RED colour space workflow

[Hugh]

We're waiting until someone we know has a RED camera, at which point we'll set up some test shoots. We're not getting one ourselves (we're a smallish London film VFX studio), but we want to make sure that we are both prepared when a client comes to up having shot on a RED, and that we know enough about it do decide whether or not to hire one when we need to do our own shoots.

 

[Cüneyt Kaya]

Not entirely (you knew that was coming, didn't you? )

If you do a filmout to actual film film, the characteristics of the film will make your images look quite a bit different from your rec709 look.

I wrote a bit of stuff about colour spaces (although more from a film VFX standpoint) on my blog - you can find the post here (I don't update that blog anywhere near as often as I should)

Anyway, if you have a look through that post, you'll see the image I used in Log, Linear, Linear-with-a-2.2-gamma and Log-passed-through-a-film-lookup.

If you wanted your DVD to look the same as your footage when printed to film and projected, you'd have to have a 3D LUT that would emulate your film print. You could then apply this to your images before creating the DVD. (They may, of course, require some kind of adjustment after, but the 3D LUT would at least have got them into the right space)

your blog was really informative

so there is gamma-corrected linear for tv/video
log/ linear for filmoutput.
lin image Log-passed-through-a-film-lookup. (TRUElight node)
you used a lin image with shakes look up table.
i guess for VFX it is better to work in linear, then add a log- passed- through a- film- look- up

you made me more curious about color spaces.

1.
is rec709 gamma corrected linear or linear?
2.
After answering nr. 1,
what is lin or gamma corrected linear in redcine?

ok one more question:
if you finish 1st in HD, and later (lets say 2 years), you want to make a filmoutput.

a)
finish hd with gamma corrected linear, then grade the whole thing again after 2 years.
b)
finish hd linear with log-passed-through-a-film-look-up, then grade everything in log after 2 years.
c)
finish hd linear with log-passed-through-a-film-look-up, then take take your dpx and output directly.

man really complicated with the colorspaces.

 

[Hugh]

your blog was really informative
so there is gamma-corrected linear for tv/video

This is because any display device doesn't output exactly the values it is passed in. If you double the colour value that you're passing to your monitor, you will not be doubling the number of photos that are being thrown out by your monitor.

log/ linear for filmoutput.

Well, filmouts are almost always done as Log. I'm sure some places will accept linear files, though.

lin image Log-passed-through-a-film-lookup. (TRUElight node)

The Truelight node, by default, takes a Log image. Our pipeline, for example, is set up so that you can easily view Linear or Log images through the film lookup - it'll convert Lin to Log internally before applying the Truelight lookup if that's what you're passing it.

you used a lin image with shakes look up table.

Not sure where I did this bit...? The gamma-corrected-linear image was done just using a Gamma node...

i guess for VFX it is better to work in linear, then add a log- passed- through a- film- look- up

Yes - definitely. Most of the major operations that you'd be doing in VFX assume that your data is linear. Comping in log can give all sorts of problems, as can working with a gamma curve on the images.

1.
is rec709 gamma corrected linear or linear?

I'm not sure - I was trying to find the definition of rec709 - you'll notice that I never actually really mentioned rec709 in my post - I'll need to find out what it actually means. I suspect it's a combination of a gamma curve and a predefined white point and colour space definition.

2.
After answering nr. 1,
what is lin or gamma corrected linear in redcine?

Lin is setting the gamma (in Format Settings) to Custom (1.000)
Gamma corrected linear is Custom (2.200)
If you select sRGB or rec709, you'll see the gamma value (greyed out) be set to 2.4 or 2.2. These two aren't exact gamma curves - if you look at the Wikipedia entry for sRGB, you'll see that it says that sRGB has an overall gamma of approximately 2.2, but is actually a linear 1.0 near black, and 2.4 elsewhere.

ok one more question:
if you finish 1st in HD, and later (lets say 2 years), you want to make a filmoutput.

a)
finish hd with gamma corrected linear, then grade the whole thing again after 2 years.
b)
finish hd linear with log-passed-through-a-film-look-up, then grade everything in log after 2 years.
c)
finish hd linear with log-passed-through-a-film-look-up, then take take your dpx and output directly.

If you shot the whole thing on HD, then you'd probably work in HD-native colour for the whole project, and then get it onto film at a later date.

If you shot it on film, for an HD finish, as a lot of TV shows do, then I'm not sure. We did VFX on a show that worked like this, and we were delivered film scans as Log images. We were also given a grading reference, and what we sent back to the clients had this grade on, which would have been useless if they had tried to put this to film. But then again they probably never had any intention of doing this.

There are, I'm sure, many people with much more experience than me on here, so I'm sure there are people who can either clarify/justify what I've said, or correct anything that I may be mistaken on.

