Converting Red to how Alexa handles saturation

[Christoffer Glans]

In direct attempt to mimic Alexa, one would be additionally limiting and conditioning sensor performance and stressing the signal. And shooting a moving target. That is neither necessary nor a proper route to achieve some of the image properties present in Alexa imagery.

In other words - that thing you like can be achieved just not that way you assume.

What does a moving target have to do with the colors and saturation levels?
But I'm interested to hear your method to achieve the look and feel of Alexa.

 

[Brett Harrison]

Here's the other side of the transform from linear to display, i.e. RED Weapon image with various proprietary transforms on top, colour coded on the corner, RED means RED, silver means Alexa XT open gate, and the borders mean the transform - green = Mistika's own OETF transform to rec709.

So this video literally transforms from ACES linear debayer to Alexa handling of gamut mapping and contrast curve for Rec709:

 

[Hrvoje Simic]

What does a moving target have to do with the colors and saturation levels?

Moving target is another camera behavior.
Which would also be a double conditioning for a signal.

But I'm interested to hear your method to achieve the look and feel of Alexa.

The goal is not achieving the look and feel of Alexa. For the reasons already mentioned in this thread. The goal is optimizing the signal for the purpose. Which consequently brings some image aspects which analytical aspect of the mind tends to associate with the first equivalent source of the sensation.

Method is highly complex and in it's large part pioneering. What can be said is that involves in depth sensor analysis and signal optimization and shares some principles used in designing IDTs, through approach termed as "scene referred".

 

[jussi rovanpera]

Y'all pixel peepers seen this? http://www.yedlin.net/OnColorScience/

 

[Christoffer Glans]

The goal is optimizing the signal for the purpose. Which consequently brings some image aspects which analytical aspect of the mind associates with the first source of the visual sensation.

Well, if the purpose is to reach the color look of an Alexa shot film (I'm not talking about over-graded films which looks the same on any camera in the end), then that is the purpose here. There's a cleanness and depth to the colors with an Alexa basic look which I'm not seeing on Dragon. Dragon is much flatter and in tungsten gets very muddy, especially in yellows and reds.

What I'm trying to get is how a transform in it's deepest form couldn't transform the color and luma values of an image. If all the information is stuffed into the image, if a pixel is a certain value of a color and luminance and you change that pixel to another value of the same color and luminance, wouldn't that add up to a perfect match and if the changed value is recorded and all colors and luma values are present in one single image, it should be possible to come as close as you could ever get for a transform/LUT.

 

[abreu-canedo]

Well, if the purpose is to reach the color look of an Alexa shot film (I'm not talking about over-graded films which looks the same on any camera in the end), then that is the purpose here. There's a cleanness and depth to the colors with an Alexa basic look which I'm not seeing on Dragon. Dragon is much flatter and in tungsten gets very muddy, especially in yellows and reds.

What I'm trying to get is how a transform in it's deepest form couldn't transform the color and luma values of an image. If all the information is stuffed into the image, if a pixel is a certain value of a color and luminance and you change that pixel to another value of the same color and luminance, wouldn't that add up to a perfect match and if the changed value is recorded and all colors and luma values are present in one single image, it should be possible to come as close as you could ever get for a transform/LUT.

Not really as simple because we are dealing with moving images. You could take a single frame and say, take pixel 1 and make it color X. But for it to work across all frames, you have to say: take all pixels of this color Y and make it color X. But the thing about that is that not all cameras take in the same information. The dyes in the CFA make a huge difference in the purity of the RGB signal captured by the sensor and thus, the signal processing to filter out chromatic overlap between pixels in a bayer pattern. The more pure the Dyes are, the less overlap. The more pure the dye, the wider the color gamut a camera is able to achieve. Not only that, but certain cameras are able to capture different colors and variations of different colors differently.

The reason why I say all of this is because my theory is that A: if red only sees 600 types of cyans (a very hard color for digital sensors based on purity of dyes on CFA), you cannot tell any color software to turn those pixels into 4000 different types of cyans. Arri has been making telecines for many decades and their CFA game is off the hook, son! Haha. The complexity of tonality in cyans and yelows, for example, is slightly ahead of Red (albeit I have not yet used Helium). And B: the signal process that happens to filter out spectral overlap to produce pure RGB from sensor data is amplifying signals in complex ways which affect the signal to noise ratio, which then affects the "grain" structure of different color and shades differently. This affects the way that colors roll in and out of each other. This is not something you can easily create an algorhythm for. Plus, I am not sure if Red does any NR in camera to manage this behavior, which would further affect color separation, tonal complexity, etc... You can go from more colors to less colors, but cannot go from less colors to more colors...at least not in an algorhythmic way. You would literally have to alter every frame individually and do so with an insane color accuracy.

In my opinion, the cost of the time spent trying to Alexafy Dragon footage is more than the cost of renting an Alexa. Of course, you are able to theoretically do lower level things such as cap saturation above 50% luminance, cool your reds a bit, and limit the dynamic range of your scene to accomodate a comparable highlight roll-off in post, all of which would make for a great Red image. But I'm afraid the differences between Alexa and Dragon is more than what a transform can account for.

