Match source to a custom target ColorChecker (match A to B)

Howdy, will try to keep this simple.

Currently there is a "Color Match" function in Resolve, but it tries to match to a theoretical perfect observer of the scene (as if it's lit perfectly), rather than a source/target (match A to B, etc.). This could for instance match a B cam to an A cam even under intentionally dramatic lighting.

It is common practice in high end VFX/post to use the main camera as a target trichromatic observer, and match other sources to it. Currently the only way to do this in Resolve/Fusion is to export EXRs, run an external 3x3 matrix solver, and then create a DCTL to perform this match.

I did make a Python script for doing this with the Colour-Science.org libraries. This is very similar to the "mmColorTarget" and "MatchMacbeth" tools in Nuke. Takes in source/target images and spits out a usable DCTL to perform the 3x3 matrix. Ideally used with Linear data. Very useful for camera matching. https://github.com/ctcwired/dctl-matrix-maker

Put simply, instead of using a ColorChecker to make something look "perfect" the goal is to simply "match source A to source B" which I believe would often be more appropriate and useful.

Thanks!

I may be the ignorant one, but Resolve has both Auto Color and match source to target as well as Color Checker functions.

https://filmmakingelements.com/color-ma ... i-resolve/

Resolve has both Auto Color and match source to target as well as Color Checker functions.

Appreciate the link. Unfortunately none of these functions are appropriate for matching the source and target of a Color Checker itself.

The "auto match" function just does basic frame averaging, and is mostly for 709 content (not log). It seems to be a legacy feature.

The "Color Match" function matches a Color Checker to a perfect (neutral) scene, but does not let you match a source to a target, for instance if the scene is intentionally dramatically lit with colored lights, but you still want to match Camera A to Camera B.

Ideally, this functionality would generate (solve) a 3x3 matrix and perform the operation in linear space, the same way it is often done in VFX.

Resolve has both Auto Color and match source to target as well as Color Checker functions.

Appreciate the link. Unfortunately none of these functions are appropriate for matching the source and target of a Color Checker itself.

The "auto match" function just does basic frame averaging, and is mostly for 709 content (not log). It seems to be a legacy feature.

The "Color Match" function matches a Color Checker to a perfect (neutral) scene, but does not let you match a source to a target, for instance if the scene is intentionally dramatically lit with colored lights, but you still want to match Camera A to Camera B.

Ideally, this functionality would generate (solve) a 3x3 matrix and perform the operation in linear space, the same way it is often done in VFX.

I don't understand why don't use the Color Checker function in Resolve using a chart like the Calibrite Color Checker Video? What color space are you in?

Resolve has both Auto Color and match source to target as well as Color Checker functions.

Appreciate the link. Unfortunately none of these functions are appropriate for matching the source and target of a Color Checker itself.

The "auto match" function just does basic frame averaging, and is mostly for 709 content (not log). It seems to be a legacy feature.

The "Color Match" function matches a Color Checker to a perfect (neutral) scene, but does not let you match a source to a target, for instance if the scene is intentionally dramatically lit with colored lights, but you still want to match Camera A to Camera B.

Ideally, this functionality would generate (solve) a 3x3 matrix and perform the operation in linear space, the same way it is often done in VFX.

I don't understand why don't use the Color Checker function in Resolve using a chart like the Calibrite Color Checker Video? What color space are you in?

Resolve's existing match function attempts to correct the shot to be perfect, which is often not a good idea: https://www.colour-science.org/posts/th ... -harmless/

I'm asking to match A to B. Not make A perfect, or make B perfect. But to pick one as a target for another, as is often a practice in VFX work.

The Color Checker could even be old, worn out, or tinted, and this would still work, as long as the different cameras saw the same chart, because you're picking one camera to be the "master" and bringing the others to it.

This stuff is already built into Resolve.

