madVR - high quality video renderer (GPU assisted)

[James Freeman]

I've disabled the LL in the settings and Enabled Asmodian's shader to find the perfect value.
1.90 is much closer than 2.20 (in my experience).

But what really bothers us is the tail in the black region.

[turbojet]

What could cause overlay to be much darker and saturated then windowed and fse on a tv?

Overlay looks fine with the same setup on a monitor.

[cyberbeing]

Originally Posted by madshi

@cyberbeing, here's another try. Which dithered image looks nearer to the original?

|- gamma -|- original -|- linear -|

Yes, it's a very low bitdepth (2bit) again, but this really does help to prove the point.

Edit: Please make sure you watch these at 100%.

I concur that Gamma light looks closer to the original there. In some ways I actually prefer it over the original since it enhances shadow detail. Linear light is way too dark, crushed shadow detail, and all those green color tones are thrown way off their original hues and saturations.

[James Freeman]

Originally Posted by turbojet

What could cause overlay to be much darker and saturated then windowed and fse on a tv?
Overlay looks fine with the same setup on a monitor.

darker and saturated?
Sounds like windows is changing to 16-bit in overlay mode with the TV.
Or an HDMI range problem.

[Ver Greeneyes]

Originally Posted by madshi

True, but this again just poses the question which transfer function to use, and not whether gamma dithering would be better. The point I'm trying to make is that linear light dithering is generally more accurate than gamma light dithering. Which transfer function should be used for linear light dithering is a topic of its own.

Yeah, I honestly don't know what to make of it. It's clear that linear light loses some of the detail near black - but is this because it's objectively wrong in some sense or is it just that this amount of detail is the best that can be achieved in that bit depth without brightening the image as gamma light does? I can say that linear light dithering is visibly different near black even in 8-bit on my monitor (probably most noticeable on a gradient test pattern though), but my calibration might just be emphasizing differences near black beyond what the encoding intended.

FWIW, the transfer function I calibrated my monitor to is Rec.709 but with the viewing conditions adjusted to those of sRGB (dispcal -g709 -a64). This should be a close match for the 'intended' viewing conditions, and should also be similar to BT.1886 (which doesn't have the linear segment, but does start out more steeply the higher your monitor's black point is). With linear light dithering it seems I would be better off calibrating to x^1/0.45 with an output offset - although I wonder whether enabling linear light dithering during calibration (is ordered dithering affected?) would automatically compensate.

[James Freeman]

Originally Posted by cyberbeing

I concur that Gamma light looks closer to the original there. It some ways I actually prefer it over the original since it enhances shadow detail. Linear light is way too dark with crushed shadow detail.

madshi, try 1.90 (1/0.52) much better.

Some parts look a little dark others look a little bright.

[6233638]

Originally Posted by madshi

I'm not sure. In any case the choice of the transfer curve will influence how bright/dark the final image will get. So I don't think your blurring & zooming trick properly shows whether the linear light dithered image is darker than the original or not. In real life this depends mostly on whether the transfer function used by madVR for linear light dithering matches the display calibration. If it does, the overall brightness will be perfect, otherwise it will be slightly off (too dark or too bright).

You're right - I just tried it after assigning different profiles to the images prior to the linear light conversion step and it is using whichever profile is assigned to the image. (sRGB for untagged images)
After using a profile with a 2.2 gamma, it does still look too dark near black.

However, viewing on a retina macbook at 100% size from a distance, to make it as small as possible, the linear light image actually seems too bright near black.

If possible, I would like to see the results from 0.51, the BT.709 curve, and maybe 2.4 gamma though.

[madshi]

Originally Posted by James Freeman

I've disabled the LL in the settings and Enabled Asmodian's shader to find the perfect value.
1.90 is much closer than 2.20 (in my experience).

You know what this *really* means? It means that your display is probably calibrated to about 1.90.

Originally Posted by cyberbeing

Honestly, applying a calibration directly to an 8bit 3DLUT behaves horribly with Argyll CMS. Using a 3DLUT without a GPU gamma ramp doesn't allow me to target the custom gamma response I desire.

The trick is to do a 1D gamma ramp calibration first via ArgyllCMS, but then to "merge" the final 1D gamma ramps into the 3dlut. That's what the "-a" switch (IIRC) is for instead of "-H". Basically the end result is similar to what you probably get now in Overlay mode - except that letting ArgyllCMS do all this will end up being even more exact, because ArgyllCMS will do the math in 64bit floating point and madVR can then perform the whole thing in one step (which means less LUT interpolations).

