RED Color - A look at REDWideGamutRGB/Log3G10

[Antony Newman]

If the colour science is affected by the transition of colour (and luma changes) in the RAW images - would images with swathes of 'no colour change' from image Charts provide insight into what Grames algorithm is doing with real images?

Ie the final look may 'tend' towards the same steady state values, but the transition could look different.

Observation

1st Image :

Whilst not looking directly at the 'bright LED lights' in the picture in RWG, my brain says 'there is something very bright' in that part off the picture. This algorithm seems to tally more accurately with my perception of blooming.

High frequency 'peppering' of the DC2/RG4 with a grey veil (or highly luma 'noise') seems reduced in RWG (noticeable in the blue in the top left of the image)

Details previously invisible - now see the light of day. Eg top of the left most guitar fret. OMG .. that shadow detail.
And the top drum emblem and details around a huge step up in colour and luma detail.

2nd Image :

Difficult to determine to what extent I am viewing makeup vs natural skin tone.
RWG looks less grey, like the skin looks more moist (or less dry).
Hue change in RWG, seems dependent on skins incident angle to light source - and gives a 'roundness' and 3 dimensionality to the image.

4th image :

RWG : Noise below the 'PHIL' text has a less noticeable grain pattern (great for low light)
Building behind : Not sure the colour of the white curtain veil. RWG seems to have slightly more DR reach in the top right (great for scenes nearer clipping)

5th image :

Massive difference. I can't comment on absolute colours (not seen Phil in person).
RWG looks moody (emotional content in the shadows that begs you to stare for a little longer).
RWG image looks more like a 'Stills' camera.

Impressive work.

Thankyou Phil for sharing your analysis & investigations, and Graeme for what must have been a herculean research effort.

AJ

 

[Brian Boyer]

...but it's certainly an improvement from what I can see.

Absolutely. Sometimes it's difficult to judge the subtle differences or be able to fully appreciate the improvements when viewing images side-by-side. So, I cut each side out and stacked them in layers within Photoshop.

Being the exacting guy that you are, each side is the same size, which removed the hassle of lining them up. After a simple cut and paste, toggling the upper image on and off allowed for a more direct comparison.

The RWG/Log3G10 + current Transform LUT appears to breathe new life into the images compared to DRAGONcolor2/REDgamma4. I mean that quite literally. Skin looks more alive and exhibits more of its complexity when light bounces around beneath the surface. I'm digging it.

Well done, Graeme.

 

[Phil Holland]

Marc, when shooting a test chart, how much of the frame would you like it to fill? e.g. I would like to include a test chart at the start of each interview. Does it really need to fill the full frame? Can it fill 50% of the frame? Lets say I want the interviewee to hold the chart in front of them, just below their face.

When shooting pure test charts it's ideal to fill as much as the frame as possible.

In practical live action and using them as a point of reference, you want them big enough so you can interpret the patches clearly without any contamination.

These are field charts used in these images and are useful for A/B comparison. Meaning you can see the difference between those two images side by side. Shooting a pure chart I'd whip out the CDM or something a bit more technically correct to a degree.

I'm not so against XRite and DataColor charts used for in frame color reference actually, but generally speaking DSC does hold a higher QC level. They'll be making a cameo in upcoming tests. Just have to haul them around. I almost always have compact charts with me.

 

[Les Hillis]

...Is anyone else seeing Marc's images with the correct colors?

It looks off to me too, with the colours skewed as you described.


I really like where these new RED colour developments are going though. Thanks Graeme and Phil and everyone for making the improvements and showing us the progress.

 

[Joel Arvidsson]

The latest SDK has an updated Log3G10. This was released last week. This matches the Log3G10 in the official release of build 42, which also came out last week. There was even a new camera FW update.

Thanks Stacey.

 

[Mike P.]

With the latest round of test shots, I definitely prefer the 3g10/RWG+709 conversion across the board (in all the sample pics).

Sure, the differences may be (very) subtle, but all of those subtle improvements ultimately adds up to a discernibly nicer image than DC2/DG4...

 

[Graeme Nattress]

Some shots are subtle, not least because DC2/RG4 is not that bad at all. However, other shots it's beyond night and day, especially obvious on shots that have out-of-gamut colours. The other thing I notice straight away is the "3D effect" (for want of a better description) and how the contrast in the image behaves over a wide range of ISO developments.

Graeme

 

[Bob Gundu]

I really like what's going on here. Along with re-maping of data, is there saturation falloff too?

 

[Marc Wielage]

Regardless of which chart you use, color chart analysis in examples like these works for a particular scene color adaptation for color grading purposes and just a general impression.

Not with the XRite charts. With the DSC charts -- which are a known standard -- yes.

If Phil wants to provide images with full-frame charts, I'd love to see them.