Resolve Color Match.jpg (47.79 KiB) Viewed 2694 times

Unless you are asking to color match one chart to the other? You can change both clips to linear, mask out everything but the charts then use the Auto match function, this could work.

If not, you could use Lattice to reverse the LUT.

Unless you are asking to color match one chart to the other?

Yes.

You can change both clips to linear, mask out everything but the charts then use the Auto match function, this could work.

This does not work, Auto Match is not a real analysis of the chart.

Thus neither existing feature in Resolve will do the feature I am requesting.

The colorist resource Mixing Light did a detailed, multi-part series on Color Match years ago and how it works, and when it doesn't work:

https://mixinglight.com/color-grading-t ... irst-rule/

One of the rules is: match the brightness first, and THEN Color Match has a better chance at accomplishing the rest. Note MixingLight is a pay service, but I think what they have to offer is worth the investment.

For me, it's fairly trivial to manually match two cameras shooting identical charts, unless one of the cameras is really whacked-out. Anything the Color Match tool does, a user can also do. The advantage of doing it yourself is that it's not done as a mysterious "behind the wall" change -- it's actual lift/gamma/gain and curve values you can see and reproduce.

Matching cameras is Color 101, very basic stuff. It helps to understand scopes and start with a calibrated display. WIthout those two things, it's going to be very rough. All the basics are covered in the manual.

Darren Mostyn has a good tutorial on how to do it:

So far everyone here has clearly not read what I am asking for, though I appreciate everyone trying to be educational.

Color Match will not work if you are intentionally blasting the subject with a bright purple light. Color Match only neutralizes/balances, it does not (currently) let you match A to B.

In particular, the goal is to solve for differences in the spectral sensitivities of the sensors for a given illuminant / scene, in linear, such that you maintain additive mixtures, and the match is exposure invariant. This is possible currently with external tools such as the colour science python library, with Nuke (mmColorTarget, MatchMacbeth), and in Autodesk Flame. Resolve does not have any tooling for this, other than analyzing externally and then generating your own DCTL, which is what I am currently doing.

Currently I export EXRs in Fusion, run the solver with python, generate a DCTL that will perform the 3x3 matrix operation, and then import that DCTL into Resolve each time I need to do this task.

For instance, here is a DCTL that matched a Canon B cam to a Sony Burano A cam, after both had been converted to AP1/Linear, being lit by an ARRI Skypanel that I used yesterday on a project:



This works quite similarly to the RGB Mixer in Resolve, only with many more decimal places, and was produced by a script.

Another way to think of it: Generating your own application / scene specific ACES IDTs.

This also came in handy when needing to match sensor native Phantom high speed camera EXR files to an A cam, since Phantom do not provide an official ACES IDT, so I was able to make a custom IDT specifically tuned for the lighting used on set to nudge it into AP1, matching it bang on with the A cam.

Hope this makes sense, thanks.

So far everyone here has clearly not read what I am asking for, though I appreciate everyone trying to be educational. Color Match will not work if you are intentionally blasting the subject with a bright purple light. Color Match only neutralizes/balances, it does not (currently) let you match A to B.\

The answer is: don't do that. We tell ALL DPs: pull the colored gels when charting the camera, then put the gel back. Otherwise, neither the chart manufacturer or the matching circuit (or scopes) knows what you're doing. It's a very, very bad practice to try to match heavily skewed, colored light. Bad, bad, bad idea.

So far everyone here has clearly not read what I am asking for, though I appreciate everyone trying to be educational. Color Match will not work if you are intentionally blasting the subject with a bright purple light. Color Match only neutralizes/balances, it does not (currently) let you match A to B.\

The answer is: don't do that. We tell ALL DPs: pull the colored gels when charting the camera, then put the gel back. Otherwise, neither the chart manufacturer or the matching circuit (or scopes) knows what you're doing. It's a very, very bad practice to try to match heavily skewed, colored light. Bad, bad, bad idea.

Haha I'm *not* doing that don't worry. I'm just figuratively demonstrating the limitation of the current implementation.