Originally Posted by turbojet

What could cause overlay to be much darker and saturated then windowed and fse on a tv?

Probably Overlay outputs 0-255 and windowed and FSE outputs 16-235. Just a guess, though.

Originally Posted by cyberbeing

I concur that Gamma light looks closer to the original there. It some ways I actually prefer it over the original since it enhances shadow detail. Linear light is way too dark with crushed shadow detail.

"concur" suggests that you agree with the others. But everybody else changed their opinion after I pointed out to them that they accidently viewed the images zoomed even though I explicitly stated that you need to take extra care to view them at 100%. If you don't view them at 100% you'll get totally incorrect results because the browser scales them in gamma light, and scaling heavily dithered images in gamma light produces totally incorrect results. Did you view them at 100%?

Please don't concentrate on shadow detail, only. Of course brightening up dark areas will improve shadow detail. But the key thing to look at is if the overall image looks correct (in terms of brightness and gamma response). Shadow detail at 2bit dithering of course is expected to suffer. FWIW, watching this in motion with 2bit dithering (especially using the "dynamic" option) greatly improves shadow detail. It's much better in motion than it appears on the screenshots.

Originally Posted by Ver Greeneyes

Yeah, I honestly don't know what to make of it. It's clear that linear light loses some of the detail near black - but is this because it's objectively wrong in some sense or is it just that this amount of detail is the best that can be achieved in that bit depth without brightening the image as gamma light does? I can say that linear light dithering is visibly different near black even in 8-bit on my monitor (probably most noticeable on a gradient test pattern though), but my calibration might just be emphasizing differences near black beyond what the encoding intended.

Hmmmm...

-------

Here's the Avatar dithering comparison again, with added images using different transfer functions for linear light dithering:

|- gamma -|- original -|- linear (pure power 1/0.45) -|- linear (sRGB 2.4) -|- linear (BT.709 1/0.45) -|

Again, make sure you watch these at 100%. Having your browser zoom these will totally screw up everything.

The key point to take from this screenshot comparison is that dithering in gamma light is not correct. I hope everybody can agree with that now? Which image looks best to you will directly depend on how your display is calibrated. The image created with the transfer function nearest to your display calibration should look nearest to the original 8bit image to your eyes.

[James Freeman]

Originally Posted by madshi

Here's the Avatar dithering comparison again, with added images using different transfer functions for linear light dithering:

|- gamma -|- original -|- linear (pure power 1/0.45) -|- linear (sRGB 2.4) -|- linear (BT.709 1/0.45) -|

sRGB 2.4 looks closest (much better than 2.2), still just a tad dark.
Gamma Light is out of the question.

Can you edit and add sRGB 2.2 too?

I can de-focus my retina at will for these tests, so no wall climbing for me...

[6233638]

I think what is throwing people off, is that the images dithered in gamma light do preserve the most amount of shadow detail, and make that all the more obvious by brightening it up.
From that selection, the pure power 1/0.45 is definitely closest to the original - though it does still look a bit bright when I compare it on the macbook.

The thing is, you really have to view it on a small screen from a distance so that you are focused on the image and not the dithering.
When I get too close to my screen, my preference changes.

[turbojet]

Originally Posted by madshi

Probably Overlay outputs 0-255 and windowed and FSE outputs 16-235. Just a guess, though.

You are probably right. If I set tv to 16-235 in madvr it looks better but still a bit dark but then I get crushed blacks in windowed/fse. nvidia cp setting had no effect, is there any way to fix this here or is it a madvr issue?

[James Freeman]

Originally Posted by 6233638

When I get too close to my screen, my preference changes.

Yep,
What will translate better in 8-bit, Sharp or Defocused?

[iSunrise]

Originally Posted by madshi

Here's the Avatar dithering comparison again, with added images using different transfer functions for linear light dithering:

|- gamma -|- original -|- linear (pure power 1/0.45) -|- linear (sRGB 2.4) -|- linear (BT.709 1/0.45) -|

The key point to take from this screenshot comparison is that dithering in gamma light is not correct. I hope everybody can agree with that now? Which image looks best to you will directly depend on how your display is calibrated. The image created with the transfer function nearest to your display calibration should look nearest to the original 8bit image to your eyes.

Thanks a lot for these. They are perfect examples where you can see that linear light does a huge difference (at least for me).

If I view these with a sRGB calibration target, the "linear (sRGB 2.4)" is simply perfect. No brightness change whatsoever, just a perfect reproduction of the original with the few bits that are available (extremely impressive to look at).

In comparison, the "linear (pure power 1/0.45)" loses a lot of detail again and is darker, but this seems to be expected, cause I am calibrated to sRGB.

