madVR - high quality video renderer (GPU assisted)

[dansrfe]

I wish there was a file generated by hcfr that we could just load into yCMS in madVR.

[6233638]

Originally Posted by dansrfe

I wish there was a file generated by hcfr that we could just load into yCMS in madVR.

Honestly, most auto-generated LUTs aren't that good in my experience.

I would much rather spend the time to tune a LUT to be optimal at specific points by hand - say 25/50/75/100% RGBCMY saturation, and 20-100% Greyscale in 10% steps (going below 20% in yCMS won't be usable until we get >8-bit output from madVR) and have interpolation take care of the rest.

You would think that programs would be best at doing this algorithmically rather than taking the time to hand-tune things, but most of the time tolerances just aren't strict enough or the emphasis is on speed and making a "good enough" profile. It isn't worthwhile if your display drifts a lot, but if your display is good enough, it's definitely worth putting the effort in. (please, no-one suggest ArgyllCMS...)

[MSL_DK]

Originally Posted by 6233638

Video is encoded to only use 16-235, so you probably shouldn't see anything outside that range. Color management systems also have a tendency to clip anything outside the display's gamut.

I do not understand why I get the flashing bars above 235 on red and blue after using ycms. The two colors also looks washed out compared to blue

[leeperry]

Originally Posted by MSL_DK

I do not understand why I get the flashing bars above 235 on red and blue after using ycms. The two colors also looks washed out compared to blue

Yeah, has anyone really put the current yRGB based CMS of mVR through test patterns and so? I haven't read about any real world testing tbh. I've only read ppl complaining and the author replying that he had better things to do than working on yCMS. That's mostly the reason why I asked madshi why he wouldn't allow custom RGBW coordinates to be input into his PS based gamut mapping magic....it looks so beautiful as is

And BTW, yesterday I played around with a UE46ES6100: the 10 points color temp correction works like a charm(and 10bit correction kills PQ-wise, 3800:1 native CR yeah baby!) but the custom primaries/secondaries are a major hit and miss because they work in R/G/B instead of XYZ, huh? Maybe I really missed something but from my limited testing, that was a major hit and miss. And anyway, it exhibited outrageous clouding right in the center of the bottom 2.35 bar so it had to go back and I didn't bother testing it any further...not dropping 900€ for this

[6233638]

Originally Posted by leeperry

Yeah, has anyone really put the current yRGB based CMS of mVR through test patterns and so? I haven't read about any real world testing tbh. I've only read ppl complaining and the author replying that he had better things to do than working on yCMS. That's mostly the reason why I asked madshi why he wouldn't allow custom RGBW coordinates to be input into his PS based gamut mapping magic....it looks so beautiful as is

Unfortunately the yCMS controls are very basic right now, which probably puts its correction on a similar level to these gamut mapping pixel shaders. However, LUT correction can be significantly more powerful. The problem is that the LUT creation tools for yCMS are very basic right now, not the LUT system itself.

Originally Posted by leeperry

the custom primaries/secondaries are a major hit and miss because they work in R/G/B instead of XYZ, huh?

Most color management systems work with RGBCMY HSL controls.

XYZ is a pain to work with. There are some very high end displays which let you just dial in some XYZ co-ordinates and the display shows them accurately, but most displays are not capable of this.

xyY is preferable to XYZ because it's easily readable - xy are your horizontal and vertical axes, and Y is luminance. (brightness)

[leeperry]

Well for each primary/secondary I was able to play around with -50/+50 for R/G/B....how the hell do you get those spot-on REC.709? HCFR talks in xy/ΔE and the green primary of that sammy was drifting like hell

A friend of mine told me that his high-end SONY TV also provides custom gamut capabilities, I've asked him for more informations.

Well yeah, I am well aware that 3DLUT's are the top of the crop in the movie industry(and that each movie goes through a truckload of them, some of them being proprietary and encrypted), but right here right now with yCMS it doesn't do anything (much) better than the PS script (AFAIK), it hasn't been thoroughly tested(I haven't seen a single real world measurement of the results of yRGB based corrections through mVR, only complains) and the author has been AWOL for a while. But anyway, I know it annoys madshi when I raise this matter, so I'll just put my hands where my mouth is

[leeperry]

All this said, I just bought a UE50EH5300(so cheap that it was hard to resist ) that can only do D70 or D59...it's got gain/offset settings for R/G/B though, so I would need a 3DLUT to convert it to D65 and map its gamut.

I guess I'll try my luck with the current yCMS helper in mVR, get it through HCFR test patterns and report back....IIRC it claims to be able to convert both gamut and white point so we'll see about that

[MSL_DK]

^^ I think I'm having a little break with ycms. Whatever I do, the result is weird. But looking forward to hearing about your results :-)

[madshi]

FWIW, a new yCMS build with an important bugfix is expected very soon.

