DX10 cards
Yes.

Anyone know how can we directly contact the manufacturers to clearly convey this message to them?
No chance in hell, IMHO.

-----

So here are the FINAL 5 Cylons, ehm no, I mean dither test builds... :D

http://madshi.net/madVRfinal5.rar

I'm tired of hide-and-seek testing, so the releases are properly named. All builds are monoColor static, to make the comparison easier.

(1) "adaptive4" is the same old build as before.
(2) "ed11" is a low-noise alternative. It's based on ed10, but should be quite a big improvement over ed10, IMHO.
(3) "od8", "od16" and "od32" are ordered dithering builds, with different amounts of random added (for good measure).

Unfortunately the combination of ordered dithering and error diffusion didn't work as well as hoped, so I've dropped that idea. I'm not too much of a fan of the ordered dithering look. It looks nice in screenshots (especially ramps and bars), but in real time video playback the dither pattern is so static that it's really annoying to my eyes. But you be the judge. Let me know what you think. If you do like ordered dithering, please let me know which of the 3 od builds you prefer.

FWIW, currently the ordered dithering algorithms run through DirectCompute in a non-optimized manner, so performance is probably not much better than error diffusion. But if you guys like ordered dithering, I can implement it as a pixel shader, which should probably make it similar in speed to random dithering.

[nevcairiel]

FWIW, ordered dithering can have rather bad interaction with cheap screens/TVs (ie. 6-bit panels), when the ordered dithering pattern happens to match the pattern the panel uses to dither the signal, you get an amplified pattern and its truely visible - a primary reason why i switched LAV Video to random dithering by default.

So here are the FINAL 5 Cylons, ehm no, I mean dither test builds... :D

:D

[The 8472]

(1) "adaptive4" is the same old build as before.
(2) "ed11" is a low-noise alternative. It's based on ed10, but should be quite a big improvement over ed10, IMHO.
(3) "od8", "od16" and "od32" are ordered dithering builds, with different amounts of random added (for good measure).


Totally subjective, "not trying to analyze individual pixels", testing with 4bit mode and only 3 of the builds:

ed11 > od8 > od32


Doesn't seem much of a difference in 8bit mode. My display's anti glare layer itself is more noticeable than the least significant bits. And so are specks of dust.

[leeperry]

A4/NL6 is the way it is. There's no way to make the noise look more diffuse, except by using a totally different algorithm.
how about ed11 with 0.97 and/or the same random numbers as A4? I'd really like to try a low noise build that would be as close as possible to A4...retaining the same clarity but with a lower noise level :)

[Buckster]

A4 reminds me (in a good way) of the subtle dithering effect on the Pioneer Plasmas

[leeperry]

It's hard to deny that ed11 looks great, tbh I've been spoiled by the vanishly low noise level of the chroma noise builds and the mid-noise of A4 is slowly but surely getting the best out of me. I guess it would indeed be great to have the abilty to roll two algorithms, the same way you can roll opamps or tubes in audio gear ^^

not sure where the OD builds would rank, I'll recompare them all together tomorrow morning with fresh eyes.

[GREG1292]

ED11>ODC ever so slight

[James Freeman]

OD8, AD4 look dirty, with OD8 the worst and AD4 right after it.
OD16 is a ****, not there and not there.
ED11, OD32 Look great, they are the smoothest.


OD32 - Most even looking dithering pattern, but less smooth for real content (bigger/fluffier dithering pattern compared to ED11).
ED11 - Slightly "less organised" than OD32, but looks way smoother in real content & test patterns (smaller/denser dithering pattern compared to OD32).

ED11 Wins for me.


madshi,
It looks that ED11 is the cleanest looking build so far, even to OppositeColor Dynamic, dare I say (EDIT: for static images (paused)).
Hope the ED11 pattern (instead the current AD4) somehow permeates into the OppositeColor build. ;).


I've created some (dithering weight arrays) with a lot of effort, but I'm not sure if they're as good as they could be.

Let me answer that.
Yes, your algorithms look better than most commercial ones from the beginning of the 20th Century.
Nothing to be doubtful or modest about... Rejoice !!!

[Q-the-STORM]

Let me answer that.
Yes, your algorithms look better than most commercial ones from the beginning of the 20th Century.
Nothing to be doubtful or modest about... Rejoice !!!

I think you mean 21st century... beginning of the 20th century is 1901...

[James Freeman]

I think you mean 21st century... beginning of the 20th century is 1901...

Floyd–Steinberg dithering is an image dithering algorithm first published in 1976 by Robert W. Floyd and Louis Steinberg.

Well not beginning, but still in the 20th century.