Oh, and I'd love to find the definition of rec709, if anyone knows it.... Also where the chromacity points of the camera RGB lie.

(Looking ion RedCine, the chromacity points of rec709 appear to be pretty much identical to those of sRGB)

 

[Graeme Nattress]

REC709 is a colour space and a tonal curve. The tonal curve is defined as a power law curve with a small linear section near black.

Best place for definitions is Poynton's book.

Graeme

 

[Hugh]

K.Berlin:

Poynton has a load of great stuff on his website.... It's a bit confusing at first, but if you're interested about a lot of this, do have a look at both the Gamma FAQ and the Colour FAQ, both linked from this page. He talks about "Linear Light" which is what I meant by Linear earlier. "Linear Light" stresses the fact that it's the physical light hitting the sensor that is being stored linearly (and it saves confusing the video people who have "linear" which is actually gamma-corrected)

Graeme: Thanks. In this wonderful world of web, there's usually some reference page floating around somewhere - the details are actually in Poynton's FAQs. If I did more stuff with video and less with film, I'd get the book....

 

[Graeme Nattress]

The book is well worth it. It's my desk reference, along with W&S for colour science.

Graeme

 

[Cüneyt Kaya]

hi graeme,
this topic was discussed over at projectred
http://www.projectred.net/forums/viewtopic.php?id=39

walter volpatto thought a little bit about this issue...
i will quote him. dont know if it is unpolite to do, if it is...sorry.

quote:
Good day to everyone. Sorry if I'll be long.

Yesterday I had the first look at some images from the RED camera. After some shooting the class of AFI DPs came in my room and they want to do an evaluation of the material and have a discussion about it.

The material was converted and loaded ( DPXs files ) into a Quantel Pablo machine ( by couple of coworkers ) and projected in a DI film-calibrated environment for quality and color judgement.

I was less than impress: hi-light burned, lack of details, noise and other bad things on the image. As soon I try to do a color correction trying to match a quicktime on the Apple of a low resolution it seems he has the HIres and I have the crap low res.

I couldn't believe it and I didn't.

After everyone left, I did a bit of research and I found the original material, the Quicktime generated ( by I do not know who ) and the software REDAlert v.1b1 BETA that i installed in my MAC and start to poke around with the software and the material and compare what I got with what the material really could look like.

The material it is a LOT better than what I got in the first place but the conversion was a lot of button pushing by someone without keep in consideration the final color spaces target involved.

And here is where my post really start.

Lets start from another project and another camera.

I had a pleasure to work with John R. Leonetti in the movie "I Know Who Killed Me" and he used the Genesis camera. The first tests I saw prior to that movie where not exciting but as for the RED camera nobody knew how to handle the material.
I had few simple challenges:
1) simulate a film stock for the final output, in this way the DP can light the scene without think about digital/film issues.
2) provide a comfortable viewing environment in the set ( I partially failed to do this in this project but I corrected that for the movie I'm involved now ).
3) provide Dailies who WILL MATCH THE FILM as best as possible if the material is printed at LAD in the lab and the new tape has backtrack the original Tapes for conform.

And some others issues now worth to be mention here.

So I asked John to shot (in the same light controlled environment) a costume scene with the genesis camera and a film camera with the film he intended to match and we went trough to do a side by side comparison in the DI room. Was not too bad. Decided which color correction to use to match the film we film-out the material in a vision stock and was matching the film quite well. In that way when the Dp on the set place the light can rate the set "as if" it was using a regular piece of film.
Then I generated a LUT who will transform the original raw tapes in a Rec.709 color space through a realtime LUT and this LUT will contain the genesis raw to film log space and the log to Rec.709 conversion. needless to say what you see in that HD monitor has to match the DI projection room and the Film projector as close as the tolerances of the different color spaces are allowing.
The missing bit is a LUT loaded to the Panavision gamma box who can ( on the fly ) grab the output of the camera and provide a proper output on the set ( now this problem is solved. )

so, given a tape with the genesis camera material in it, all the output look alike and the quality it is preserved due to the minimum transformation involved between the media ( raw ) and the final output ( Film logarithmic color space ). All the heavy transformation where for the HI-Def dailies or for the simulation on the set.
Given that EACH LUT was builded starting from the Film-Log color space emulation in the DI room, no illegal colors where allowed in the dailies so nobody was seeing things that later on where not possible to achieve in film.

Therefore we have the three point above with the conversions:
1) [genesis raw] {process} [transformation to Film-Log space]---->[simulation in DI room in DLP color space of the vision stock] ( for film-out lets call all of this [FILM LUT] )
3) [FILM LUT] + [Rec.709] for dailies
2) [FILM LUT] + [Rec.709] - [Gamma loader 3D_LUT type] for set confidence environment.