So I would say that we finally nailed it. Now you just need to include this into madVR and I think we should all be satisfied.

[6233638]

Originally Posted by James Freeman

Yep,
What will translate better in 8-bit, Sharp or Defocused?

I think people will pick the sRGB image if they are too close to their display.
The 1/0.45 image is the one which is actually closest to the source though. (yet it is still too bright near black)

This is why I'm doing testing at 100% size on a retina macbook. The image is only about 6x4" in size, so I can make the comparison without defocusing my eyes.

[iSunrise]

Originally Posted by 6233638

I think people will pick the sRGB image if they are too close to their display.
The 1/0.45 image is the one which is actually closest to the source though. (yet it is still too bright near black)

I actually tested both to be sure. Very close to the display and at 3-4 meters distance from it.

And for both of these cases, the sRGB 2.4 example is a very tiny amount brighter in darker shades (almost indetectable), while the 1/0.45 image is _noticeably_ darker.

That´s on a display with perfect reproduction of all 0-255 shades of grey and sRGB calibration. Results may differ if the display is not able to show really dark shades or if someone chose a customized sRGB target, though.

[James Freeman]

Originally Posted by iSunrise

If I view these with a sRGB calibration target, the "linear (sRGB 2.4)" is simply perfect.
In comparison, the "linear (pure power 1/0.45)" loses a lot of detail again and is darker, but this seems to be expected, cause I am calibrated to sRGB.

Agreed.
sRGB 2.4 definitely better than Pure Power 2.2.

Originally Posted by 6233638

I think people will pick the sRGB image if they are too close to their display.
The 1/0.45 image is the one which is actually closest to the source though. (yet it is still too bright near black)
This is why I'm doing testing at 100% size on a retina macbook. The image is only about 6x4" in size, so I can make the comparison without defocusing my eyes.

I think defocustin is a big part of this test because we are testing how the eye sees light not detail.

Originally Posted by iSunrise

And for both of these cases, the sRGB 2.4 example is a very tiny amount brighter (almost indetectable),

For me its brighter when focused and darker when defocused... What will equate better at higher bit depths?
Still much better than Power Curve.

[cyberbeing]

Originally Posted by madshi

The trick is to do a 1D gamma ramp calibration first via ArgyllCMS, but then to "merge" the final 1D gamma ramps into the 3dlut. That's what the "-a" switch (IIRC) is for instead of "-H".

That's not a trick, that "-a" switch is what I was talking about which always gave horrible results every time I tested it. I've given up on it.

Originally Posted by madshi

Basically the end result is similar to what you probably get now in Overlay mode - except that letting ArgyllCMS do all this will end up being even more exact, because ArgyllCMS will do the math in 64bit floating point and madVR can then perform the whole thing in one step (which means less LUT interpolations).

I strongly disagree. madVR loading the Gamma ramp in 16bit via Overlay yields superior results to Argyll CMS's attempts at merging it into an 8bit 3DLUT.

madVR loading gamma ramp via shaders = expected results
Gamma ramp loaded into GPU = expected results
Dispcal Calibration merged into 3DLUT with "-a" switch = unexpected results
3DLUT created with gamma which doesn't match your display = unexpected results

Originally Posted by madshi

Did you view them at 100%?

Please don't concentrate on shadow detail, only. Of course brightening up dark areas will improve shadow detail. But the key thing to look at is if the overall image looks correct (in terms of brightness and gamma response). Shadow at 2bit dithering of course is expected to suffer.

Yes, I viewed them at 100%, outside my browser. The linear light version looks extremely wrong. The colors tones are no longer even close to the original, as I stated before. All the colors in the gamma light version look similar to the hues and saturation of the original. There is really no contest there.

Originally Posted by madshi

Here's the Avatar dithering comparison again, with added images using different transfer functions for linear light dithering:

|- gamma -|- original -|- linear (pure power 1/0.45) -|- linear (sRGB 2.4) -|- linear (BT.709 1/0.45) -|

Again, make sure you what these at 100%. Having your browser zoom these will totally screw up everything.

The key point to take from this screenshot comparison is that dithering in gamma light is not correct. I hope everybody can agree with that now? Which image looks best to you will directly depend on how your display is calibrated. The image created with the transfer function nearest to your display calibration should look nearest to the original 8bit image to your eyes.

I still find Gamma wins in these new comparisons, as it's closest to hue and saturation of the original. The only image from the linear ones which is not horrible seems to be "linear (BT.709 1/0.45)", but it's still way off in its color tones.