What it does now: Attempt to match the chart to a theoretical reference (often inappropriate).

What I want it to do: Solve for differences between two sources in a mathematically robust & scene-referred (ratio preserving) way. The "reference" in this case might be your main production camera.

The Color Checker could have had coffee spilled on it and this would still work, as long as both cameras saw the same chart at the same time, you can match A to B. Does that make sense?

I already do this type of matching, regularly, and so do many Nuke & Baselight users. It works great (seriously if you've never tried 3x3 matching, or making your own IDTs, you should). However it requires external tools, or to not use Resolve, and it would be trivial for BMD to implement. It is common practice in VFX and virtual production to match tri-chromatic observers.

You can do this in Nuke.
You can do this in Baselight.
You can do this in Flame.
You can do this in ColourScience Python.
You can *not* currently do this in Resolve, natively.

I made a mockup of how I imagine the UI for this could look:

Please excuse my lousy photoshop skills lol



Presumably if project Color Management was enabled, Resolve would auto convert both sources to Linear for measuring & performing the match.

Meanwhile, here's a screenshot of one of the popular tools for this in Nuke. Notice there is a checkbox for "Use Reference Values as Target" but it's optional, and not the only way to use it.

Apologies it's blurry, it's been a while since I had a Nuke license so had to dig this up in a tutorial.

After understanding what the original poster wanted, I agree that this needed in Resolve. Lattice solves this issue but it's not on Windows.

I shouldn't have to balance a shot, recreate the original look, the balance the second shot, copy the recreated original look to the second shot just to match shot B to shot A.


+1

I agree that this needed in Resolve.

Thank you!

Lattice solves this issue but it's not on Windows.

Yeah in general external tools can do this already, but thus the reason for it being a "feature request" for Resolve.

Haha I'm *not* doing that don't worry. I'm just figuratively demonstrating the limitation of the current implementation.

You brought it up. I'm just telling you: don't do this. You'd be shocked the number of people who try to shoot color charts (or even B&W grayscale charts) under a color cast. It doesn't work. Bad idea. Your very concept is flawed and will not work. Believe me or not -- I can't stop you from driving off a cliff.

Haha I'm *not* doing that don't worry. I'm just figuratively demonstrating the limitation of the current implementation.

You brought it up. I'm just telling you: don't do this. You'd be shocked the number of people who try to shoot color charts (or even B&W grayscale charts) under a color cast. It doesn't work. Bad idea. Your very concept is flawed and will not work. Believe me or not -- I can't stop you from driving off a cliff.

I appreciate you wishing to provide advice but, way ahead of ya.

Already doing this operation externally, just would be nice to have in Resolve. This is about matching tri-chromatic observers (essentially custom ACES IDTs), no "color casts" involved. I just gave a bad example, apologies.

Haha I'm *not* doing that don't worry. I'm just figuratively demonstrating the limitation of the current implementation.

You brought it up. I'm just telling you: don't do this. You'd be shocked the number of people who try to shoot color charts (or even B&W grayscale charts) under a color cast. It doesn't work. Bad idea. Your very concept is flawed and will not work. Believe me or not -- I can't stop you from driving off a cliff.

I appreciate you wishing to provide advice but, way ahead of ya.

Already doing this operation externally, just would be nice to have in Resolve. This is about matching tri-chromatic observers (essentially custom ACES IDTs), no "color casts" involved. I just gave a bad example, apologies.

What are you currently using for it? I have this project that I shot color charts for and used Color Match but they aren't lining up.

So far everyone here has clearly not read what I am asking for, though I appreciate everyone trying to be educational.

Actually, we did read it, and I'm telling you we've been doing this for decades. Matching cameras manually is not that hard once you learn how to do it. Have you ever observed an experience colorist in a session presented with this situation? No charts, limited time available, 3 or 4 unidentified cameras, all shooting the same thing, with a fixed budget and schedule. Can it be done without any automatic controls? Yes. Yes, it can.

