Research | Colour ManagementCan I Make D-Log Linear?
July 15th, 2018
PEAK LUMA, D-CINELIKE, D-LOG, Y-WAVEFORM, DISPLAY REFERRED, SCENE REFERRED
In an effort to ‘Linearize’ the 8bit REC709 video delivered by the DJI OSMO – as is the case with all 8bit REC709/sRGB video files – it must be mapped to a 1 – 0 scale. To this end, the exposure data you capture from the camera when trying to create a linear conversion LUT (Look Up Table) must represent the very lowest, Black or 0 point of the scale and conversely; the very highest, pure white, 1 point on the scale.
From some simple experiments and subsequent observations of the d-Log footage from the OSMO, it seems that it is impossibly to both produce a pure black (0) or pure white (1) value, mitigating its ability to be converted to a linear 1 – 0 scale. This it seems is there to recover highlights and shadows details as observed in earlier posts. This is perfectly acceptable for Colour Grading and standard film workflows, but it does mean that we are sacrificing data to scale 11 stops of dynamic range into an 8 bit image. To this end, no matter how high the exposure settings are taken the OSMO in D-Log mode will not go higher than 234 in a 0 (0)- 255 (1) scale.
What follows are my observations and findings in this regard alongside some tests with other picture profiles in the camera to find the best balance between full range data capture and pleasing, organic dynamic range for colour grading.
- Validity of Research 80%
- Complexity of Content 100%
- BENEFIT OF ACTIVITY 100%
Above is a screengrab of the Y-Waveform scope from Davinci Resolve 15, that displays the inherent Peak Luminance (Y) values of the D-Cinelike DJI OSMO footage prior to scaling to REC709 display transform. This is in stark contrast to the D-Log equivalent, which cannot reach 1023 data levels. You can see this comparative analysis in the video evidence below.
D-Log EXP Limits
When researching a methodology for creating a LUT to convert the footage from the OSMO display refered, REC709 colour space to a Linear, Scene refered colour space, it was clear that the data captured in camera must represent the full range from 0 – 1/0-255 in 8bit video.
From my previous research around the quality of ‘gradable’ dynamic range in the D-Log colour gamut in the OSMO app, this seemed like the logical option to preserve 11 stops of dynamic range across the full spectrum. Right? Wrong!
When practicing the manual camera exposure settings on the camera prior to my data capture for LUT creation something became clear: no matter how bright the lighting or exposure settings, the D-Log gamma curve would flatten out highlights and thusly not exceed 234 in a 0 – 255 8bit luminance scale. From what is gained in subtle graduation, data is lost at the top and bottom extremities to make sure that this is contained within the limitations of 8bit imagery.
This can of course be remedied in post production colour grading processes, but this is not the aim of this exercise. For accurate VFX integration across all composting programmes linear data is best and we need to reverse engineer the linear data at the heart of the camera. This we cannot do with this camera; with this setting.
You can see this issue in the figurative images below and in the compairtive video I have made, where data has been taken from both a Y-Waveform inside Davinci Resolve and from the histogram and RGB indexes in DJV Imaging Viewer, both very accurate and industry standard monitoring software.
D-Log clips whites at 234 IRE (954 Approx.)
D-Log unfortunately limits peak white/brightness to approximately 234 IRE (954 on a 0 – 1023 scale) even though the in camera monitor applies a Zebra overlay for 100% brightness. This is not only misleading, but not usable for generating full-range exposure data for Display to Scene Referred conversion.
D-Log crushes blacks at 15 IRE
D-Log also crushes blacks at approximately 15 IRE (30 on a 0 – 1023 scale) which also mitigates the generation of 0, black data for data gathering later; even if the camera was ‘capped’ with the lens cap.
d-log sacrifices data to scale 11 stops into 8bit
To this end, it seems that data is sacrificed at both the peak and low ends of the signal when using the D-Log setting to preserve shadow and highlight detail. This is great for grading in post, but not adequate for full-range exposure data capture for 1D LUT generation. Hence, D-Cinelike will be used for both data capture and VFX plate acquisition.
Here are some recent posts that I have written as part of the Digital Media Practice 3 project – ‘FUSE’
September 7th, 2018 Chiefly, once the lighting is set up correctly using the Linear EXR data as an environment map, and by setting the colour management to 'Filmic' - the colour management system written and developed by Troy Sobotka - much of the heavy lifting is...
August 30th, 2018 Continuing on from my modelling process, I moved on to the texture painting process in Substance Painter. I not only chose this programme for its procedural, non-destructive, scalable workflow, but also so that I could try a separate Normal/Height...
August 24th, 2018 Following my successful shoot in The Peak District, I began the modelling process. Chiefly, I used my research from previous projects in The Arcane Welder project to combine both mesh-based and Boolean-based deformation with procedural kit bashing...