 

[Brett Harrison]

The reason why I say all of this is because my theory is that A: if red only sees 600 types of cyans (a very hard color for digital sensors based on purity of dyes on CFA), you cannot tell any color software to turn those pixels into 4000 different types of cyans.

This process is called gamut extension and is a very active area or research, particularly for the conversion of old media to wide gamut colour spaces for HDR formats.

I believe in a modern debayer algorithm there's thousands of lines of code. There's a lot going on to get from colour filter arrays including interpolation.

 

[abreu-canedo]

This process is called gamut extension and is a very active area or research, particularly for the conversion of old media to wide gamut colour spaces for HDR formats.

I believe in a modern debayer algorithm there's thousands of lines of code. There's a lot going on to get from colour filter arrays including interpolation.

I suppose I should have included "yet" in my statement. But in fairness, I did say, any color software, which implies "currently". Maybe it is implied? I don't know, cut me some slack, here.

OP is speaking about the present moment. But you bring up a fair point. I am sure through machine learning and AI the complexity of color interpolation algorhythms will become infinitely capable. However, we are not currently there. All those lines of code are there to deal with the result of the CFA and what it provides the sensor. CFA is very important. At least at the moment.

 

[Patrick Tresch]

Y'all pixel peepers seen this? http://www.yedlin.net/OnColorScience/

It's a lot of generic talk good for some beginners.

Btw "Red's Weapon is 8k" is wrong. Helium or Dragon VV is 8k not Weapon.

 

[Brett Harrison]

It's a lot of generic talk good for some beginners.

Btw "Red's Weapon is 8k" is wrong. Helium or Dragon VV is 8k not Weapon.

Actually the stuff that Yedlin is doing goes way beyond the usual tools in grading systems; colour scientists know whats up - that's why he recommends we all go chat to them and engage with their skills and knowledge to start developing the potential of digital image processing beyond the traditional lift gamma gain mode.

One thing you can do right away without the study of dizzying math is to collaborate with a color scientist at your regular post house. Properly trained color scientists are under-appreciated, and I bet they'd love to collaborate with the type of clients who appreciate their unique expertise. And if your post house doesn’t have a color scientist, push them towards more rigorous methods. If we're going for a film look, we want a comprehensive perceptual model, not empty lip service. Just saying the words "color science" or "film print emulation" doesn't mean you've done the work. As filmmakers, a powerful step forward is merely to recognize that there exists a special skill set unique to a proper color scientist that has little to no overlap with the important but different expertise of a colorist, an on-set digital imaging technician, or a workstation engineer.

 

[Hrvoje Simic]

Well, if the purpose is to reach the color look of an Alexa shot film
...

You shoot on Alexa.

There's a cleanness and depth to the colors with an Alexa basic look which I'm not seeing on Dragon. Dragon is much flatter and in tungsten gets very muddy, especially in yellows and reds.

Since misconception is a drawback and patience a virtue I'll try with the .3rd attempt:
Optimizing camera's image response does not necessitate (or benefit from) attempting to copy a look from another camera.

What I'm trying to get is how a transform in it's deepest form couldn't transform the color and luma values of an image.

Because:

a) transform has limitations
b) image signal has limitations
c) values tend to vary

 

[Patrick Tresch]

Actually the stuff that Yedlin is doing goes way beyond the usual tools in grading systems; colour scientists know whats up - that's why he recommends we all go chat to them and engage with their skills and knowledge to start developing the potential of digital image processing beyond the traditional lift gamma gain mode.

That's exaclty what we do since 2008 with Graeme Nattress. And it's also the believe of R3D since 2007 to give people a digital negative and not a baked in out of the box look. R3D is unsuited for the ingnorants but a powerfull tool for the photographer to acchieve his image.

Pat

 

[Patrick Tresch]

There's a cleanness and depth to the colors with an Alexa basic look which I'm not seeing on Dragon. Dragon is much flatter and in tungsten gets very muddy, especially in yellows and reds.

Strange your findings are exactly the contrary of what Vincent Mathias AFC found out :

"J’ai trouvé que l’Epic Dragon restituait mieux les nuances dans le mélange de lumières chaudes et froides et marquait plus les écarts de température de couleur. Plus généralement, j’aime aussi les plus grandes nuances dans les carnations avec la RED.

...

Autre différence importante, c’est la restitution des lumières sodium orange, dans les rues, de nuit. Alors que l’Alexa les restitue orange avec une tendance magenta, la RED donne un résultat plus jaune vert que je préfère largement, plus proche de ce que je vois à l’œil."

http://www.afcinema.com/RED-Epic-Dragon-retour-d-experience.html

 

[Anton Esteban]

I wanted to add something to this conversation. I think the bigger question is not how to make RED an ALEXA but how to make the overall image quality and color of footage shot on RED's better. Now, everyone is absolutely correct in this discussion when they say that a RED will never be an ALEXA no matter what fancy transforms to get some kind of magical translation. First of all, I think the ALEXA camera is not the be all end all of image quality, there is room for great improvement. And, honestly, I think the HELIUM sensor or what 8k brings to the table is a huge advancement. 8K is not only a sharpness improvement, or something that helps image reframing; 8K can allow for more detail in the transition between light and dark as well as in the presentation of the gradation of color since there are, literally, more pixels to show this change and therefore give a much more nuanced and vivid image. What we need in this discussion is how to get RED or another third party like Company 3 to come up with a number of different transforms to get better image quality out of RED’s existing awesome sensors. I mean, this is the same company that gave us variable OPLF’s, why can’t they give us some high quality color transforms for different looks, lighting situations? I know I’d pay some $$ to get that. RED, hook it up!!!!

