REC709 Sources, using ACES

Tags: #<Tag:0x00007f964cb04c50>

(Bruno Munger) #1

Hi All.

First time i post here.
Since ACES does not have an official IDT for REC709 Camera sources (i.e. drones, GoPRO, etc…) What is the recommended practice for this type of cameras? Just to be clear, i am talking about Scene referred ungraded camera (not 709 graded material). Obviously, a straight conversion from (as an example) Rec709/Gamma2.4 -->ACEScct/AP1 will give you an image but not a proper one.

I am curious what are studios/Post houses doing in this case? Do folks create custom IDTs per Cameras, for every projects? I.e. Shoot a Macbeth chart for every camera under multiple conditions ?



(Jim Houston) #2

Hi Bruno,

Good to see you here.

The Rec.709 encoding equation for cameras goes from scene-referred to a Rec.709 encoded signal, then often, to get an output-referred signal on a monitor, a simple gamma is applied (now 2.4 in standard Rec.1886) Gamma 2.2 is sometimes used as a legacy to look at material and sometimes a less aggressive unbuilding can be helpful is the colorist work is still to be done… A simple gamma curve doesn’t work perfectly because in 709 there is a straight line portion going down to ‘0’ from a knee point. Use of puregamma always need a slight adjustment near black.

So now I can try to answer, for ungraded material straight from a ‘unpainted’ camera, the inverse of the encoding formula from Rec.709 (which is readily available) should be used in the IDT to get to linear and then matrix from Rec.709 to AP0 — then you can convert to ACEScc/ACEScct.

If you are working with Rec.709 that has been graded, or ‘painted’ in the camera with hue/con/bright, then inverting from the output-referred image is the right thing to do (the inverse of the Rec.709 ODT).
This puts you in an artificial ‘linear-appearing-to-the-eye” space where you still need to stretch the resultant black values to a ‘0’ point.

The downside of either is that you only have the range available that was in the original encoded image, and there is no guarantee that the noise characteristics of the image after these contortions is still going to be satisfactory. A benefit here of ACEScct with the tone curve at the bottom can serve to compress the noise region of an image.

So lastly, on the direct question of getting different cameras into ACES on a project – the main customization is getting the right dynamic range from these cameras to map into ACES in a consistent way (in which a valid 18% chip is an essential component) AND is the color actually Rec.709 consistent. Many times it is not. So having a grey ramp image and a known 18% card in consistent lighting is the best control in a practical sense for these cameras. The MacBeth can be a good color check for consistency of conversion, though by itself, there are insufficient colors and grey range in my opinion.

The Macbeth is a method for generating an IDT. A larger range of colors like DuArts can help in creating a better matrix if you have separate calibrated measurements of what colors they are supposed to be. In this case though, you may have to assume no knowledge of color space and derive a new matrix from ‘X’ to ACES as a linear optimization of source to target for rows of measured color data compared to the desired target color for each row. Then use this new matrix in place of the Rec.709-to-ACESAP0 matrix.

The third grey chip from the right is the middle grey value and that should get ‘close to’ 0.18 in ACES. If it falls between 0.15 and 0.19, I leave the image alone and let the colorist adjust, if outside that range, I would consider the exposure (ISO) to be off and would do a technical grade to bring it within range. (no different than what used to be done on film scans all the time).

Hope this provides enough hints about the possible complications. Because there is more variability in the Rec.709 cameras, many of which are from the consumer market, there are sometimes specific changes that are needed in an IDT to bring it into the ballpark. An example of this, is that some cameras from the still world may build in a reproduction curve for print (about a gamma 1.1-1.2) which has to be removed to get to true linear. There is where eye judgements of the result is sometimes needed, because it is reverse engineering of an unknown process and when is the conversion ‘good enough’.

If you just want to know, is there an IDT for Camera ‘Y’, that is a different question. :slight_smile:


(Cacho Briceño) #3

Hi Bruno and Jim.
I am an editor and layman in this aspect of the color business but I need to ask you a question: The problem is that every camera has its own color behavior. Which is a consequence of the way it camera was built and the processing it does to the image after the picture has been taken. That’s why, each brand of camera and model, should come up with an exact and proven IDT made by the camera builder. So every body wold trust the conversion to ACES. Do you think, if that would occur, it would help to fix the confusion.?
Thanks a lot, guys