[Xaurus]

madshi, is there any way a donation can help speed up the 3D support?

[leeperry]

Originally Posted by madshi

FWIW, a new yCMS build with an important bugfix is expected very soon.

OK, thanks for the heads-up!

In the meantime I could always use a 8bit Argyll CLUT in order to convert D70 to D65, but there's no point in going through very time consuming and annoying HCFR-based manual tests of yRGB based corrections if there's a bugfix bound to be released soon.....which still raises my question as to whether it would be a lot of trouble for you to allow custom coordinates for PS based gamut mapping in mVR. As we say in french "never put all of your eggs in one basket", and that would do the interim nicely until yCMS corrections will have been proven to be reliable. I haven't checked your gamut rolling stuff in HCFR, but it looks great to me and I'd be glad to oblige with a whole bunch of test results.

[dansrfe]

Is there a guide or book I can read to get better knowledge of calibration and 3DLUTs?

[leeperry]

I like this one.

[pirlouy]

Leeperry, your determination to search always for better quality is really funny and enjoyable to read !
Even if it's off topic, I hope you'll continue to tell us about your research.

[dansrfe]

This issue has existed for a while for me but I never raised an issue about this before. I actually don't know what it is attributed to. Either MPC-HC, madVR, or LAV Filters.

The issue is that when I have subtitles on (in windowed mode) and I drag the playing MPC-HC window across to my other monitor, the subtitles become horizontally stretched and I have to resize the window (just trigger a resize basically so even +- 1 pixel does it) to correct the subtitles width. This happens with SRT or SSA embedded in MKV files.

[6233638]

Originally Posted by leeperry

Well for each primary/secondary I was able to play around with -50/+50 for R/G/B....how the hell do you get those spot-on REC.709?

It doesn't sound like that set has a proper CMS and only offers RGBCMY chroma adjustments. (saturation and brightness combined)

Originally Posted by leeperry

A friend of mine told me that his high-end SONY TV also provides custom gamut capabilities, I've asked him for more informations.

I don't think this is correct.

Originally Posted by leeperry

right here right now with yCMS it doesn't do anything (much) better than the PS script (AFAIK), it hasn't been thoroughly tested(I haven't seen a single real world measurement of the results of yRGB based corrections through mVR, only complains)

I'll try and get measurements off a display this weekend.

Originally Posted by leeperry

All this said, I just bought a UE50EH5300(so cheap that it was hard to resist ) that can only do D70 or D59...it's got gain/offset settings for R/G/B though, so I would need a 3DLUT to convert it to D65 and map its gamut.

There is no such thing as D70 or D59, only D50, D55, D65 and D75.

If you have white balance controls, you should be able to calibrate it to D65 however. I would suggest starting at the 7000K preset as you are generally best to use subtractive adjustments when calibrating.

Originally Posted by leeperry

I guess I'll try my luck with the current yCMS helper in mVR, get it through HCFR test patterns and report back....IIRC it claims to be able to convert both gamut and white point so we'll see about that

Unless your display is really bad after calibration with its own controls, I would suggest only using the RGBW gamut section, and using the "greyscale/gamma" section to only make gamma corrections from 20% up, until we get 10-bit output from madVR and/or better controls for yCMS LUT creation.

To do this, set xy to 0.312713, 0.329016 for all points (preferably 20-100% in 10% steps) and enter your measured Y values, then adjust each point as necessary to hit your desired gamma target, working from 90% downwards. (the calculated LUT will probably only be accurate to within ±0.10 for each point) As your display contrast is <10,000:1, I would suggest that you use the BT.1886 curve adjusted for your contrast ratio. (this spreadsheet should help)

I would also suggest using the display's controls (if it has any) to calibrate gamma as much as possible, prioritizing 0-20%, and yCMS for fine-tuning gamma from 20-100%. yCMS corrections below 20% can introduce visible errors, which is why I would recommend using the display controls for that. (I suspect this is the result of madVR's 8-bit output)

As display gamma controls tend to be relatively coarse, if you can't get a point at the specific brightness required (some controls might jump from 2.35 to 2.45 on a point for example) you are better to calibrate it to the brighter of the two (lower gamma - in this case, 2.35) and correct that via the 3DLUT, than be too dark and brighten it with the LUT.

[leeperry]

Originally Posted by 6233638

If you have white balance controls, you should be able to calibrate it to D65 however. I would suggest starting at the 7000K preset as you are generally best to use subtractive adjustments when calibrating.

