Red Dragon: Candle and Match Low Light Tests

[Ergin Ozturk]

Computer generated.

from Phil's chart?

 

[Filip Orlandic]

from Phil's chart?

Nope. X-Rite numbers. You can use it as a target reference.

 

[Phil Holland]

I'll chime in as I got a second before chomping on food things.

Here's the chart with the straight 5600K REDcolor3/REDlogFilm against the computer generated calibration based on the same image.

Link to 16 bit TIF: http://www.artbyphil.com/temp/reduser/phfx_redEpicDragon_ColorChartCalibrated.tif

And one at ISO 250:

Link to 16 bit TIF: http://www.artbyphil.com/temp/reduser/phfx_redEpicDragon_ColorChartCalibrated_ISO250.tif

If you're curious the variance mostly stems from saturation, which is due to the "neutrality/flatness" of REDlogFilm - which is expected. Tonal values are extremely close per chip. Hue Shift isn't particularly large either.

What is important to take from this is that the color is extremely good and painfully close to chart accurate. Under a more calibrated and tuned source, meaning missing the variables of the ambiance of say the sky dome or bounce from the concrete below the chart and other chaos due to the sun or sky pollution, this would produce even better results.

So, long and the short of it.

Color is very, very good on Dragon. Like crazy good. Crazy good straight out of camera. Under all light conditions that I've tested.

And this is without whatever new color science Graeme's been cooking over the last few months. I'm actually curious what "can be done" from here even.

Color is extremely important here. As is the Workable Latitude found in the captured Dynamic Range. As is the Calibration. As is the IR Cut found in the newly designed OLPF tuned to the Dragon Sensor itself.

I should add that this test setup also showcases a realistic extreme Dynamic Range stress test with the white shirt, cast shadows, and shadow box on the chart itself.

 

[Bruce Allen]

If you're curious the variance mostly stems from saturation, which is due to the "neutrality/flatness" of REDlogFilm - which is expected. Tonal values are extremely close per chip. Hue Shift isn't particularly large either.

What is important to take from this is that the color is extremely good and painfully close to chart accurate. Under a more calibrated and tuned source, meaning missing the variables of the ambiance of say the sky dome or bounce from the concrete below the chart and other chaos due to the sun or sky pollution, this would produce even better results.

So, long and the short of it.

Color is very, very good on Dragon. Like crazy good. Crazy good straight out of camera. Under all light conditions that I've tested.

And this is without whatever new color science Graeme's been cooking over the last few months. I'm actually curious what "can be done" from here even.

Color is extremely important here. As is the Workable Latitude found in the captured Dynamic Range. As is the Calibration. As is the IR Cut found in the newly designed OLPF tuned to the Dragon Sensor itself.

Phil: Great test - and I agree.

However you and I both know that a color chip does not have precisely the same spectral characteristics as human skin.

So to be pedantic: all that test conclusively proves is that Dragon can render those color chips well when well-lit.

It IS very encouraging though!

I really look forward to testing on real live humans once they have that new OLPF in place (since IR and deep red is kinda important to skin tone rendering - as long as they didn't go too far and cut off the deep red wavelengths with the IR cut, we'll be OK)... plus the new color science to match.

The thing is: that's going to take a while. And once that is all set, it's going to take a little while longer for the DPs to learn how to get the most out of that sensor (though the less quirks the sensor has, the easier that task is).

This is all good stuff though!

Also, I am kinda chuckling since for years everyone was so angry at me on Reduser for daring to suggest that there could be something wrong with MX skin tones due to sensor issues. I got insulted by SO MANY people who said "It's RAW, Bruce! If you don't like the skin tones it's because you don't know what you're doing!" - haven't had received any apologies once I've been proven right though.

Bruce Allen
www.boacinema.com

 

[Antony Newman]

Bruce,


On MX, I think "you know what you are doing" when "RAW data can be consistently transformed in a 1 to 1 mapping to your desired colour target". The issue with this on MX was red wavelength 'pollution' in the blue sensor readings stopped this being a 1 to 1 transform - necessitating infra red light removed from the scene during filming.


What Phil is stating is that :
+ All light sources do not 'phase the colour - which means that it doesn't have a transform issue, and
+ Out of the box colour doesn't need a transform (other than saturation) to hit the finishing line (simpler workflow).


Armed with a sensor with far lower noise, different RGB filters over each pixel that have far lower colour bleed (ie improved out of band colour rejection) - I guess Graeme is balancing processing power requirements necessary to get this transform dead on a standard target colour curve ... something he would need quite a few dragon sensors from the oven to incorporate variance.


AJ

 

[Phil Holland]

Phil: Great test - and I agree.

However you and I both know that a color chip does not have precisely the same spectral characteristics as human skin.

Thanks Bruce. I will end up re-shooting this exact test and I'll show the other charts I shot under different choices once the new OLPF become available. That will be the "only truth" moving forward, so that's what people will likely be interested in. Same for the Match Test as well.

It's nice to see that Red has developed new pleasing and accurate color straight out of camera under various lighting conditions. That's a big one. The chroma separation and definition is what I'm after.

Lighting skin is a crazy long conversation. These modern sensors in general battle with IR and UV in interesting ways. As you mentioned that IR plays a role in the warmer palettes. Red's done a nice job in how it renders with the original Dragon OLPF, just the occasional bounce IR reflection thing isn't lovely for certain conditions. They got that solved and I'm confident the new color science/calibration on the new OLPF will do the job, but we'll see in a month.

I have a VFX background and instruct at the Gnomon School of Visual Effects on Sundays. One thing that I enjoyed showing students was the effect of different types of lights on skin and the different penetration you get in subsurface scattering, which is a subtle but visible effect. Those subtle hues are sometimes difficult to recreate with certain diffused sources that experience things like droop or softer spread.

I'll update this thread in a month so we have a nice A/B comparison of the new OLPF.

 

[Joel Arvidsson]

Phil do you think the dragon olpf is as strong on dragon as on epic? I like to se a it weaker. Of course i can get as much detail I want but i thinks its unnecesary strong. Would be fun to try one without it and test what kind of scenes that would suffer without it and if it could be fixed in post.

 

[Phil Holland]

Phil do you think the dragon olpf is as strong on dragon as on epic? I like to se a it weaker. Of course i can get as much detail I want but i thinks its unnecesary strong. Would be fun to try one without it and test what kind of scenes that would suffer without it and if it could be fixed in post.

To me, and while there's no easy way to truly measure it without ripping both OLPFs out and comparing them in a 1:1 way, the new Dragon OLPF is indeed weaker. Also Dragon's pixels are smaller. It's seeing just a smidge more fine high frequency detail than Mysterium-X's OLPF. Subtle, but there.