LUTs - using and building

Steve,
Welcome to the monkeyhouse. I don't know if you've had the opportunity to open Apple COLOR and examine its internal "LUT builder", but it does exist, and essentially generates a grade that can be stored and applied as a re-mapping filter, sitting just outside the active correction "rooms". It is supposed to be of the mga form, but externally generated mga's themselves are incompatible with COLOR. I found that out first-hand on a Genesis project.

Most of the LUT inquiries on this, and most of the pro-sumer forums, are directed at presentation/display LUTs, intended to lift inexpensive computer monitors into "calibration class" grading stations. There have not been many calibration LUT inquiries, as filmout would be virtually unknown at this level.

I'd be interested in knowing if you have spent any time investigating the LUT capabilities of the AJA Kona3 I/O card. It has a couple of preset cubes, but as far as I can tell, they aren't relative to anything.

As far as re-mapping source values to output values.... I've always carried an internal model that every correction, every grade, is a sort of "temporary" Look-Up Table -- so when a project has been treated, a list of dozens, or hundreds, of "LUTs" exist for the content -- that manages it into its own personal space. Then we need to move the "whole house" into whatever the real presentation medium space might be. DLP projection, Cinema projection, Internet, SD Broadcast, HD, what have you.

Thanks for dropping by and hope to hear more from you.

Quick little proofreading thing....
check: "to be really accurate you'll need to be generation a Calibration LUT on at least a daily basis..." I think that should really read "generat ing..."

jPo

Yes, this interests me a lot.
I do have a Windows machine (we say that it is the mentally ill relative that we keep locked away in the attic like some Edwardian novel) and this would have been good for me last year when I was struggling with throughput on the Genesis project that I mentioned.

The challenges remain, however, as the desire for filmout won't go away -- but getting it sorted at a comfortable budget level will help re-stimulate that sector, at least for my market.

I've been away from the CML for a bit... did you re-brand "Praxis"?

jPo

Just a wee note to say I purchased the LUT Builder from Steve at the beginning of the year, and am successfully using it generate LUT's to grade a feature film at 2k which will be recorded to an Arri Laser next month.

Initially I was using Steve's Luts with my HDLink Pro. However although good, the Blackmagic HDLink Pro appears to mis-represent the LUTS with a bias towards red and adds contrast. I had to use a slight compensation grade in the primary out room of Color to get the image correct.

Now with Steve's new builder that generates LUTS for Color, it's purrfect, and what's more I use the Shake LUT out of Cubebuilder for all my cgi stuff.

Great product Steve.
Rgds

Toby

<rant on>

What's amazes me about this LUT issue is that the ColorSync/color managed workflow in still photography has handled output rendering very well since c.1994. An ICC profile for your printer/monitor/printing press is, essentially, a LUT. A map of a devices gamut and I/O response curves. Some current monitor manufacturers (e.g., Eizo, Samsung) that have dedicated pucks and calibration/profiling products are embedding LUTs in the monitor. The ICC monitor profile is stored in the ColorSync>Profiles folder is used by various software programs (Photoshop) for accurate rendering. It's an easy thing to work with.

What I don't understand is why making and/or acquiring a LUT for Color is so difficult. I've looked for a week and cannot find any end-user solution except for the one by Steve Shaw (and it only runs on Windows!). In the still photo biz, ICC profiles can be found, for free, from most inkjet and paper manufacturers. Making one for your monitor starts at $150. Heck, even making a profile for a film recorder was easy and accurate. Making a color-managed workflow is not difficult to do in photography. In video, it's considered difficult and attempted only by seasoned experts. I'd think with the proliferation of televisions, HDTVs, cable & broadcast the task of making a LUT for editing would be readily available. How 'bout ATSC provide various Rec. 709 LUTs on their website like the ICC does? That'd be nice. Or how 'bout TI provide LUTs for their various DLP projectors.

To ice the cake, ColorSync is Apple's baby yet they don't implement it in FCP or Color. I was stunned to see the pref setting for "Display LUT" and not a setting for "Output Profile".

Perhaps I'm mistaken. Perhaps ICC profiles are a much more advanced way of rendering images. They can be made 1-D or 3-D. They are designed for different rendering intents. They are accurate to 16-bit color (with 32-bit coming?). Perhaps the video/film industry is stuck with old technology. When I opened the LUT generated from Color and saw a simple list of numbers, it sure looked that way.

<rant off>

Oh, yes, amazing indeed.

I'm so glad to see somebody else wondering about this. Most of the time I'm a still photographer (and a physicist by training) and I thought I had understood color management. My displays are calibrated, my large format printer is calibrated and I never worried about colours. Then I started to play around with video color grading and I had to find out that even within the FCP suite the "Color" preview window, the FCP viewer and the Cinema Desktop preview all show the same pictures differently on the same monitors. I, too, spent a couple of days now to figure out what's going on and it seems to me that the video world simply evolved in a different way. Video was electronic before there were computers so everything centers around that external "Broadcast Monitor" that produces colors in response to analogue voltages specified in certain ranges (all based on YCbCr color space).
It is this reference monitor that all output is processed for and all renders you see in the different preview windows are probably supposed to be approximations. Unfortunately, the different programs seem to go about their own mysterious ways to generate those approximations.

Video people will tell you that you need to purchase a reference monitor ($5000 and up) to get dependable colours. Which is bizarre since the days of YCbCr-based CRTs are over and any LCD-based device will use some kind of LUT to generate RGB intensities. Obviously, there are only expensive high-end solutions (like CineSpace) to generate proper tables and, weirdly enough, those use the same rather cheap spectro photometers that we still people use to do colour management.
For now, I plan to dedicate one of my displays to video as "external" display using a Matrox MXO2 box (or a Blackmagic card) and adjust that visually using colour bars (and maybe my Eye One for overall colour temperature).

In some way it does make sense to adjust that "video only" display visually since, after all, the video audience will also look at it at some mis-adjusted output device. So at least having your own single external display takes out the guesswork about which of all the different preview windows to grade your footage to.
If integrating the proven colour sync layer is too much to ask - couldn't someone (XRite?) write a piece of software that would calibrate a decent LCD screen to video specs? After all, today's wide gamuth displays should well exceed the colour space needed.
Now that all DSLRs have amazing video capabilities and everyone is talking about convergence of still and moving pictures I wonder how that convergence is supposed to happen downstream of the workflow.

everyone is talking about convergence of still and moving pictures

Really?

jPo

Hi Steve,

when a 3d lut stores its values 'compressed' with a log curve, as can be the case for example with .cube luts, how exactly are they stored?

Is the maths for log storage in 3d luts the same for all 3d lut formats?

(Note that I am not talking about x->log conversion luts here, but the way data is stored in the luts.)


kind regards
fred

Hi Steve,

and thanks for your reply.

I have been trying to work out how to generate LUT's that take input values >1.0. The problem is that if I assume a max value around 10.0, then with a lut of size 32 I loose a lot of precision.

So I was assuming that one stores the data non-linearly,
like in this article, (see 'Nonuniformly Sampled Lattices').

http://http.developer.nvidia.com/GPUGems2/gpugems2_chapter24.html

How does other vendors deal with this problem?


P.S I did check out your software a while ago. Very cool! 🙂


cheers
fred