 

[Graeme Nattress]

I really like what's going on here. Along with re-maping of data, is there saturation falloff too?

Thanks Bob! There's an awful lot going on!

Graeme

 

[Hrvoje Simic]

Not with the XRite charts. With the DSC charts -- which are a known standard -- yes.

If Phil wants to provide images with full-frame charts, I'd love to see them.

The quote lacks the last part. The statement wasn't negating your point with differentiation of charts but stressing the double edged sword of using them for judging camera performance, due to chart limitations and properties of light.

This is an image analysis tool for color grading and shooters, which shows only what raw recording camera does through a particular transform on particular chips bathed with particular light.

It doesn't work for deeper insight of camera spectral response, it is not applicable in many scenarios even for color grading (because light varies due to light source properties and environment casts and bias, and light may easily differ 1 meter away) and on many shoots no one is going to wait for the chart shot. If you do get lucky to get a chart shot, balancing strictly on it may easily screw up other elements of the shot, it may easily be unusable three shots later and good luck convincing a large(r) cast and crew to adapt their m.o. for a chart shot every time the setup changes.

 

[Mike Nagel]

The quote lacks the last part. The statement wasn't negating your point with differentiation of charts but stressing the double edged sword of using them for judging camera performance, due to chart limitations and properties of light.

This is a color grading approach (half usable) and rough and general camera analysis, which shows only what raw recording camera does through a particular transform on particular light. It doesn't work for deeper insight of camera spectral response, it is not applicable in many scenarios even for color grading (because light varies and environment throws casts, and light may differ 1 meter on the side) and on many shoots no one is going to wait for the chart shot.

that is why u take the chart shot right in front of the foreground talent - that is what the scene is lit for...

if done correctly the chart shot is done in the exact setup as the actual shot - there is no delay or hold up... only need to get it once, until light, lens or angle changes...

 

[Marc Wielage]

that is why u take the chart shot right in front of the foreground talent - that is what the scene is lit for... if done correctly the chart shot is done in the exact setup as the actual shot - there is no delay or hold up... only need to get it once, until light, lens or angle changes...

Well, the chart basically gets us in the ballpark... if we can believe the chart, and if the chart relates to a known color space (like Rec709). To tell you the truth, I'm generally good with a grayscale chart, provided it's in the exact same light as the actors, it's in focus, and it fills the screen. Just having that is helpful, and that goes back to film.

One alarming thing about Phil's charts is that there's very little Magenta or Green content, and it's in that spectrum that Rec2020 and future systems like it are supposed to provide more color range. I would bet it's not Phil's problem or a camera problem -- it's a chart​ problem.

it is not applicable in many scenarios even for color grading (because light varies due to light source properties and environment casts and bias, and light may easily differ 1 meter away) and on many shoots no one is going to wait for the chart shot. If you do get lucky to get a chart shot, balancing strictly on it may easily screw up other elements of the shot, it may easily be unusable three shots later and good luck convincing a large(r) cast and crew to adapt their m.o. for a chart shot every time the setup changes.

I know. I have 37 years experience in color correction and have been down that road many times.

I generally get a call into the 1st AC and request that they give us a color chart maybe a couple of times a day or on a major lighting change, like indoors / outdoors or nighttime set / daytime set. Just 3-4 seconds of full-frame chart is fine, and that takes no time to do. We're talking a half dozen times a day, tops. Nobody needs it on every take or every scene, but it's nice to see one on the first shot of the day.

Having said that: I've done many thousands of major projects with zero charts, and we somehow managed to stumble through them OK. But in a camera test comparison, I very definitely would want to see charts on everything, especially if you're trying to see a difference between A and B.

 

[Mike Nagel]

Well, the chart basically gets us in the ballpark... if we can believe the chart, and if the chart relates to a known color space (like Rec709). To tell you the truth, I'm generally good with a grayscale chart, provided it's in the exact same light as the actors, it's in focus, and it fills the screen. Just having that is helpful, and that goes back to film.

yeah, neutralizing the 1D via at least a few greyscale chips is the very minimum I do

 

[Simran Dewan]

Phil, apologies if the question has already been asked, but are you using Graeme's 709 Transform or have you created your own?

 

[Phil Holland]

Phil, apologies if the question has already been asked, but are you using Graeme's 709 Transform or have you created your own?

Graeme's of course.

Here's a couple new grabs:

I'd post more, but I'm currently working on a Dragon and Helium test!

 

[Patrick Tresch]

I love what I see in the color level. I love the saturation of the skin tones. I would love to see this on high saturated led colors. To see if there are some clipping problems. I bet not...

 

[paulherrin]

very nice!

 

[John Brennick]

As always, thanks Phil!!

 

[Adam Montville]

Looking side by side to the matching Oct frames, I really like the slightly increased contract (or maybe less brightness), and the punchier colors, enhancing that "3D" effect. I can't wait!