I got past junior high you know... :sly:

[leeperry]

So even when I add native 10bit output, you'd still prefer to use 8bit with A4? Because that's what your comment suggests. Error diffusion was never meant as a way to deblock, enhance edges or anything like that
Truth be told, A4 does extremely obvious grain-based EE....it's fun and all but not quite neutral and you made clear that the more visible the grain the lower the bit depth, I also kinda suspect that ED11 is 0.97 but if not I'd love to see what that'd do if even doable. All this said, I haven't gone through the trouble of rolling them again as I just woke up.

[James Freeman]

madVRDirectCompute3 = First DC build, worm pattern.
madVRDirectCompute4 = No worm pattern, more noisy, random dots.
madVRDirectCompute5 = No worm pattern, less noise, no random dots.

madVRedtest (8 ED builds) = 5 & 7 are selected.
madVRedtest2 (9th ED build) modified Floyd-Steinberg, low noise.
madVRedtest3 (10th ED build) even better than 9th, vely low noise (madshi's favorite).

madVRednoisy (4 variants of 5&7) = "Noisy 2" is selected.
madVRnoisyLimited = Improved/Fixed "Noisy 2" build.

madVRnl = NL6 selected (Which is "Noisy 3"), (from here, all the builds are based on this one).
madVRedLinear (NL6 with Linear/Gamma Light) = Gama Light chosen.

madVRedAdaptive = Improved NL6 with Gamma Light.
madVRadaptive2 = Improved edAdaptive.
madVRadaptive3 = Improved Adaptive 2.
madVRadaptive4 = Improved Adaptive 1 (Adaptive 4 chosen best by many).

madVRed4 (based on Adaptive 4) = Mono/Multi Color Static/Dynamic introduced.
madVRopposite = OppositeColor Static/Dynamic builds introduced.

madVRfinal5 = MonoColor Static builds based on "Adaptive 4", "ED11 (Improved old low noise madVRedtest3 (ED10))" & new Ordered Dithering.



ED5: floyd-stein, weight sum 1.00, old random generator
ED7: filter lite, weight sum 0.97, old random generator

noisy 1: floyd-stein, weight sum 1.00, new random generator
noisy 2: floyd-stein, weight sum 0.97, new random generator
noisy 3: filter lite, weight sum 0.97, new random generator
noisy 4: filter lite, weight sum 1.00, new random generator

nl1: filter lite, 0.97, old random generator
nl2: floyd-stein, 0.97, new random generator
nl3: floyd-stein, 1.00, old random generator
nl4: filter lite, 1.00, new random generator
nl5: floyd-stein, 0.97, old random generator
nl6: filter lite, 0.97, new random generator
nl7: floyd-stein, 1.00, new random generator
nl8: filter lite, 1.00, old random generator


Just to remind you what we've gone through. :)

[Shiandow]

I tried comparing them in 4bit, in this case oppositeColor dynamic looked best (by far) and I didn't really like ordered dithering because of the static noise. But after watching in 8bit I found that the temporal noise caused by the dynamic algorithm was far more noticeable than the static noise caused by ordered dithering. So in the end I'm not really sure which I prefer; in 4bit ed11 seemed to be best followed by ad4 and then the ordered dithering builds but in 8bit I wasn't really able to tell the difference between them. They did all give a pretty noticeable improvement over random dithering so if ordered dithering can be made as fast as random dithering then that will probably be my new default.

[6233638]

(1) "adaptive4" is the same old build as before.
(2) "ed11" is a low-noise alternative. It's based on ed10, but should be quite a big improvement over ed10, IMHO.
(3) "od8", "od16" and "od32" are ordered dithering builds, with different amounts of random added (for good measure)From my testing so far, "ed11" seems to look better than "adaptive4" as adaptive4 adds a very strong texture to the image.
However, it did seem like ed11 may add patterns to the image - I'm sure that testing on grayscale ramps etc would make that obvious.

I don't really like the ordered dithering builds at all. Very obvious patterns introduced onto the image.


Comparing any of these builds to "oppositeColor Dynamic" or even the monochrome dynamic build in 4-bit makes me wonder how anyone could prefer builds with static dither patterns.
More than just being able to see patterns, the bit-depth just appears to be so much lower with them.

As for the dynamic builds "adding energy" to the image, in the 4-bit modes, yes it's clearly there, and the dynamic builds do add "energy" across the image fairly uniformly.
The non-dynamic builds still have this "energy" in a lot of places due to film grain/compression artifacts in the source, but it seems to happen a lot more sparsely, and on areas of contrast in the image. (edges etc.)
So rather than most of the image having this "energy" relatively uniformly, only certain parts of the image have it, and really stand out from the rest, rather than blending in.


