Inverse IDTs


(Matthias Tomasi) #1

Is there such a thing?


(Alex Forsythe) #2

yes, most are pretty easily inverted as they are usually a 1D function and a 3x3 matrix. Do you have have a use case in mind?


(Matthias Tomasi) #3

Hi Alex, thanks for the reply! The specific application that made me think of this is film recording of an ACES source master.


(Alex Forsythe) #5

@Matthias_Tomasi This gets a little tricky and isn’t officially supported in ACES 1.0. The issue is that going through the inverse of the ADX to ACES transform can yield some corner cases where the results can be quite ugly. If you think about the ADX to ACES transform as an inverse film model, it’s designed to predict the scene colorimetry that was in front of the camera when the film was shot from a particular film density. This works reasonably well. The issue going in the other direction is that the creation of density on the film depends not on the colorimetry of the original scene, but the spectral radiance. That information isn’t available in ACES. That said, if you’re not really interested in getting a negative that prints perfectly but is able to be scanned to recover your ACES file, this is something we’ve been testing. What exactly are you trying to do?


(Matthias Tomasi) #6

Thanks a lot for that answer, although it certainly opens up new questions for me… I take it then that the ADX IDT is more complicated than e.g. Alexa (1D curve and 3x3 matrix)?

My use case is indeed for striking a print but It wouldn’t need to be “perfect”. My idea was that the hypothetical ACES to ADX transform creates something which, when seen through the film print preview LUT of the lab that does the recording and printing, looks reasonably close. Then, with the director in the room, I would do a quick pass on those shots that stick out.

Alternatively I’d have to render to DCDM or Rec709, and the lab would apply an inverse recording LUT in the Arrilaser software.


(Jim Houston) #7

The essential part of what you are describing is that you have a close tie in with a Lab
that can predict what colors are going to result from a recorded negative. The color space is
so different in a subtractive system like film, that some colors will still appear to react more
strongly than desired. We basically left the system as – a ‘close coordination’ with recording vendors
is required. The Rec709 path is marginally safer, but it can be limiting in the look achieved.
The wider gamut of P3 is achievable in part with the right recorder LUTs for conversion.

ADX to ACES is not really an IDT like Alexa, but the principle is similar. In this path, there is an attempt
to ‘unbuild’ the film back to scene relative exposure. An approximation of the average film performance in terms of interlayer cross color effects is used in this process. Going the other way is similar – ACES to ADX – there is an attempt needed to simulate the effects of the film dye interactions so that then a film print preview would get reasonable close results. Closing the loop though between the hypothetical model and the actual output on a recorder is really a job for the house doing the recording that depends on what aim densities are being calibrated to on the recorder.

The inverse path can work as well, although there are sometimes bit-depth resolution issues that can cause banding in an inverse path. The important part here is that it would be better to have made the grading decisions for Rec709 rather than the wider values in a DCDM. The results will be a little easier to get in the smaller gamut.