Gamma >> linear (BT.709 1/0.45) >>> linear (sRGB 2.4) >>>>>>> linear (pure power 1/0.45)

P.S. It may be helpful if you did your next comparison with monoColor dithering, to ensure the oppositeColor dithering isn't at fault for these color issues. Also try including a (BT.709 2.4) & (BT.709 2.6) image. One of those curves may come decently close to my calibration, depending how you scale the curves vs Argyll ambient scaling.

[JonnyRedHed]

Originally Posted by madshi

Ok, there are still 2 things you could do to help find the cause of this problem:

(1) In the situation where playback doesn't work, anymore, create a freeze report (Ctrl+Alt+Shift+Break), and mail it to me. The freeze report should appear as a text file on your desktop. It might take a couple of seconds after you pressed those keys.

(2) Activate the madVR debug mode by double clicking "activate debug mode.bat". After doing this madVR will log everything down to a big log file on your desktop. Be careful, the file will grow very large very quickly. If it takes too long to reproduce the problem, it might make sense to stop zoomPlayer, delete the log file and try again. Otherwise the log file might grow too big (several GBs). Once the problem has occurred, stop zoomPlayer right away. Then zip up the log file (it will compress very well) and send it to me. If the log file is too big to send, you can delete everything until the last maybe 200MB of text. Of course you'll need a tool which can do that. E.g. the freeware HxD should be able to. Zipped up it should be less than 10MB, I think.

Thanks!

I've emailed the log and freeze reports. Log (zipped) was about 14mb.

[madshi]

Originally Posted by turbojet

You are probably right. If I set tv to 16-235 in madvr it looks better but still a bit dark but then I get crushed blacks in windowed/fse. nvidia cp setting had no effect, is there any way to fix this here or is it a madvr issue?

The easiest solution would probably be to use the madLevelsTweaker.exe shipping with madVR to force your GPU to output 0-255 in all modes.

Originally Posted by iSunrise

Now you just need to include this into madVR and I think we should all be satisfied.

Which is not what I wanted...

Originally Posted by cyberbeing

That's not a trick, that "-a" switch is what I was talking about which always gave horrible results every time I tested it. I've given up on it.

I strongly disagree. madVR loading the Gamma ramp in 16bit via Overlay yields superior results to Argyll CMS's attempts at merging it into an 8bit 3DLUT.

Ok. When did you last test this? Did madTPG already exist then? If not, it might be worth a try testing this again, just to double check the problem still occurs. madTPG changes things a bit because the whole calibration (even 1D lut creation) runs through madVR this way, using high bitdepth rendering. Even when using ArgyllCMS to create GPU gamma ramps, when using madTPG the GPU gamma ramps are never actually used.

Originally Posted by cyberbeing

Yes, I viewed them at 100%, outside my browser. The linear light version looks extremely wrong. The colors tones are no longer even close to the original, as I stated before. All the colors in the gamma light version look similar to the hues and saturation of the original. There is really no contest there.

I still find Gamma wins in these new comparisons, as it's closest to hue and saturation of the original. The only image from the linear ones which is not horrible seems to be "linear (BT.709 1/0.45)", but it's still way off in its color tones.

You're looking at everything (shadow detail, hues, saturations) except what I wanted you to look at , namely overall brightness and gamma response. But I guess it's my fault for using oppositeColor which I knew you hated.

Maybe you can test this yourself with an appropriate movie on your display? Using monoColor, in 2-4bit, then in motion toggle linear light dithering on/off, and also switch between 2-4bit and 8bit. I think you will find that linear light dithering is overall nearer to the original, even if it might not be perfect. At least that's what everybody seems to agree on atm. But then, most users also liked oppositeColor, which you don't like. And leeperry is on your side this time, too. So maybe it is a matter of taste again. But FWIW, I'm sure that linear light dithering produces a more correct overall image brightness and gamma response compared to gamma light dithering.

Originally Posted by JonnyRedHed

I've emailed the log and freeze reports. Log (zipped) was about 14mb.

Yeah, thanks, got them. Will look at them later today.

[nevcairiel]

Originally Posted by madshi

Here's the Avatar dithering comparison again, with added images using different transfer functions for linear light dithering:

|- gamma -|- original -|- linear (pure power 1/0.45) -|- linear (sRGB 2.4) -|- linear (BT.709 1/0.45) -|

Again, make sure you watch these at 100%. Having your browser zoom these will totally screw up everything.

Watching on my PC screen, BT.709 looks the closest watched from a distance for me, although only Gamma has a very obvious difference, for the others its just minimal.