CF Weapon ARRI XT common colourspace comparison - seeking advice

[Stacey Spears]

Would love to get your thoughts.

D'oh! Sorry, completely forgot about this thread. Will try and spend some time thinking about this over the weekend. I am always looking at ways to get the most out of RED footage. e.g. I hope to get a quick HDR grade of a Seattle skyline shot I took last weekend. Trying to figure out the best way to deliver to Dolby since I want to use A.D.D. I also shot with Frame Average 2, so it will be nice and clean in the dark. If I get this done, I will share an HEVC. Its only a couple of 20-second static shots.

That is totally pedantic 'for the sake of this arguement' logic but there are some areas where the Wide Gamut and SGamut3 primaries go outside of the ACES primaries.

Wondeful post, thank you for sharing Sam!

 

[Brett Harrison]

Sam, whenever you post it's educational to the max. Thank you for contributing to the thread! Always nice to hear from the makers of the tools, too.

I want to again apologise for the delay, I am still waiting on both RED and ARRI (who I met with at NAB) to get back to me to confirm they are happy with the settings/general approach - I'm already winding up for another more thorough test, so it's all worth the wait (I hope!).

But from what I've seen it's very revealing to put the two cameras side by side in an all-other-things-being-equal environment. It certainly highlights the importance of debayer settings!

Very soon,
Cheers,
Brett.

 

[Brett Harrison]

Addendum:

The purpose is also to have a look at what the respective manufacturer's 'beauty transforms' do to the other's linear data - because IMO that's been an aggravating factor in comparing the true potential of these two cameras.

Here's the updated workflow for the current draft of the test:

The bottom tiles will form a colour coding for the test results.

 

[Brett Harrison]

Hi All,

Here are the much-delayed results. A combination of waiting to hear back from RED and ARRI about the test, as well as moving interstate to start a new job, are most of why it's so late. I can provide more from the test if people are keen to see. I won't provide the RAWs but I intend to do so for the next test.

Thanks for taking a look!

This test is designed to compare the RAW data of the RED and ARRI cameras in the same colour space, using ARRI, RED and SGO Mistika transforms (the latter is intended as a neutral point of comparison).
I am using the ACES debayer option of both the RED and ARRI SDK, which is an agreed-upon transformation from the respective native camera colour space to a common colour space.

https://en.wikipedia.org/wiki/Academy_Color_Encoding_System

From there I have transformed to three different gamuts and their respective contrast curves:

The Mistika system has a feature named the ‘UniColor’ node which allows a transformation between one proprietary colour space to another using mathematical formulas supplied by RED and ARRI, such that these transforms don’t destroy information because they are using floating point image processing. So for all intents and purposes it’s as if ARRI has adopted RG4/DC2 and visa versa.

From there I’ve applied the respective LUT over the log space, e.g. ARRIWideGamut/LogC to the photometric LUTs available onarri.com, e.g.:

AlexaV3_K1S1_LogC2Video_Rec709_LE_iridas3d.cube
https://www.arri.com/camera/alexa/tools/lut_generator/lut_generator/

DC2/RedLogFilm to RLF2RG4CUBE3D.cube supplied on theRED.com website.

With this method we get to see ARRI through the appearance of the RED transform and RED through the appearance of the ALEXA transform, as well as a completely independant transform for the sake of comparison, using Mistika’s own colour science based on REC-BT.1886 Optical-electro transform:

http://www.sgo.es/mistika/

ITU BT.1886 OETF
http://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.1886-0-201103-I!!PDF-E.pdf

In this test I have applied no manual chroma adjustments; all the ‘grading’ done is at the RAW parameter level and this is simply to make the kelvin and tint the same, and to balance exposure by choosing the appropriate ISO setting. There is one exception to this: for the -4 exposures I have applied a slight gamma and white point adjustment (after ACES, before log space) to ensure that the white point and black point and gamma match, as there is some difference between where everything lands when you select 6400 ISO on a .ari file vs. a .r3d file.

CMOS cameras doesn’t have a curve; it’s just a straight ‘linear light’ representation of reality with no ‘roll off’ or ‘shoulder’, ‘knee’ etc. These characteristics are added after the fact. Linear is a gamma of 1.0, whereas Rec. 709 is 2.4.

The differing sensor qualities (for example size of photosite, optical low pass filter) will mean that there are differing levels of light over time (flux) before a given sensor photosite fills (i.e. clips) but this doesn’t mean roll-off, it just means a varying threshold before clip.

Therefore the point of this exercise is to show the considerable difference a differing curve and colour gamut transform make to the linear information, i.e. what kind of ‘look’ they entail, sensor notwithstanding.

In the test I’m just using the plus 3 and minus 4 sets to show a typical ‘expose to the right’ exposure and a useable low-light exposure.

