LUT Stress Test Image

John Schick Jan 4, 2017

  1. Hey, I just wanted to share this free LUT Stress Test Image I've been working on. Not all LUTs are created equal, and this is one way to check the build quality of your LUTs - to see if they exhibit any noticeable banding or other artifacts.

    After applying the LUT, check the individual red, green, and blue channels. Also try increasing the gamma (after the LUT) to see what’s happening in the shadows.



    Download link (16-bit tiff):
    http://truecolor.us/downloads/lut-stress-test-image/

    More information:
    http://truecolor.us/news/not-all-luts-are-created-equal/
     
  2. Hey, it's 2017, and there's already a new LUT out there!

    BTW, the demo is very well-lit and has excellent graphic design.
     
    Rodrigo Solera likes this.
  3. Great to see a LightSpace CMS customer adding to the field of colour management.

    Mind you, true film emulation LUTs will indeed cause artefacts and banding, especially if you push/pull contrast/brightness, as that is exactly what film does... It is inherently non-linear and even suffers foldback issues (such as reciprocity failure).

    Just goes to show that using true 'film emulation' is often NOT what you really want when grading.:D

    Steve
     
    Marc Wielage and Rodrigo Solera like this.
  4. Isn't it a bit misleading to compare a technical gamut transform LUT (which can be done using straight matrix transforms) to a FPE LUT, which is built with scans?

    This is like taking a low res jpg of a photo image and comparing it to an vector graphic EPS. Zooming in at 500% and going, "see ours is better!"

    Well yeah and a fighter jet is faster than a horse, but it's not a fair comparison.
     
    Last edited: Jan 4, 2017
    Marc Wielage and Rodrigo Solera like this.
  5. Hey, yeah I agree that the LUTs examined in the test do different things. Though I wanted to show that artifacts exist on a variety of types of the transformations.

    I still maintain that it's possible to create a film print emulation LUT without those banding artifacts. I've attached a quick example of a repaired version of the LUT I had used:

    LUT_repair_test.jpg

    Clarification: I agree that in order to accurately match a particular film stock, it will have some inherent unevenness. And when creating a smoother LUT (or repaired version), the colors will be less precise than the original. Though when using something like film emulation as part of a look, my approach is I'd rather trade some color accuracy for better LUT quality.
     
    Last edited: Jan 4, 2017
  6. I'm not convinced anybody who puts out a film emulation LUT actually provides something that is going to change a digital image and make it look like it was printed on 35mm print film. (I concede that Baselight's Truelight system is extremely well-documented and accurate, and Steve Shaw's Lightspace LUTs are also very good.) I've been using LUTs since 2002, when I worked for Kodak, and it was a very nebulous, vague, and ephemeral area subject to a lot of opinion and interpretation. Even at Technicolor, where we had Thomson's LUTher box, we also had to employ what we jokingly called "the LUT of the month club," where everything had to be tweaked and adjusted by eye, because nothing was consistent or predictable.

    I don't have a problem with anybody saying, "hey, if you use my LUT, it'll make your pictures look different" or even "we think it'll make your pictures look good." Just don't tell me it's going to emulate film, or (god forbid) a specific emulsion. Anybody who's worked on actual film projects will tell you, even film itself is inconsistent: I've done shows where the first batch of 5219 wasn't identical to a batch of 5219 used a week later, and then when the lab had a bad day, the same batch of film comes out looking 20% different. Trust me -- it happens. It's not an exact science. And there's even more chaos theory the moment a film recorder and answer prints are involved.

    I also believe you can come up with perfectly acceptable starting points just with the controls already built in to color-correction systems. I think it's misleading to beginners and neophytes to tell them, "hey, buy our $29 or our $79 or our $699 LUTs, and they'll color-correct your picture!" I don't have a problem with a custom LUT designed for a specific project and exposure when the DP is involved. At the same time, on sessions where I'm handed a LUT, I'll typically reverse-engineer it with a grade, A/B the results with the client, and I have yet to experience them saying, "naaaa, I'd rather go with the LUT."
     
    Rodrigo Solera likes this.
  7. Cool!

    Would it be helpful to have one for video levels and in YUV color space or is the scaling not relevant for 16bits?
     
  8. Those are some good points about film, Marc. I didn't mean to imply a film stock always had one exact look, or that you could completely match it with a LUT. It might be possible to get a close match in one scene, but like you said, it would be different every time. I was mainly trying to make a point about overall LUT build quality (I'd rather sacrifice some accuracy for better quality).

    And yes the lighting from the video was well done, lit by Sean Ryan: https://vimeo.com/user4735410

    I think since most LUTs are built for RGB material, you'd want to apply it to that. You could always try converting to YUV after the LUT, and analyze the color information separately.

    Regarding video levels, the main idea of this image is to see the entire LUT's range of values from 0 to 1.0, but you're welcome to experiment with other combinations.
     
  9. Perfect idea. But, probably a version that has more colors for real world colors would be better? I mean real shot colors like skintone, sky, vegetation.
     
  10. I'd like to see real-world footage shot badly. That's always a challenge.
     
    Cary Knoop likes this.
  11. I mean more gradations for these test images. I see most of colors here are oversaturated.
     
  12. i have a load Marc .....will FED EX over the RAID :):)
     
  13. For checking how a LUT handles skintones, sky, etc., that would probably work best on another image. Since different LUTs expect different input parameters (Cineon/LogC/SLog & WideGamut/Rec709 etc.) the expected skintone pallet would be different for each of those combinations.

    What I needed for the chart was to see the maximum range of values that a LUT handles (0 - 1.0), in order to check if there are any artifacts at the extremities. I did include the Cineon values though, since most LUTs I work with expect a log input.
     
  14. The Cineon side of your image appears simply to be the patterns from the left side with a linear to Cineon conversion applied. I'm not entirely sure how useful a test that is. A linear ramp in linear light (and hence a Cineon conversion of that) is not particularly representative of anything photographically meaningful.

    But the left side of your image is definitely useful to show up distortion in LUTs.

    I would argue that it might be more useful to have the right side go linearly between e.g. 240 and 732, which is the Cineon range representing 3 stops under mid grey to 3 stops over. 204-623 would be the equivalent for LogC. So maybe 200-750 would be a reasonable compromise, representing the range of values where most of the pixels of a log image sit. Or even going from black to diffuse white (95-685) as you do, but linearly rather than exponentially. After all, linear intervals in log are equal steps of exposure.

    Make sense?
     
  15. Jason Myres

    Jason Myres Moderator

    3,625
    2,263
    0
    You're right, but a good take-away from this is if you're going to use third party LUTs, have a process to vet them so you actually know what they're doing to your images.
     
    Marc Wielage likes this.
  16. Thank you for the feedback Nick, and those are some great ideas. I just took a look at applying your suggested offsets, and they do provide a nice alternative to the Cineon transform.

    Those interested could also add the adjustments (like desaturation or scaling offsets) to the image prior to adding a LUT, if desired.

    Regarding usefulness, I would say that for me, the Cineon values were very helpful when creating & checking the REDWideGamutRGB to REDcolor4 LUT. Since the encoding is non-linear, it gave a different perspective on how the color transformations affected the image.

    And you’re right about it not being photographically meaningful - I just needed to see how the LUT held up at those extremes. When I was testing everything, I used plenty of normal footage as well.
     

Share This Page