Yup thanks for the tips, you nailed it! A few hours after writing this, I fired up my i1d2 and managed to get the D70 preset to reach D65 without the help of Argyll so all is well

I'm not too OCD'ed about gamma tbh, but I'm quite sure that I should be able to get it to behave because I can play around with the backlight brighness, picture brightness, contrast and it's got a gamma setting that goes from -3 to +3 so that should get the job done. Either way, I like a darkish 2.4 cinema-like gamma.

The "10p white balance" feature that Sammy provide in their mid/high range TV's seems pretty pointless because the gain/offset settings for R/G/B can easily reach D65 anyway. Many old Mitsu DLP's only had those settings and that had never been a problem before.

xvYCC also seems utterly pointless at this point as far as I can tell, and concerning the gamut stuff their user manual says "Colour Space: Adjusts the range and variety of colours (the colour space) available to create images"(engrish at work ^^)....I tried to set extreme settings and it did move the primaries drastically in HCFR, but as I said it seemed to be too much of a hit and miss to bother with it. Either you implement it well or you don't, providing sloppy features to look good in the datasheet is not a good business practice.

I dunno what's your take on this, but when I fired up the 46" Hitachi, its sujective "pop effect" really struck me! Everything looked like it was literally popping out of the screen, and that Sammy UE46ES6100 looked even more 3D and made all the other TV's in the shop look flat and dull. The salesman understood that I was a video nutcase so he let me set them all to cinema modes and disable all their noise reduction/dynamic contrast features so I wasn't comparing apples Vs oranges.

And strangely enough, the "pop effect" of those 3/4K'ish:1 CR Hitachi/Sammy TV's is much more striking than on a 20K:1 flat screen CRT.....that same salesman told me that a good TV is not just a panel, it's also a DSP! And I think he nailed it because when playing noisy SD videos, the 6100 was literally cleaning the picture(even though all its NR settings were disabled ) and after watching it for several hours, I came to the conclusion that Sammy being well aware that there's only so much CR you can squeeze off a LCD panel, they had to "cheat" in order to increase the subjective "pop effect". But strangely, it's not silly halo-based EE nor agressive gamma so I don't really know how the hell they do it

But the 46ES6100 looked very very "colored" to me, in the sense that everything kinda looked like a cellshade cartoon and very "digital" when OTOH the 50ES5300 looks far more natural and doesn't try to impress you with unrealistic pictures

I guess they must be doing some smart NR + playing around with the contrast of edges or something, hard to tell..but there sure is something on.

Originally Posted by 6233638

I'll try and get measurements off a display this weekend.

Sounds like a plan

Originally Posted by pirlouy

Leeperry, your determination to search always for better quality is really funny

Well, I'm a proud member of team "enough is never enough"

Anyway, I think I finally won the panel/clouding lottery because this 800€ 50" Sammy really seems to hit the spot: it can reach D65 in 10bit using its internal processing(w/o resorting to a sloppy dynamics-killing 8bit CLUT on the PC), its clouding is very reasonable(only a slight stain in the middle of the left edge), its picture is very sharp and quite contrasty and it's got that killer looking 1cm thick only shell around the picture.....I have to admit that most other TV's look kinda dated to me now.

I couldn't care less about cellphones and tablets, but I can see why Apple are so afraid of Sammy.....and I guess the latter learned their lesson quite well when they commissioned Joe Kane for some of their videoprojectors because even their low-end TV's provide offset/gain settings for R/G/B & 24Hz support, when getting this kind of features from their competition usually raises prices drastically(x2/x3) and forces you to pay for all kinds of features you might very well never use(wifi, artifacts feast frame interpolation, video recorder, smart TV, anaglyph 3D, etc).

At this point, all I need is 1) gamut mapping(but we'll see what 6233638's test results with yCMS will look like) 2) mirroring(which should be possible when mVR will support PS scripts) 3) levels conversion, because I've got many HD videos that went through failtastic processing and end up being 24-210 or stuff like that....I can't use the 8bit ffdshow processing anymore so I'm really SOL on this one....hopefully madshi will pity me/us and allow custom levels filename tags such as [levels=24-210]

Sitting 2 meters away from a 50" TV(with 10bit colorimetry correction) should really allow Jinc3 AR luma upscaling to shine! It's about time I'd fire up the 660 that's been sitting in its box for the past 2 days now

[MSL_DK]

Originally Posted by leeperry

Yup thanks for the tips, you nailed it! A few hours after writing this, I fired up my i1d2 and managed to get the D70 preset to reach D65 without the help of Argyll so all is well

The same here ... but with color clipping (Samsung UE40D5005 /5000/5520)

[vomanci]

Got 2 crashes one after the other, using PotPlayer, trying to right-click in widowed screen. After that it worked finehttp://www.sendspace.com/file/c8wvle.

[MSL_DK]

Originally Posted by madshi

FWIW, a new yCMS build with an important bugfix is expected very soon.

Thanks for the info