 

[Anton Esteban]

I know this is possible because I'm always teasing out better color and image quality from my Dragon footage in Da Vinci Resolve. The problem is I'm not a professional colorist. I know its possible we just need the best eyes in the world looking at what the different sensors are capable of and find ways to get it there. At present, I think what comes out of my camera, without color work, is underwhelming to say the least. But I have great hope in RED cameras and just upgraded mine to a Helium Weapon. It's finally coming in a week or so and I'm really hoping my dream of these new awesome professional color transforms becomes a reality. Come on RED you guys invented this digital cinema game, its time to take it back once and for all!

 

[Brett Harrison]

That's exaclty what we do since 2008 with Graeme Nattress. And it's also the believe of R3D since 2007 to give people a digital negative and not a baked in out of the box look. R3D is unsuited for the ingnorants but a powerfull tool for the photographer to acchieve his image.
Pat

Yes Graeme Nattress is a colour scientist; yes you can think of RAW as a digital negative - it's not a bad analogy but a bit limiting IMO, because a digital camera measures scene luminance in a more precise and usable way for image processing down the line.

However, not sure what you're saying about Graeme in relation to Yedlin - you'd have to elaborate with specific reference to Yedlin's article.

 

[Hrvoje Simic]

I think the bigger question is not how to make RED an ALEXA but how to make the overall image quality and color of footage shot on RED's better.

Bingo.

I know this is possible because I'm always teasing out better color and image quality from my Dragon footage in Da Vinci Resolve. The problem is I'm not a professional colorist.

It's possible.

 

[Steve Conry]

Thought it might be worth mentioning that Canon seems to have tackled this exact task with the C300 Mk II and had success. Here's an article about it with relevant excerpt:

http://noamkroll.com/looking-for-an...meras-have-you-covered-at-a-lower-pricepoint/

"The C300 MK II now comes with a “Production” camera profile that is designed to mimic the color science of the Arri Alexa. When combined with Arri’s Rec. 709 conversion LUT in post, the resulting images between the two cameras are almost too close to call the difference on. For this reason, the C300 MK II is often used as a B-Camera to the Arri Alexa or simply as a cost-effective alternative for the A-camera."

---

Presumably if Canon can match Alexa in a way that is mostly passable, it isn't impossible as some have argued. I do however believe that a LUT, or something of the sort, would be the wrong approach as there are too many color and lumanance parameters that would need to behave differently depending on the conditions. It seems it might be something that would have to happen earlier in the process--likely by RED themselves. My guess is this is what their team is doing when they make updates to the color science, at least to some degree. Question is whether or not they are using Alexa as the benchmark or something else. Either way, does seem possible to get two cameras to match pretty closely, given the right method and manpower.

 

[Gavin Greenwalt]

The Alexa/Red differences are most easily going to be solved optically not electronically. If RED offered a Red, Yellow, Green, Cyan, Blue pattern sensor then you could digitally match a RED perfectly probably to an Alexa, however the filters on the bayer pattern and the OLPF irreparably affect how the image is recorded.

The ultimate example is the color yellow.

- Let's say for every 10 photons of yellow light your green sensor picks up 5 and your red picks up 5. The camera records "Yellow" perfectly balanced.
- Now let's say you have a different bayer pattern where the green picks up 8 yellow and the red only picks up 2 yellow. The camera now records yellow as a more greenish color.

You could digitally try to extract that greenish yellow color and push it towards red, but what if it's actually supposed to be a greenish yellow? Or even worse, what if that greenish yellow isn't from any yellow light at all but it's actually an 8 : 2 ratio of green light and red light.

Trying to capture a nearly infinite spectrum of light exclusively as a ratio of 3 different colors is extremely limiting. If the Alexa reacts to Yellow light one way and RED Reacts to Yellow light another way, but both react to Green and Red light the same, you'll never know if it's reacting to a mix of red and green light or if it's reacting to yellow light. It's like the old math problem. If you have the number "x+y = 4" you can't solve for x or y because it could be 1+3=4 or 2+2=4 or (-8) + 12 = 4. There are an infinite number of possibilities.

That's not to say that the gap can't be closed substantially because it can. And that's not to say that Alexa isn't doing some really pleasing things on the image processing side that RED couldn't also adopt( and should adopt). But, there are limits.

 

[Derek Carlson]

Diffusion filters, RWG/3g10 and a good Kodak LUT like Impulz is about the closest I've gotten to making the RED footage look like Alexa, but it's almost impossible to make them exact. Hate to say it, but on the cheap, Blackmagic is a better option if you're going for the Alexa look. RED and Alexa are like Kodak and Fuji -- different looks for different applications.