Correct ADX10 handling for ACES 2065-1 DSLR scan of film negative

Discussions - Using ACES Tech/Engineering
, ,
1

I’m interested in experimenting with Kodak Vision stock (50D, 500T etc) for still photography printed as ECN-2 and then ingesting the negatives into ACES using a DSLR with an IDT based on spectral sensitivity data for that sensor. This is just for a personal creative project and I’ll be using Nuke for all my color transforms.

Assuming a good accuracy with the camera sensor and that IDT, my starting point is a negative scan in the form of an ACES 2065-1 linear EXR. I assume I would then want to transform it using the ADX10 IDT, but I’ve been doing more of a generalized Cineon type of approach in my early tests.

As far as inverting the negative with the aim of placing the base emulsion as RGB Dmin 95 CV, from my initial tests with regular photographic film negatives I’ve had decent results in Nuke by either:

  1. Dividing out the base emulsion color from the raw negative scan and then inverting, followed by log to lin.

  2. Subtracting the base emulsion color sandwiched in between a lin to log round trip.

However since I’m not currently using the ADX10 IDT, I think these approaches are relying on the native logarithmic response of the printing density of the negative. ie, there is no attempt at “unbuilding” the film stock, or conforming to ADX code values - so there’s room for improvement, even if my stated purpose means that there’s a point where it’s “good enough”.

To understand more about ADX, I had a look at TB-2014-005.pdf in the ACES documentation that can be summarized as “essential notes on using APD and ADX”. My question then is, do I need to do a printing density to ADX transform before applying the ADX10 IDT?

And for that printing density transform, do I need to do a single APD to ADX10 transform as per page 14, or since this is not going to be scanned on a film scanner that encodes with ADX code values do I first need to do a Status M to APD Conversion as per Annex C in that document, followed by the APD to ADX10 transform?

Bonus points: within this home-brew DSLR negative scanning community, I’ve read about some people using individual red, green and blue LED lights along certain wavelengths to further split up the negative scan to get a better result than a high CRI white light and then I would recombine them in Nuke.

For example, the hardware portion of this article:

Any comments on this? It obviously would be closer to the result of a more high end RGB film scanner.

3 Likes
2

This looks right up my alley, although sadly I don’t have any concrete advice for you. What are you hoping to achieve by ingesting film negatives into the ACES ecosystem? Pairing film with digital footage? Do you plan on working with the ACES RRT or rely on the film’s response for your “tone mapping”?

Regarding the RGB lights idea, it’s an interesting idea. My experience with saturated lights is that my camera sensor is much more sensitive, and prone to blowing out, certain wavelengths over others, which would imply the lights would need to be at different intensities to perform at their individual ideal exposures. Also, it’s likely that any RGB combo will have spectral overlap, so by compositing them, I wonder if you’re adding more energy into the scan than there should be.

3

I’m sure this counts as a zombie thread here in 2022.
But I’ve recently become interested in the same area.

Curious if either of you guys ran with this any further?

4

Alex, and Ryan. Thanks for your interest. This was a thought in the depths of the pandemic that kind of died on the vine. But I have continued thinking about it, as well as answering some of my own questions as posted back then.

Ryan: I had a look at your online presence back then, and I saw some pretty interesting experiments you were doing in this space.

My updated view is to maybe not worry about unbuilding the film scan, since this is obviously rather involved, but instead mapping linear sensor data to a scanned film look as a kind of LMT with all its inherent characteristics left untouched - possibly via a log-log transform where the digital is pure log, but the film scan has the inherent s-curve on top of the log along with all the interesting cross-talk.

I also had the idea in late 2020 of applying Nuke’s machine learning to this task, and since then they’ve released CopyCat. This looks pretty good because the prior TensorFlow solution was based more around sRGB images, whereas CopyCat can deal with linear floating point and is obviously just generally more useful and mature for a non-developer like myself. I’m steeling myself to pay the backlogged maintenance fees and buy some beefy graphics cards.

As far as scanning the film in a more pure uncontaminated way with separate RGB records, I’ve looked into using a Foveon sensor so that it could be done in one pass and capture more of the spectral response than a monochrome bayer sensor. There was a certain amount of DIY hobbyist attitude in my original idea, and this looked like an interesting pragmatic solution - particularly with the later SD Quattro cameras that can write DNG files, where the DNG metadata could be ingested to ACES via Resolve/oiiotool/rawToACES so that we have hopefully a passably neutral linear image of a negative scan that can then be transformed to a Cineon log scan in basically the same way as it would be by an Arri scanner with no unbuild, retaining all the baked in film characteristics - as if we were still working in log in Shake in 2001.

1 Like