AXIOM looking for color science developer

Started by Sebastian, November 13, 2021, 12:39:56 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Sebastian

We have recently made very exciting progress with uncompressed 12 bit 4k/UHD raw recording on the AXIOM Beta (entirely open hardware/free software high end camera community developed  project: https://www.apertus.org/axiom in case you have not heard of it before).
Its based on the experimental raw recording mode: https://www.apertus.org/axiom-beta-uhd-raw-mode-explained-article-may-2016 that we developed together with a1ex at the time - and it uses HDMI as "transport platform". The original approach turned out to be rather cumbersome to use though as recording devices altered the image in various ways (chroma subsampling, blurring, sharpening, compression) and a big part of the things we tried to do was reversing those steps to acquire the original raw data again - which was limited to 8 bits per channel at the time.

Now with the new approach we use a USB3 HDMI capture device which does not alter the image in any way and allows us to regain 12 bit raw directly as it arrives from the image sensor in the camera so all the above headaches are a thing of the past.
We still use a HDMI signal - pack our own custom pixel data into the video stream though. With some tuning we are even able to get non standard HDMI resolutions transferred (4096x2160, 4096x3072 - full image sensor resolution etc. at 24, 25 and 30fps). A1ex created a whole suite of conversion tools (https://github.com/apertus-open-source-cinema/misc-tools-utilities) and measurement routines to benchmark and analyse some aspects of the images and applied processing (darkframe subtraction, gainframe, non linear response compensation, clipframe and dynamic row noise compensation) but is currently not able to continue his work.

Here is a quick sample of recently captured footage with/without applied darkframe:


We are hoping to find someone interested in rebooting the color science work together with us and continuing building on top of the foundation that A1ex created.

2blackbar


ilia3101

Is it colour science that you guys are looking for, or is it more like signal processing?

Anyway I'm interested :)

Sebastian

Quote from: 2blackbar on November 13, 2021, 05:32:35 PM
what format it is? dng frames? + pcm wav?

Yes, DNG sequence is the primary target format. We go through raw12 (which is basically just sequentially written 12 bit raw data from the image sensor) as intermediate step: https://wiki.apertus.org/index.php/RAW12

Full processing block diagram - anything after raw12/DNG is optional of course or depends on the planned processing pipeline.
https://paste.pics/c1ba7e506cb4e801d8c4b06eb148af66


Quote from: ilia3101 on November 13, 2021, 06:03:25 PM
Is it colour science that you guys are looking for, or is it more like signal processing?

Anyway I'm interested :)
Great, we are hoping to work on both things actually :)

2blackbar

GReat, i cant find dng samples with human face tho, to convert from raw16 i need to compile converter for win
Details like this( not having dng samples for people to tinker with) can and will harm/limit interest , id like to check how footage behaves under tonemapping algos in mlvapp cause not all dngs give same results, it would probably also need cam matrix.

Sebastian

Quote from: 2blackbar on November 14, 2021, 04:10:10 PM
GReat, i cant find dng samples with human face tho, to convert from raw16 i need to compile converter for win
Details like this( not having dng samples for people to tinker with) can and will harm/limit interest , id like to check how footage behaves under tonemapping algos in mlvapp cause not all dngs give same results, it would probably also need cam matrix.
There are some DNG samples to download from the end of the above article:
https://www.apertus.org/axiom-beta-uhd-raw-mode-explained-article-may-2016

2blackbar

I got them, cant open them in mlvapp with dng to mlv converter tho...