LUT Strees Test: HighEnd Film Emulation Kodak - Dual Hue

Evaluating LUTs with a Stress Test

May 16, 2018

Look-Up Tables (LUTs) are black boxes that make it difficult to know what precisely they're doing to your image. Colorist Jason Bowdach shows you how to stress test and evaluate your LUTs.


Series

Team Mixing Light Note: We are pleased to welcome Jason Bowdach to our roster of Contributors, here on MixingLight.com. He kicks us off with a great article & video on an important subject. Please join us in welcoming him to Mixing Light, in the Comments below! You can find out more about Jason (and all of our Contributors, on the About Us page)


How do you know if your favorite LUT is destroying your images?

 “A little knowledge is a dangerous thing” – Alexander Pope

Over the last few years, LUTs (or Look-Up Tables) went from being an esoteric technical tool, understood by only the most technically inclined, to a common tool that most Directors, DPs, and Producers are now somewhat familiar with. I emphasize somewhat because the potential usage of a LUT has been stretched far beyond its original purpose, making creative discussions with clients difficult without clarification about WHY we are using a look-up table in a specific workflow.

As LUTs can perform a wide range of color manipulations, they’re used for a huge variety of both technical and creative purposes. In one workflow, a LUT may be used to technically transform the source image from the camera’s capture format (typically stored RAW or logarithmically) into a more visually pleasing Rec709 image. In another workflow, a LUT may be used to apply a creative look to footage that has already been normalized and balanced. In a third workflow, a LUT might combine a technical transform with some creative adjustments, making a hybrid LUT that is both technical and creative. The possibilities are so vast, that it’s potentially destructive. Therefore, a golden rule to remember is that all LUTs are not created equal. This situation is exacerbated by the fact that LUTs are essentially “black boxes,” which obfuscate our ability to see whats going on “under the hood” of a LUT.

a 3D LUT illustrated as a three-dimensional cube.
a 3D LUT illustrated as a three-dimensional cube.

How well do you understand your LUT?

As I’m sure many of you can attest, it has become common practice to provide the colorist \ finishing artist with the LUT used in-camera (while shooting) or for dailies color prior to beginning final color & finishing. The LUT provided may be a generic transform LUT supplied by the camera manufacturer (great!), a custom LUT developed specifically for the production (even better!), or a random LUT that someone from the production found online in a forum (not so great).  Unfortunately, from a colorist perspective, it can be difficult to objectively judge the quality of the LUTs being provided, even when applied to a variety of real footage.

Could using the provided LUT during final color result in artifacts, banding, or noise that might not be present in the source?

Potentially. Aside from applying the LUT to a huge library of test footage, is there a way we can evaluate the LUT more objectively? Yes, there is.

Test charts & objectively stressing your LUTs

In previous Insights, Patrick demonstrated how we can use a greyscale ramp to visualize the effects of a LUT.  While evaluating on a greyscale ramp is certainly helpful, we can get even more granular. In my research on the subject, I found several unique test charts that when combined on a single timeline, make for a powerful “LUT Stress Test” and a great addition to any colorist’s toolkit.

Stressing the LUT: The Baseline Images

Let’s take a look at the test charts & clips that I’m currently using in my stress test timeline:

TrueColor's LUT Stress Test with no LUT applied. One side represents linear (video) and the other represents cineon film log.
TrueColor’s LUT Stress Test with no LUT applied. One side represents linear (video) and the other represents cineon film log.
Two color wheels, one representing highly saturated hues and the other less saturated hues.
Two color wheels, one representing highly saturated hues and the other less saturated hues.
A high bit-depth gradient with hue represented on the horizontal axis. On the left (L) side, saturation is represented on the vertical axis. On the right (R) side, lightness is represented on the vertical axis.
A high bit-depth gradient with hue represented on the horizontal axis. On the left (L) side, saturation is represented on the vertical axis. On the right (R) side, lightness is represented on the vertical axis.
A real-world test image in Arri LogC (Originally REDLog3G10 RWG, converted using Resolve Color Management)
A real-world test image in Arri LogC (Originally REDLog3G10 RWG, converted using Resolve Color Management)

Stressing the LUT: Arri LogC to Rec709

Note: The images I’m sharing in the remainder of this Insight do suffer some JPEG compression artifacts… but I think the degradations I’m talking are clearly visible and distinct from the compression artifacts.