One issue is nobody will shoot charts, so your opening assumption is flawed. And even if they do shoot charts, the charts are shot badly, or the images have no relation to reality. It's very rare you get a) good charts, b) charts shot well, c) DPs who understand how to use the cameras they're working with (even if they're different). If the DPs are really good, you can get by without charts; but even if the charts are really good, you can't get by with a bad DP and expect great results.

My standard line is, "we can always make it look better, but we may not be able to make it look perfect." I can match two cameras manually in less time than I would bet anybody could take two camera charts and try to match them with Color Match. And even after Color Match, you still have to eyeball them and use experience and good instincts to make the whole show work. There are no shortcuts.

So far everyone here has clearly not read what I am asking for, though I appreciate everyone trying to be educational.

Actually, we did read it, and I'm telling you we've been doing this for decades. Matching cameras manually is not that hard once you learn how to do it. Have you ever observed an experience colorist in a session presented with this situation? No charts, limited time available, 3 or 4 unidentified cameras, all shooting the same thing, with a fixed budget and schedule. Can it be done without any automatic controls? Yes. Yes, it can.

One issue is nobody will shoot charts, so your opening assumption is flawed. And even if they do shoot charts, the charts are shot badly, or the images have no relation to reality. It's very rare you get a) good charts, b) charts shot well, c) DPs who understand how to use the cameras they're working with (even if they're different). If the DPs are really good, you can get by without charts; but even if the charts are really good, you can't get by with a bad DP and expect great results.

My standard line is, "we can always make it look better, but we may not be able to make it look perfect." I can match two cameras manually in less time than I would bet anybody could take two camera charts and try to match them with Color Match. And even after Color Match, you still have to eyeball them and use experience and good instincts to make the whole show work. There are no shortcuts.

Marc, I work in a car crash test facility in Virginia. We use Phantom cameras under high power lighting.

Phantom cameras do not have a standard working colorspace when you export EXRs, it's just sensor native RGB primaries, therefore I NEED and DO use charts ALWAYS in LAB ENVIRONMENT. There is no "DP" this isn't a film set. This is not for cinema or television or web.

Please spare me the patronizing remarks on "budget and schedule" when I am working in a scientific lab. Especially when I am feature requesting tools that ALREADY EXIST in COMPETITOR SOFTWARE. Not everyone works in the same industry, or has the same priorities or needs.

Our team already use this workflow, we simply do it externally to Resolve and boy howdy would it nice if it was an internal feature.

Thanks, and have a great weekend.

What are you currently using for it? I have this project that I shot color charts for and used Color Match but they aren't lining up.

I'm using some ColourScience.org libraries via a Python script I wrote.

You give it two linear EXRs (source and target), it will find the chart in the image, solve for the differences (least means square), generate a 3x3 matrix, and then output a usable DCTL that performs said 3x3 matrix operation which you can insert into Resolve's LUT folder (and restart Resolve).

Sometimes I export EXRs from the Fusion page, after first converting the image to Linear and often I'll zoom in on the chart and add a bit of blur just to get rid of any noise. In the case of my day job, I'm typically exporting EXR frames from the Phantom camera "PCC" software.

On the color page, you can either sandwich a node between two CSTs to run the DCTL in linear, or if your timeline colorspace is set correctly for condition of the footage you can right click the node and set the gamma to "linear" and Resolve will do that sandwich conversion automatically.

Here's a link to the tool I made: https://github.com/ctcwired/dctl-matrix-maker

Example of it running in PowerShell:



Resulting DCTL it creates:



Applied like this on color page:



Demo image, in this case matching them all to the A1 on the left as the source.

You should post the video for all to see. It makes a compelling case on why it should be implemented.

Yeah in general external tools can do this already, but thus the reason for it being a "feature request" for Resolve.

