Lets talk about LUTs...

B

Brian_Broz

Active member
Hi,
Can somewhone describe the purpose of LUTs (lookup tables) for calibration on set and through the production chain? Forgive my ignorance, but I am hearing the term LUT thrown around a lot lately...and stilll don't understand (fully) the different options for colorspace and how to keep things consistent thru production. I understand the importance of keeping consistency in terms of the waveform monitor and vectorscopes, but how does colorspace relate?
I have been working in video for years where it was a betacam and an 8044/45 Sony for color reference:-) And 12-15 NP1 batteries...but that's another story :-)
I hope I'm not the only one on these boards that doesn't fully understand LUTs. I'm curious, as I've heard of guys setting up custom LUTs before working on a series....but I feel like a PUTz for asking.
I'm hoping someone can chime in with their wisdom and experiences. I'm trying to know all this stuff before those RED cameras arrive!!
Thanks for any help.


Brian Broz
 
Hey Brian, don't feel so bad. I've been around a while and don't thoroughly understand the need for multiple LUT's. In all the film xfers I have attended, LUT's were never mentioned, until the displays started changing from old fashioned SMPTE CRT's to the unholy number of display options now available, including film-out.

Why for instance the Arri D20 needs it's own LUT? Is it essentially a sort of color correction tool to bring the camera image into some kind of "normal" setting?

I see on hdforindies a mention of "nondestructive vs. destructive" LUT's in regards to the Silicon Imaging camera and Iridas. Why would anyone WANT to work in a destructive mode?

I have a vague understanding, but now LUT's are thrown around like, well, filters in a matte box. It seems like you need a box of them to work.

So how do multiple LUT's help, what problem are they solving?

When using custom LUT's is there an analogy to choosing a warm card to white balance on?
 
Lookup tables can also be called by the more colloquial term "Palette" and comes from the world of indexed colour. Essentially you have a table with one or more entries. This consists of the entry number and sufficient space for the length of number to represent a colour for the full colour model. In a simple 8 bit RGB model this would mean 24 bits for the colour model and one or more bits for the entry number into the table depending on how many entries you want for the table.

The actual colour value for each pixel in the source image is replaced by the table entry number to get the original value you use the entry number and "lookup" the actual colour value from the table. Destructive tables tend to be smaller, have fewer entries and may use adaptive models where several really close shades of a colour may be replaced by a collective intermediate value and a single index number.

LUTs were used as a simple form of compression for file formats like PICT, PCX, TGA, GIF. However, they can have quite a few uses and for Digital Video you can apply them in a whole slew of ways to provide custom remappings for looks, correction curves etc. etc.
 
My understanding of LUT's has been that every electronic picture generating device, whether a camera or a computer video card, has it's own native LUT.

A "custom" LUT would remap those values to a different color range. I can understand using that as a way for instance to adjust a raw image so that it is displays correctly within the native color space of a monitoring device. That would apply to working in log color space when doing effects on a scanned film. You would remap the display output to linear to see what you were doing on a regular monitor. But it would leave the the original log file untouched, only changing the output to the display.

So, does every source file type, whether a D20, a Dalsa or Red need to have their own set of LUT's to cover every display device you might use when field monitoring? Would there be another set for the edit system?

And how would using a LUT as a destructive filter on the original recording be better than doing the color work in a purpose built grading system?
Or do the grading systems do auto converts to various preset LUT's?

On the Iridis system, the LUT's can be embeded non-destructively in the metadata. It's a way to pass on the intent of the DP and still leave the raw footage intact. Would the grading system import the LUT as a starting place, converting the look defined by the custom LUT into the color space of the grading system? Would the colorist then be free to tweak from that starting point as normal?

On the set, would a tech and the DP do rough color work on the Iridas using typical types of color control tools and then does the Iridas convert that result to a LUT as an intermediate tool to get the information to the grading system?

Is there then two purposes for LUT's?

1. To alter or force an image to display correctly on a particular display device which may not natively be accurate to the source material.

2. To essentially use as a destructive color altering tool like a matte box filter, or a non-destructive communication device between the set and the colorist.
 
Think of a lut as a curve in photoshop. You can do a lot with them - log to lin or lin to log, a gamma curve, a constrast s-curve, brightness, contrast etc. etc.

