Research | Colour Management

Exploring DJI D-Log Dynamic Range
July 10th, 2018


Following the previous post(s) regarding colour management, the ACES pipeline and the Sobotka’s ‘Filmic’ colour management in Blender, information around live action acquisition formats, colour space and dynamic range is very important.

This is particularly relevant in the context of this study to not only optimize explorations of colour management but also test and ratify the impact of the ‘Filmic’ colour management inside Blender on Live Action plates, especially following the observations made in the initial tests within this post

To this end, the hope of this post is to explore, test and clarify the Dyamic Range of one of the proposed aquistion devices for this project: DJI OSMO. In partiucalr it will compare the Dyamic Range of both the internal REC709 scaling and the D-Log colour space avaiabke in the DJI settings. This post will also explore the differences – if any – of scaling when both formtas are imported to Blender, both in its Standard default colour manament and in the Blender ‘Filmic’ piepline.

  • Validity of Research 80%
  • Complexity of Content 80%
Above is a screengrab of the Y-Waveform scope from Davinci Resolve 14, that displays the inherent Luminance (Y) values of the D-Log DJI OSMO footage prior to scaling to REC709 display transform.

Secondary Research


Using the official DGI D-Log grading documentation as a source it is clear that the standard OSMO, firstly, does not have a RAW video recording mode as an option. This would have been the ideal and typical way in which to gather all of the exposure data from the sensor.

From this document it is also clear that DGI alludes to the OSMO sensor being capable of capturing 11 stops of dynamic range. This said, it also clarified that in the absence of a RAW video recording option, the 8 bit video recording codec compresses this 11 stops from the sensor to about 8 stops. This is typical of 8bit video recording.

This said, the D-Log LUT contained within the advanced options of the camera, applies a colour transform inside camera to better expand the dynamic range being compressed and scaled into the 8bit compression. To this end, even though it does not represent the true, RAW data from the sensor, it better approximates it. See the Exposure/Greyscale Chip charts within the DGI image below.

Core findings

OSMO chip CAPABLE OF 11 stops

In D-Log it seems that OSMO is capable of 11 tops of dynamic range

No raw video recording option

This sensor data cannot be 1:1 transferred to a video file due to no option for RAW video recording option in the OSMO.

8 bit video compress to 8 stops (approx.)

The only available video options inside the OSMO are 8bit 4:2:0 video compression inside either a (.mov) or (.mp4) container. Due to the limitations of standard, ‘normal’ 8 bit compression, only 8 stops of the 11 stops on the sensor can be retained.

D-log colour transform scales 11 stops to 8 bit (approx.)

The D-Log colour transform better scales the 11 stops on dynamic range off the sensor prior to compression, so as to better preserve the wider dynamic range.

Primary Research

DJI OSMO CAMERA – DLOG versus REC709 Internal

In order to test the relative ‘gradability’ and perceived dynamic range of the different colour spaces, I simply set up the OSMO on my dining room table looking to the garden outside. The theory was that the huge exposure difference between inside and out would test the relative dynamic range of each.

Lighting conditions were also very balanced and sustained so as not to skew the results. It was clear the in camera REC709 display referred colour space was fairly manipulatable and had quite a naturalistic gamma curve and saturation. This said, the D-Log did scale more EV stops of dynamic range into the 8 bit display referred colour space of the camera.

Highlight details and shadow details were also notably better when the D-Log was colour managed and scaled back to REC709 display refereed colour space. Please watch the video for my primary results.

Core findings

D-Log is very gradeable

From both the images recorded and from the Y-Waveforms it is clear that d-Log preserves much more shadow and highlight detail similar to a logarithmic curve.

REC709 (display referred) is still gradeable, but crushes highlight details

The internal, Display referred 8 bit footage directly from the camera in the Standard profile is still quite manipulable but does tend to clip in the highlights/specular.

The Curious Engine

My Professional Portfolio
Go There Now

Recent Posts

Here are some recent posts that I have written as part of the Digital Media Practice 3 project – ‘FUSE’