As you can see there is a strong correlation between all the color spaces involved and the RAW material is touched only one time and never left the log color space.
For whoever want to have a "cool look" and not just a "Film Look" I suggest to see the final movie: John Daro ( the actual colorist, I only supervised his color ) pushed really hard the style.

After this experience I can piece together what went wrong with the RED camera.

the material, loaded in the REDAlert software was converted by someone from the R3D to quicktime pushing a certain amount of buttons to have something he/she liked. this is a raw to rec.709 conversion direct, without keep in mind the final output constrains ( film ) nor from where the material come from or where has to be in order to do a color session later.

Then, another person, having no reference whatsoever of the first conversion, did a second conversion ( very flat ) in the attempt to have a better latitude for the manipulation in the Di room ( at 4K ). unfortunately it choose to output the material in Rec.709 (DPXs) as well.

This mean for me grab material who as been converted and re-convert it to log color space, with all the noise coming out from all over the place and color that the film simply cannot hold.

As i mention before, the material it is lot better than what I saw and the problem of the workflow is how to handle the spaces conversion.

I have a solution in mind but at the moment the software REDAlert is not letting me do it, but I want to post it to see:
a) someone has a better idea
b) there is a software that can do that

the main problem is the branching of the color spaces: from raw to rec.709 and from raw to Film Log color spaces. There is no correlations and your dailies will not be representative of the material printed in a standard condition in film.
.......................................................................................................
what I need is this.
imagine for a second that the REDAlert software can load a 3D_LUT and that the LUT will be AT THE END of the math process.
I calibrate my Mac to Rec.709 ( or a different output, better if I have a calibrated monitor output ) and I grab my raw file.
I load the 3D_LUT who is the combination of the [ Film-Log simulation to DI color space]+[Rec.709 conversions]. The first will depend on the lab you're printing but for the sake of exercise I'll assume there is a average standard we can safely assume valid.
Now, when I manipulate the picture with the REDAlert software I'm actually working in Film-Log color space and the 3D_LUT is helping me to view in a environment who is simulating the final print.

Then I've three deliveries I have to produce:
a) Quicktime dailies, simply leave all the LUTs on and bake it in the quicktime file ( may be a simpler version trough am ICC profile or something similar )
b) 4K pictures for DI environment, turn off the 3D_LUT simulation and render just the REDAlert software part then make DPXs
c) store the REDAlert files settings for future use with the project.

All the deliveries have the same look and the same is saved for tweak if necessary.

The Log mode and the Rec.709 mode in the REDAlert software seems not to give what i need at the precison i need...

Am I missing something?
......................................................................................................
Walter Volpatto
DI Artist, Fotokem.
Current working on "Not easily broken" : quote ends
------------------------------------------------------------------------

will ra/rc do this or do you know a workaround, or is it implemented and i dont know how to use?

 

[Jeff Brue]

We've had some of the same issues ie being unable to recover a shade of Red in any kind of LOG workflow, but that can be rendered out at Rec 709 out of red alert and if viewed at Rec 709 its there and looks decent. The same thing rendered out in any of the forms of LOG means that shade of blown out Red is irrecoverable. Its a problem when a major DP at home pulls a shade of Red that you can't. The solution, well banking on my experiance with the Phantom and DSLR files we are going to have to color correct the R3d files natively. I was able to get that shade of red out of scratch with a film LUT, something I couldn't get out of a Davinci or Nucoda even with 16 bit Tifs.


Jeff Brue
Post Supervisor
Digital Film Company

 

[Seth Larney]

This is extremely interesting..

I am currently upgrading our post facility to centre around Scratch, our in-house projects will all be Red based which is what has initiated this change.

One part of this solution may be :

I think that there are plans for the Red workflow to include LUT's (or RED versions of LUTS) to be applied via SD card to the camera, to be added in realtime to the SDI/HDMI outouts for onset viewing, which then should be able to be added down the track as well for the DI. Which I gather is what Walter is after.

Jeff, have you spoken to Graeme directly about why this may be happening ?

One of the main reasons that we are looking at Scratch is to be able to conform in native R3D and also then apply a LUT and grade in native R3D, all the way through, hopefully eliminating colorspace issues at least partially.

What I would very much like to know, is whether I can load LUT's from Scratch onto the camera for viewing on set ? (I don't think so) Or if not, how exactly we create a consistent viewing space between onset-DI-Filmout with Scratch as the DI backbone.

Cheers,
Seth.

 

[Jeff Brue]

Well if you think about it in reality the color really isn't a color its the result of an algorithm and changing that algorithm can drastically change/effect the quality of an image.