Did you see this thread on Lift Gamma Gain?
Blackmagic should really jump on this before someone else does.

https://www.liftgammagain.com/forum/ind ... phy.18728/

Caleb,

I believe Marc was assuming a more "normal" video production workflow. While yours, which wasn't clear in your initial post, is a very highly specialized process. Not even close to regular story or episodic work.

Hence his comments may not be applicable to your situation. I don't see any patronizing in his comments, just a misunderstanding of the process in use in your case.

And from your clarification it's clear such a thing would be highly useful for your needs.

I can certainly understand. I work a fair amount in Premiere, which doesn't have nearly the range of targeted tools that Resolve has. And have found that IF I can get properly shot chip charts of each camera in a project, in a few minutes I can create curves that far more closely match the tonal and hue responses of the cameras involved.

Making a mostly usable end result possible. Saving hours of work, or the necessity of porting out to Resolve for color.

So yes, we'll shot chip charts, appropriately matched, can be a huge asset. And having a functional automated process would make it ever so much faster.

Sent from my SM-S908U using Tapatalk

While yours, which wasn't clear in your initial post, is a very highly specialized process. Not even close to regular story or episodic work.

Thanks for your response. Perhaps, though the method I am using is common practice in VFX. I also totally understand coming at it from a "I've always done it this way and it's good enough!" point of view. But take for instance the advent of workflows like ACES, which might not have made sense to most people 10 years ago, but over time has become an essential, useful and practical tool, even for someone making "just a low budget documentary" or whatever. It's there in the menu, and it works for everyone, which is great! Even if most users don't realize how many countless hours engineers spent arguing in committee meetings over it's design heh.

One of the things that made Resolve way ahead of it's time compared to other tools was having a true floating point workflow. A huge difference in design compared to something like Premiere. Now I'm sure if you explained that to someone who has two hours to edit a corporate promo video, they wouldn't understand why you're wasting their time with such details. But in the world of VFX and Hollywood that is in fact the massive difference that made it truly usable as a professional DI tool.

So with the mindset of future workflows that will only ever have more CGI, scene-referred camera footage & processing power, It's good to place value in precision pipelines that are easy to use.

In terms of the science of this matching method:

The important trait is that this match occurs to linear data, and is solving for differences in tri-chromatic observation. Theoretically, there is no such thing as "accurate color" or "accurate stimulus reproduction" in the tri-chromatic world. By only having three channels (cameras and humans) there is always an implicit bias for each observer, because it's not the full spectrum of light. This creates phenomenon like "metamerism failure", and is why different camera manufactures still have a different feel to their color science, even with log footage. It's all about the (careful) design tradeoffs in the 3x3 matrix that goes from sensor native to the working space. (SGamut3.cine, Arri Wide Gamut 3, etc.). Usually you can simply accept those biases. However in some conditions it's useful try and eliminate it, or do the color transformation yourself in a way that is optimized for your scenario.

Fun fact you can also use charts to reverse engineer whatever the sensor native colorspace was for your camera, which can allow you to do perfect white balance adjustments in post, even for log footage. The "magic" that allows RAW codecs to do this is simply the software SDK knowing ahead of time (metadata) what the 3x3 matrix is for that sensor to get back to it's native observer space (or sometimes LMS colorspace), but that's a topic for another day.


source: https://community.acescentral.com/t/res ... ation/2229

video from FilmLight on additive light relationships in scene-referred data:

What are you currently using for it? I have this project that I shot color charts for and used Color Match but they aren't lining up.

I'm using some ColourScience.org libraries via a Python script I wrote.

What charts does it use? I have the Color Checker Video but it's not reading it.

What charts does it use? I have the Color Checker Video but it's not reading it.

I'm using this library: https://github.com/colour-science/colou ... -detection

On that page it says:

"The model has only been trained on ColorChecker Classic 24 images and will not work with ColorChecker Nano or ColorChecker SG images."

There might be a different library that could be used in place of it but I'm not sure.