Before we take a look at anything too outlandish, let’s evaluate a common LUT that most of you are familiar with and happens to be bundled in all versions of DaVinci Resolve: Arri’s LogC to Rec709 Transform LUT.

A real-world test image with Arri's LogC to Rec709 LUT applied.
A real-world test image with Arri’s LogC to Rec709 LUT applied.
TrueColor Stress Test with Arri's LogC to Rec709 LUT Applied.
TrueColor Stress Test with Arri’s LogC to Rec709 LUT Applied.
Dual Hue Saturation Circles - Arri LogC to Rec709 LUT.
Dual Hue Saturation Circles – Arri LogC to Rec709 LUT.
Smooth Color Gradient with Arri's LogC to Rec709 LUT applied.
Smooth Color Gradient with Arri’s LogC to Rec709 LUT applied.

Reviewing all three charts, its easy to see what the LUT is doing and more importantly, if it causes any image degradation such as artifacts, noise, and banding. For example, Arri’s LUT appears to significantly increases the saturation and color contrast, causing a fairly clean “edge” line between colors that can be seen quite clearly on the 1st chart. While the more visually attune might be able to see the same in the real world test image, it is not as apparent and easy to see. It’s also more difficult to see any imperfections introduced by the LUT with the real-world image unless the image is pushed so far that it “breaks”. In this case, I see nothing of concern in either the real-world test image or the charts, but given this is a proven look-up table by the camera manufacturer, I would be concerned if I did see any signs of image degradation.

Now that we have a reference, let’s evaluate a few different LUTs using out Stress Test Timeline and see how they fare.

Stressing the LUT: Kodak film emulation

First up, I’m going to test a popular 3rd-party Kodak film emulation LUT, which has been backed by some pretty well-known talent:

High-End Kodak Film Emulation LUT applied to a LogC real-world test image. A fairly appealing image with nothing obvious awry.
High-End Kodak Film Emulation LUT applied to a LogC real-world test image. A fairly appealing image with nothing obvious awry.
High-End Kodak Film Emulation LUT applied to TrueColor Test Chart. Artifacts and color noise are visible, especially in the magenta hues.
High-End Kodak Film Emulation LUT applied to TrueColor Test Chart. Artifacts and color noise are visible, especially in the magenta hues.
High-End Kodak film emulation LUT applied to the Dual Hue Circles chart.
High-End Kodak film emulation LUT applied to the Dual Hue Circles chart.

High-End Kodak Film Emulation LUT applied to Smooth Color Gradient Test Chart. Artifacts and color noise are visible.
High-End Kodak Film Emulation LUT applied to Smooth Color Gradient Test Chart. Artifacts and color noise are visible.

While the test image looks pretty good with the Kodak film emulation LUT applied, it becomes clear there are some problems when we look at the charts. The LUT appears to create artifacts on the test charts, especially with magenta hues. Diving deeper (as seen in the video Insight below), adding a serial node and raising the gamma allows you to see any degradation more clearly, especially if it exists in the shadows. Adding a splitter-combiner node allows you to inspect the individual Red, Green and Blue channels to see a LUTs effect on each individual channel, if you want to go to that extreme level.

Stressing the LUT: Kodachrome Look Lut

Evaluating this Kodachrome-inspired LUT using our stress test, we see nothing obviously wrong with the real-world test image but the charts show a different story:

A Kodachrome-Inspired Creative LUT intended for LOG footage. No apparent issues, aside from underexposure.
A Kodachrome-Inspired Creative LUT intended for LOG footage. No apparent issues, aside from underexposure.
A Kodachrome-Inspired Creative LUT intended for LOG footage. Significant issues are visible in the highlights.
A Kodachrome-Inspired Creative LUT intended for LOG footage. Significant issues are visible in the highlights.
A Kodachrome LUT applied to our Dual Hue Circles Test Chart.
A Kodachrome LUT applied to our Dual Hue Circles Test Chart.
A Kodachrome-Inspired Creative LUT applied to Smooth Gradient Chart. Same artifacts visible in highlights.
A Kodachrome-Inspired Creative LUT applied to Smooth Gradient Chart. Same artifacts visible in highlights.

Strange artifacts appear as we approach the highlights and some checkerboard-like patterns appearing in the blue-cyan color gradients have me very concerned. I wouldn’t be surprised to see increased noise and artifacts in blue skies using this LUT, based on what we’re seeing on the above charts.