Example: last night I was working on a Phantom shot that exhibited aliasing in the red channel. By adjusting how the algorithm reconstructed the red channel from the bayer data I was able to reduce the amount of aliasing.

So here's the question what really is your image and what should you set your base look in. I can only equate it to opening a RAW image in photoshop, ie setting the ISO, adjusting the tonality. Well what you're actually effecting there is the result in 32 bit float of a debayerd rgb image with adjustments to the algorithm based on the desired results of the color.

So the question then becomes is it really fair to take a Red and chop it to 10 bit log and put in stasis the effects that effecting the algorithm allows? Because putting a naturalised image like Red through a film lut can be a harsh process. In any case thats the question I'm faced with right now.

 

[M Most]

So the question then becomes is it really fair to take a Red and chop it to 10 bit log and put in stasis the effects that effecting the algorithm allows? Because putting a naturalised image like Red through a film lut can be a harsh process. In any case thats the question I'm faced with right now.

But the fact is that in a multiple delivery platform world - which we are likely to have for many, many years - there are always going to be color choices that are valid on one platform and out of gamut on another. It's a simple fact of life. Film prints cannot be made to reproduce every color that is possible in Rec 709, even though Rec 709 is itself a restricted gamut. And neither one can reproduce every color that is theoretically possible in digital cinema - which is more limited by current projection technology than the specified color gamut.

Converting values between linear and log scales is not in itself a severely limiting factor. Using a 3D LUT to mimic one particular delivery platform's color space on all other platforms is. However, that is why in most cases one platform is chosen as the primary delivery target and all other formats are derived from that. In today's world for feature releases, that primary delivery target is almost always film because that is the primary method of delivery at this point in time by far. In time, digital cinema will likely emerge as the primary target, thus providing for more color bandwidth. It's used that way for some animation releases now. But it's really not worth lamenting the pallette restrictions imposed by targeting to the color space of film prints when that is the primary method of delivery and is therefore the major determinant of what the vast majority of your viewers will see.

Besides, I think that in general, the HD and other video deliverables generated from a properly done DI process look pretty darned good, whether they're technically color restricted from their original capture or not.

 

[Jeff Brue]

Mike,
Actually I'm not lamenting the palette. I'm lamenting being able to achieve a color that is achievable by debayering in another debayering palette. Lol its a warm uncomfortable seat when you've stayed up all night trying to bring a blown out red out of a sign in a log mode of redcine when the DP can get that same shade of red out of a rec 709 render, and you can as well. However it means throwing away a lot of info in the shadows, that he can see is there when he adjusts the levels himself in red cine. To me that means that something kind of large is being tossed away when applying any of the default log curves.

 

[Graeme Nattress]

I can assure you that the log curves don't toss stuff away. All matrix math is done linear light (as it should be) then whatever curve you choose is applied. At that point, you could apply an inverse curve and get back to linear light without loss (other than the limits of precision of the output format if you go out 10bit).

Now, I'm pretty sure DI apps don't ALL treat log data with full respect and often they expect a limited range of data and ignore data above or below that.

I'm not sure why you're mentioning debayering as that seems to have nothing to do with your issue.

Graeme

 

[Deanan]

same shade of red out of a rec 709 render,

Can you elaborate more specifically?
(rec709 colorspace or mapping/lut? , or log through a print lut vs 709 output?)

 

[Jeff Brue]

[/IMG]

In the upper right hand side the Hilton sign on any log based transfer using red log pd log 685 or 985 is lost and blown out to white. Yes an inverse curve was written here based on the Scratch LUTS that detail red log values for transformation back to linear light, from there we applied our standard batch of viewing LUTs to try to recover that sign.

The above image is the Rec 709 render but it did not have as much room in the shadows and color detail is lost as well as the noise pattern is much cludgier.


As far as conversation about the debayer settings and how they can be used to manipulate color, I don't think I'm being terribbly clear my apologies. The general statement I wanted to make is that I and the colorists here believe there is more definition in coloring by natively color correcting the r3d files than by going to and color correcting off of any pipeline usable uncompresed file format.

 

[Jeff Brue]

Oh and just a note this is from Claudios test using our f23 as well. The camera holds up damn well for dynamic range, I'm just trying to figure out how to get that range into a DPX file. In any case Graeme this footage is being shipped to Jarred as i type this if you want to take a look.

 

[Graeme Nattress]

Thanks Jeff. I'm up here in the cold north, but I'll see if Jarred can get a key file to me to look at.

Graeme

 

[Deanan]

Which log2lin luts did you use for scratch?

 

[Simon Blackledge]

Doesn't Scratch now have a red lut ? Or do you have to cook your own ?

s