The only time I could see someone preferring the static builds is if you are watching digital animation which doesn't have any film grain at all, and is comprised of completely flat colors - but I'm still seeing problems around the edges of objects due to compression artifacts, and I still think it looks terrible when anything moves.
I can only think that if you don't find it to be a real problem when anything moves, something about your display is masking it. (slow response time, adding its own dithering etc.)


Edit: If you are going to have options, I feel that they should be "mono/colored dither" and "static/dynamic" dither.

[Shiandow]

Comparing any of these builds to "oppositeColor Dynamic" or even the monochrome dynamic build in 4-bit makes me wonder how anyone could prefer builds with static dither patterns.
More than just being able to see patterns, the bit-depth just appears to be so much lower with them.

I thought so as well but I noticed that the noise from oppositeColor Dynamic actually was noticeable in 8bit whereas the static noise from the non-Dynamic algorithms wasn't.

[James Freeman]

To me, ED11 looks better than OppositeColor Dynamic in terms of apparent Luma Noise (obviously no Chroma noise), EDIT: for static images (paused).
ED11 is far better than any of the previous MonoColor builds.

Its has some minor ordered pattern, but that what makes it denser, smoother and less noisy to my eyes.

Maybe an ED11 based OppositeColor Dynamic build will further improve this current one?

[*Touche*]

Ok, went through new builds.

Panasonic 55ST60 plasma, calibrated
4:4:4

Tested only 8bit and real video content (720p BR rips and 720p HDTV rips), so I have no idea what patterns these form in 4bit or gradients and such.

I wanted to test this as objectively as I could, so I had my GF change the builds on me randomly, without telling me which one she set or if she even changed them. I also renamed them randomly before so even she didn't know which was which.

After multiple runs, this is how I would rank them:

BEST
1. adaptive4
2. od32

It was hard to tell them apart sometimes. Further comparison would be needed to determine which one is clearly better, but it seems that adaptive 4 has a slight edge. Both of them clearly stood out from the rest.

MIDDLE
3. od08

Landed somewhere in the middle, but closer to bad ones than to good ones above.

WORST
4. ed11
5. od16

Clearly the worst with most visible noise/dot patterns.

EDIT: Redid the testing and added some new sample scenes. It was much harder to tell the difference. Maybe my brain locked to different things when I did the first runs in the morning. I am not as rested now. Also did additional testing between adaptive4 and ed11, which most prefer, and ended up switching my preference between them, which goes to show how hard it is to tell between them (for me, on my screen). Don't care either way, as I found oppositeColor_dynamic better more often than not, so I'll keep away from static builds.

[Asmodian]

I decided to also do some testing in 4-bit and 8-bit with these builds, along with oppositColor_dynamic using smooth motion, 1080p24->1440p60 (same settings as my last test).

Testing with 4 bit for the first time; I can actually see differences in 4 bit. It makes me doubt the differences I "see" in 8-bit. :o

4-bit oppositColor_dynamic looks amazing, the noise is almost invisible. :scared:

in 4-bit I like:
oppositColor_dynamic > ed11 > adaptive4 > od32 > od16 > od8
This is mostly based on perceived noise level but the patterns in the od8/16/32 builds were also a negative.

in 8-bit I like:
ed11 >= adaptive4 > oppositColor_dynamic > od32 = od16 = od8
ed11 seems very clean but I would happily watch od32 (probably without noticing).

The only time I could see someone preferring the static builds is if you are watching digital animation which doesn't have any film grain at all, and is comprised of completely flat colors - but I'm still seeing problems around the edges of objects due to compression artifacts, and I still think it looks terrible when anything moves.
I can only think that if you don't find it to be a real problem when anything moves, something about your display is masking it. (slow response time, adding its own dithering etc.)

I am watching digital animation which doesn't have any film grain at all and on a "120Hz" IPS panel so most of that probably applies to me and is why I don't mind static dither very much while enjoying the lower perceived noise. Small pixels probably help too.

Opposite color dynamic sounds like the most "technically" correct way to dither to me, taking into account the HVS. In 8-bit and in my situation I think that the slightly sharper/crisper look from the very subtle noise looks good to me but maybe that shouldn't count as better? After all dither isn't intended to be image enhancing only as accurate and I hope as low noise as possible. I guess this is the scientist point of view.

But if we wanted a way to add noise/grain as an enhancement maybe the dither step is a good place? :)

Edit: Watching for a while in 8-bit using opposite color dynamic and there seems to be something I don't like about it. I cannot say what. I hope it isn't placebo as I think it should be the best. I changed to ed11 and I like it more. :confused:

[James Freeman]

Is the refresh rate of the dithering, is the same rate set for the monitor?