Colour space conversion etc. is usually handled by a colour matrix. Put a matrix and a lut together and you can do pretty much, if not all the colour image processing you'd want to do in a camera to achieve a nice looking image, and also take it wild enough to envisage a great multitude of "looks".

All the above is handled by metadata in the RED system, non-destructively if you're shooting raw and can be edited in camera or in REDCINE.

Graeme
 
Some good answers here. I guess my question is what type of LUTs are you talking about specifically?

LUTs can and are used for multiple purposes.

1) In camera RGB manipulation. RAW files consist of image straight off the CCD (or CMOS) sensor(s). The thing is that the data coming straight from the sensor doesn't have any specific value until one is given to it. As Graeme has said, it basically acts like a photoshop curve which can manipulate the way RAW data is interpolated to produce a final RGB image.

2) Matching disparate hardware display and capture devices for accurate color. In post production houses that deal with very accurate color requirements (grading etc) these LUTs are used to map one set of RGB values to fit inside another either smaller or larger RGB colorspace.

Contrary to what many probably think there isn't a single Red, Green, Blue colorspace. Monitors use differing phosphors which give different RGB responses, different films use different dyes which have different responses, different scanners have different CCD responses etc...

LUTs are used with film to ensure that the grader sees what the footage will look like when output to film again. By knowing the colorspace of the filmstock, the scanner, and the monitor, the film colorspace can be remapped to fit the smaller colorspace of a monitor in such a way that the monitor can display a roughly accurate example of what the footage will look like when scanned back out to film. There are many ways of displaying a larger colorspace in a smaller one (none of which are perfect by a long shot).

So, in short, a LUT is simply a way of expanding or compressing data; whether that be a full RGB image or the values of an image straight off the sensor. Think of LUTs and color matricies as choosing a film stock. As for the difference between 'destructive' and 'non-destructive' LUTs - the former is 'burned' into the image (think of your standard video camera) the latter is not and can be thrown away without affecting the RAW data.

Hope this helps and isn't too full of mis-information =D
 
"manipulate the way RAW data is interpolated to produce a final RGB image."

Luts generally happen once the RAW data is converted to a RGB representation via a de-mosaicing operation. Just to fill in the gaps incase anything thought that a LUT turns RAW to RGB.

Good point about the multitude of RGB spaces!! There's lots!

Graeme
 
"Non-destructive", as some have pointed out here, means the LUT information is metadata that can be modified or removed, and is a sort of "layer" above the true RAW information. In this layer-based approach, the RAW data is never modified. Furthermore, our LUT's are floating point, so there is no clipping if values go above 1.0 (white) or below 0 (black) with the LUT.

"Destructive" is actually like Photoshop. If you apply a curves or some other color filter, and save and close the image, the next time you reopen the image, there's no way to return to the information before that color-correction was applied. The LUT permanently modified the image. Furthermore, since Photoshop does not support floating point in it's "normal" modes of operation, if you apply a color filter, and clip image information in that filter, there's no way to retrieve the data in the clipped areas.

So that is what we mean by "non-destructive" in our IRIDAS/SI workflow . . . there's actually quite a lot of programs/workflows out there that work in either integer-based bit-depths (8-bit, 16-bit), and clip information above or below their max/min values, or don't allow for metadata-based manipulation, so once you apply the correction to the image, it's permanently "baked" in. Our floating point-based CineForm RAW engine with its metadata-based color-correction LUT's avoids both issues.
 
Last edited:
Do we mean different things went we decribe a LUT that creates a color-graded "look" on some footage than say the output LUT that calibrates a monitor or projector to display the colorspace of rec 709 or the DCI Spec?

It seems like a color-grade LUT has to be bounded by the set described in the output LUT whereas the original RAW file might contain values that are outside the display LUT we are using.

How do we know the rec709 LUT looks the same on different devices? Or is it that a device needs it's own specific LUT to make sure it is displaying the color gamut decribed by the 709 color space?

I feel like it may soon be the right time for a Video Color Spaces for Dummies book. Does anybody know of one?
 
Well, it would be nice to understand a workflow just for the LUT's.

From on set use including calibration of reference monitors and creative grading of the images. Post production, various display problems encountered through the post process plus final grading, passing LUT's to FX people, etc.

Just how many LUT's are you going to encounter in the brave new world of a pure data workflow?
 