ARRI offers a range of different shoulder and toe curves to their LUTs. The default (which you see in the name of the LUT I am using) is ‘K1S1’ i.e. Knee 1 and Shoulder 1 (0-3 range, lower = heavier). RED offers 3 contrast curves which are designed to work with the Cineon curve of RedLogFIlm, which is a steeper curve than LogC. Keep this in mind when comparing the graph below, i.e. the linear combination of the RG4 and Cineon will produce a lift, comparatively, to the combination of K1S1 and the LogC curve.

Image sourced from https://community.sony.com/sony/att...ry_for_S-Gamut3Cine_S-Gamut3_S-Log3_V1_00.pdf

ARRI provides a gamut mapping in the form of their LUT generator (i.e. a compression, to avoid clipping) so that the gamut is mapped/scaled ideally to Rec709, whereas RED only provides a contrast curve (1D, luminance/saturation correction), not a fully-blow 3D ‘look’ LUT (I.e. R G B).

To my understanding, DC2 and RC4 etc. are scaled ahead of time for Rec709, hence their display LUTs are just contrast (1D). I may be wrong on this.

https://graphics.stanford.edu/courses/cs178/applets/gamutmapping.html

I’ve turned off all sharpening so it’s as pure an image from the sensor as possible, but I have turned on noise reduction for RED as well as DEB to match what I see going on with ARRI in terms of noise and sharpness.

Mistika uses very recent SDKs from both RED and ARRI.

.R3D settings:

Kelvin: 3200
Tint: 0
Denoise: maximum
OLPF compensation: off
D.E.B.: on
Detail: low
Bit Depth: 16
Colour space: ACES colour

.ARI settings:

Kelvin: 3200
Tint: 0
Decoder: ARRI
Colour Space: ACES
ARRI debayer: ADA 3
Filter Chroma: none
Sharpness: 0.0

The ISO rating varies between the +3 and -4 exposure sets.

Let me stress that this isn’t a latitude test and the varying exposures are meant to offer a variety of appearances to compare under different colour spaces.

The images are labeled using the colour coded boxes from the flow chart above as well as the slate in-shot for camera info.

 

[Brett Harrison]

Splitting post due to length limit

Tiled cross-section of 0 exposure with Mistika transform to BT.1886

+3 exposures​

-4 exposures

​

​

 

[Martin Stevens]

Thanks for posting this, Brett.

 

[Michael Lindsay]

Thanks for posting Brett... Interesting work

Ps

With such different colour balance between the 2 shots (Irespective of processing) would there be any value to you in shifting them a little closer?

For me the Arri looks like it was shot under tungsten light and red does not (beyond the noise which tells me this).... I would slightly prefer them to have the same colour grey. If white and black fall away a little so be it.

Just a thought..

 

[Brett Harrison]

Thanks for posting Brett... Interesting work

Ps

With such different colour balance between the 2 shots (Irespective of processing) would there be any value to you in shifting them a little closer?

For me the Arri looks like it was shot under tungsten light and red does not (beyond the noise which tells me this).... I would slightly prefer them to have the same colour grey. If white and black fall away a little so be it.

Just a thought..

Hi Michael,

I've stuck to the results of matching the kelvin and tint settings between cameras, and making no chromatic changes whatsoever. It's interesting to note the added warmth of ARRI where RED seems to be a more neutral colour balance to the illuminant.

 

[Michael Lindsay]

Hi Brett

I understand you not wanting to get into the vagaries of subjectively matching skin but since you have a grey ref in the picture it would have been really interesting to see where the rest of the image fell if the greyscale matched. Since silicon is less sensitive to blue light... shooting under tungsten light and rendering a cool image is hard to do. Simply speaking the Alexa is doing less work. Also with a marched grey you may have felt less inclined to use DEB?

Thanks again for the experiment... Still very informative.

Michael

 

[Brett Harrison]

Hi Michael,

Next time around I intend to shoot a DSC Labs colour chart and derive a 3x3 matrix for matching the cameras to see what that looks like, but for this test I wanted to expose what's on offer out of the box - which makes me regret the slight contrast adjustment (which will have some slight chrominance shift).

Hi Brett

I understand you not wanting to get into the vagaries of subjectively matching skin but since you have a grey ref in the picture it would have been really interesting to see where the rest of the image fell if the greyscale matched. Since silicon is less sensitive to blue light... shooting under tungsten light and rendering a cool image is hard to do. Simply speaking the Alexa is doing less work. Also with a marched grey you may have felt less inclined to use DEB?

Thanks again for the experiment... Still very informative.

Michael

 

[Michael Lindsay]

Hi Michael,

Next time around I intend to shoot a DSC Labs colour chart and derive a 3x3 matrix for matching the cameras to see what that looks like, but for this test I wanted to expose what's on offer out of the box - which makes me regret the slight contrast adjustment (which will have some slight chrominance shift).

Completely understand...;-)