Now, this isn’t saying you should avoid this specific look-up table but the stress test provides us with more information that allows us to make an informed decision about using it and what to expect. It also provides something more visual to use with clients in the discussion about using a specific LUT or if you should re-create a look from scratch in Resolve.

Stressing the LUT: Teal & Orange Look

Let’s evaluate a 3rd-party teal & orange creative look LUT I found on a forum:

A 3rd-party creative look LUT applied to a real-world LogC test image. Some clipping is visible, but it could be creative intent (purposeful as part of the look).
A 3rd-party creative look LUT applied to a real-world LogC test image. Some clipping is visible, but it could be creative intent (purposeful as part of the look).
A 3rd-party creative look LUT applied to the TrueColor Chart. Some minor artifacts visible in blue and green hues.
A 3rd-party creative look LUT applied to the TrueColor Chart. Some minor artifacts visible in blue and green hues.
A 3rd-party creative look LUT applied to the Dual Hue Circles Test Chart. Some minor artifacts visible in blue and green hues.
A 3rd-party creative look LUT applied to the Dual Hue Circles Test Chart. Some minor artifacts visible in blue and green hues.
A 3rd-party creative look LUT applied to the Smooth Color Gradient Chart. Some minor artifacts visible in blue and green hues.
A 3rd-party creative look LUT applied to the Smooth Color Gradient Chart. Some minor artifacts visible in blue and green hues.

 

Looking at the test image, there is no obvious reduction in quality or artifacts. However, evaluating the test charts I see unnatural looking artifacts visible in the blue and green hues. Nothing too major but certainly something to keep in mind if accuracy in those hues ranges is of concern for the specific project you’re evaluating this LUT for. That being said, the look is a teal & orange variant so its not surprising to see those colors being manipulated.

Stressing the LUT: Generic Film Emulation

Let’s evaluate one more look-up table, a freely-available film emulation LUT from the wonderful folks at Light Illusion:

Real world test image with Light Illusion Film Emulation LUT applied. Looks pretty good, although a bit underexposed.
Real world test image with Light Illusion Film Emulation LUT applied. Looks pretty good, although a bit underexposed.
TrueColor Test Chart with Light Illusion Film Emulation LUT applied. No artifacts or degradation visible.
TrueColor Test Chart with Light Illusion Film Emulation LUT applied. No artifacts or degradation visible.
A film emulation LUT from Light Illusion LUT applied to the dual hue circle test chart.
A film emulation LUT from Light Illusion LUT applied to the dual hue circle test chart.
Smooth color gradient with Light Illusion film emulation LUT applied.
Smooth color gradient with Light Illusion film emulation LUT applied.

Compared to many of the other LUTs we’ve evaluated, the film emulation LUT available from Light Illusion is far less destructive to the image. I see no visible artifacts, noise or banding in any of the four images or charts (aside from what negligible amount would be considered part of the film emulation look and curve). While the results of this stress test are up to a certain amount of interpretation, this specific look-up table passes with flying colors as far as I’m concerned.

Additional Thoughts

If you’re interested in learning more about what a LUT is doing “under the hood”, you can use the concepts demonstrated above to visualize the changes using the same charts we use as the stress test. These charts clearly indicate the changes in hue, saturation, and contrast you can expect from a LUT.

If you’re a Resolve user and want to sharpen your skills then grab a still of a chart with a LUT applied and match it by hand – when you have some time between sessions.

I hope this insight has given you some ideas you can use and develop your own stress test. Please share your feedback, as I’d love to hear your thoughts.

Special Thanks

Thanks to the lovely folks at True Color & 3D LUT Creator for creating the test charts used in this insight. All the charts used in the article and video are available to download for free on their websites. Click the links for the download pages and further information.

The real world test footage used in this insight is available from RED’s website. I converted it to LogC using Resolve Color Management as I felt it was more standardized and therefore, a better test for look-up tables.

And the LUT from Light Illusion is freely available at their website.

-jason


Comments

15 thoughts on “Evaluating LUTs with a Stress Test”

  1. Great Insight, Jason! You’re systematic and technical knowledge is greatly appreciated. This is a long overdue peak behind the LUT curtain I was pleased, but not surprised, about Steve Shaw’s film emulation LUT. It looks sound. Although I don’t use look LUTs, I often use Steve’s LUT with dialed down gain (.250-.450) in a lot of my stuff as my clients love it as icing on the cake so to speak.