Luts galore and luts at plenty. In the RED, there are 3 kinds of luts - lin to log / log to lin, "look" and output. I don't think you get to choose the lin to log / log to lin, but you do get to choose what lut you want for your look, and what you want for output, say rec 709. In REDCINE, I think there will be options for luts that match to film or whatever also.

Graeme
 
Apple's Colorsync for Film

Apple's Colorsync for Film

This may help one understand the need for LUT's:

http://en.wikipedia.org/wiki/ColorSync

Apple has tried to maintain some way to guarantee color accuracy for printing, but as Wiki says "Human color perception is an incredibly complex and subtle process" and film has many steps between the image and the projector where things can get out of whack. LUT's are whacky fixes.

Besides how do I know the Red you see is not dark pink in my brain?

Rosco
 
Rec 709
Ok, so then would every digital camera, whether Red or another, require it's own set of LUT's for every display type you might use in order to achieve rec 709? Or would the camera companies give you a limited set of LUT's and force you to use displays they designate?
 
In camera we transform from, say, camera RGB space to REC709. Once it's in REC 709, any monitor properly calibrated for REC709 should show the image the same. Similarly, if you're transforming to Adobe1998 or sRGB, any device that knows this, or is calibrated for it should show the image correctly. It's all about transforming the unknown - ie, the RED camera, to the known, ie 709.

Graeme
 
Here is a better resource for understanding LUTs:

http://www.theasc.com/magazine/april05/conundrum2/page1.html

"The Colonel has his recipe, Coca-Cola has its formula, and DI facilities have their “secret sauce,” which undoubtedly includes proprietary Look-Up Tables (LUTs). Most projects that are shepherded through a DI strive to maintain the CMY filmic look and quality, but you don’t see that naturally on an RGB color space display."

Happy Reading,

Rosco
 
Jason Rodriguez said:
So that is what we mean by "non-destructive" in our IRIDAS/SI workflow . . . there's actually quite a lot of programs/workflows out there that work in either integer-based bit-depths (8-bit, 16-bit), and clip information above or below their max/min values, or don't allow for metadata-based manipulation, so once you apply the correction to the image, it's permanently "baked" in. Our floating point-based codec engine and metadata-based color-correction LUT's avoid both issues.
Click to expand...
Jason, are you with Iridas? Your comment drew me to the Iridas website where there is some helpful information. Is SpeedGrade DI a full color correction workstation ala Lustre, etc? Most of the Iridas products are priced quiet reasonably but there is no price for SGDI so I assume it's a full-on DI color workstation and thus pricey. Does SpeedGrade On Set work with other color correction systems or is the data completely proprietary to the Iridas pipeline?
 
Jason Rodriguez said:
... Our floating point-based codec engine and metadata-based color-correction LUT's avoid both issues.
Click to expand...

Hey Jason, Please add our name where appropriate. So "Our floating point-based codec engine" should read CineForm RAW. ;) I'm not doing all the 3D LUT work for some "other" camera.
 
LUTS and RED

LUTS and RED

A different perspective on the question....

A LUT is simply a Look Up Table. A device to map one value to another.

As an example of using a LUT, consider the Gamma Correction that must be applied to the straight line (i.e. linear) response of an electronic sensor so it compliments the non-linear response of a CRT monitor to provide the optimumal TV viewing experience.

Video cameras apply this gamma correction LUT to the video before it is recorded - i.e it gets burnt in. If you now take that video to a film out, you would probably wish that you hadn't used that particular LUT, as film has its own non-linear response. To optimize for the film viewing experience you would want a different LUT.

Same thing happens with in-camera color correction - do you want to do it in-camera or in post? Your in-field choices can effect your post flexibility.

A better approach for these kind of problems - especially when you have higher dynamic range recording systems (i.e 10 or 12 bit not 8 bit) is to apply a LUT into the camera monitoring paths only, so that the correction can be previewed, but NOT modify the recorded data. Then re-apply these preview choices in Post. You have created yourself an UNDO. And if you have an UNDO nothing has been "destroyed" / irreversably altered.

It does not matter if your recording is RAW or RGB, the priciple holds. In the RED workflow, we have just this capability for RAW and RGB; i.e. LUTs in the monitor path only (RAW and RGB recording) or LUTS in the monitor path AND record path if desired (RGB recording).
 
You must log in or register to reply here.
Back
Top