#apertus IRC Channel Logs

really cool experiments with the deriving of the sensor calibration with the spectral charts off the sensor page.

regarding going with a linear transformation 3x3 vs a fancier thing, this is mostly a limitation of CinemaDNG and the ecosystem of software available to people working with media files for cinema.

if we want to try supporting more advanced calibration cases beyond a 3x3 into XYZ space, we pretty much have to write our own software but also ask that users only use that software to process the CinemaDNG files, which is pretty much not going to happen.

for now I think the straight-ahead path is to see how far we can get with a linear 3x3 calibration and conforming to CinemaDNG spec and then putting a lot of effort into the color science after the calibration step: establishing a "native" gamut, the log curve, etc., all of which can be used without issue in industry standard software provided we make the LUTs and/or shader/DCTL files

there might be something I am not considering so let me know if I am missing a piece of the puzzle :)

I suppose the only other option I can think of is providing a pass-through calibration matrix in the DNG metadata and providing a LUT derived from our secret-sauce calibration that the user can use to bring the camera from "native" space to some useful intermediary to begin grading

despite being open source they appear to have some legal reason to be throwing out litigation to any companies which are using the format

polyrhythm: we could consider writing pure raw12 sequences and putting all the "color science" and processing we come up with into the software that translates those raw12 files to DNGs (it doesn't need to hapen in realtime while recording) - could be a frameserver

like, in the CinemaDNG flow, Resolve reads the raw data, reads the matrix in the metadata, and then maps the data from XYZ -> whatever working space the user has selected using those calibration matrices. I think if you tried to "pre-calibrate" the data directly in the DNG, that might have unforeseen consequences

I can be more confident that I'm thinking about this the right way with a little more reading o the DNG spec and looking up how calibration is done

I know that other raw image vendors are preprocessing their raws for the same reasons

https://ieeexplore.ieee.org/document/7289691

maybe a linear transform into a color space that is a nonlinear transform of XYZ can improve things aswell

polyrhythm: DNG does in fact support nonlinear transform if I understand it correctly

for this, see p. 82 of the DNG v1.4 spec

polyrhythm: also, at this stage I would not hesitate to just try to squeeze out the best possible thing and then see if we can map that to CinemaDNG and with what sacrifices

anuejn: regarding the HSV mapping, Resolve doesn't support that metadata field :(

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=165986

i look forward to receiving the test chart shots so we can compare results and see how best to proceed