  2. Thanks for such informative insight.

    It’s my understanding that a creative LUT necessarily is going to introduce some color shift and luminance levels.

    Analyzing the LUT through the charts, I would like to know where is the visual-crossing line between the intended color and luminance shift and the damage the LUT is doing? In other words, what are the criteria and parameters to define the line between the shift in color and luminance produced by the intended look and when it become a damage? How to visually distinguish between one and the other?

    Coincidentally, the LUT showing the least damage on the chart and the sample image (the Light Illusion) is the one with less intended alteration in color hue, saturation and luminance. The LUT looks to me more like a bleach bypass than a teal and orange. Then validating this LUT as the one introducing less artifact or damage seems to not be fair compared with the LUTs presenting a more pronounced intended look. I would appreciate some comments here to understand better the procedure and criteria in analyzing the LUTs.

    Welcome, and thanks once again for the great insight.

    1. Hi Willian,

      We’re not so much looking for shifts in color and luminance, but unnatural, “digital” looking aspects as a side effect of the color luma manipulation. For example, take a look at the attached examples and you can see some areas of concern indicated in red (one example has minimal color shifts, but major distortion and the other has significant color shifts but no distortion). The harsh, digital-looking “artifacts” (visible in one example) are not the intended result and will likely results in reduced image quality. Its more about HOW the LUT is going about the changes more than what it is doing, especially as we can’t see how they were created (what tools were used, etc).

      I hope that helped clarify a bit and answered your question. https://uploads.disquscdn.com/images/84d17a97ea1797550b96386d50774e5a5985e5f762ffe65b5f8b5ea0a5372162.jpg

      https://uploads.disquscdn.com/images/5cb842ba7fc5a4bba9a853c32293815ebe8c48719770d4572a2897ad1dd6191c.jpg

      1. Hi Jason,

        Thanks a lot for the clarification and the uploaded images. For sure it helps to understand better the measurement. I wish there was a more precise numerical method to measure these type of distorsions, like the tools we have in ScopeBox to measure in 2D similar properties in an image. Looking foward for your next insight.

  3. Hi Jason great post! I saw a post on Facebook about a similar thing (someone was showing what a lut a dp had brought to him did you a gradient). You also posted a link to your insight there. The waveform from the post showed heavy banding but Walter Volpatto pointed out that the banding might be happening because the interpolation was set to trilinear and not tethraedrical, and the difference on the waveform significant. So out of curiosity what setting did you have enabled? Is there a visible change in the image when changing from one interpolation to another? Also do you thing that this might occur because the low quality luts are perhaps lower resolution (Is that the term?) like 17x17x17 or 24x24x24 instead of say 33x33x33?

    1. Hi Jose, I tend to use tetrahedral when working in Resolve, but trilinear if I’m involving other applications in my workflow (as trilinear is the default for many applications, although thats changing). While there is a small difference, i dont feel the interpolation method would cause my findings to be drastically different. That being said, the interpolation method is more important the lower the quality of the LUT (say 17x17x17 instead of 33x33x33), as it needs to “interpolate” more points. Hope that helps!

      1. Thanks for your reply Jason, I just started a project where the DP really liked the Fuji D65 Lut that comes with resolve. Later I remembered this insight and tested that lut on a greyscale and there’s a small spike on the red channel in the shadows. I’m wondering how bad that is and if there’s something I can do to change that? Here’s a screen grab of the parade for that lut

  4. Terrific information, Jason. While I’m often “Mr. Anti-LUT,” I have to say there are technical LUTs I will reach for under some conditions, and if a DP insists on a custom LUT, I don’t have a problem at all using it if it does the job. The Arri Log-C -> Rec709 LUTs are pretty mild and aren’t that bad at all. I would point out that, using the test charts above, it’s possible to not only see what the “creative” LUTs are doing, but also demonstrate a way to get close to matching them with normal color-correction controls… but in a way that doesn’t break the image. Note also the “LUT Preview” mode in Resolve 15 also works with PowerGrades, and I think the PowerGrades are more useful for many situations.

  5. Jason,
    thanks for the great article and video!
    I am trying to replicate your project in Resolve 15. Did you set up RCM for this, or did you go straight DaVinci YRGB?
    Cheers

Leave a Reply

Hundreds of Free Tutorials

Get full access to our entire library of 900+ color tutorials for an entire week!


Start Your Free Trial
Loading...