What They Say
Dolby conducted a survey of content professionals to find out what they thought of SDR content that was automatically derived from Dolby Vision mastered content. The background and results of the survey were presented in a YouTube video.
Dolby said that the result of the survey was very positive, but with only limited facilities to manipulate the look as there are just five simple controls (Lift/Gamma/Gain/Saturation Gain and Chroma Weight) and a single CMv2.9 tone curve. Sometimes there were challenges in mapping P3 or Rec 2020 content to Rec. 709 and ‘outlier’ images could cause problems.
So, in version 4.0, Dolby addressed the isses with nine primary controls and 12 secondary controls for the process. The firm then tried matching the shots to hand-graded content. Dual monitors could not be used, but a Sony OLED monitor could be used. The group then tried to check if industry professionals could tell the difference between the automatic and hand-graded content. 8 clips were used and subjects were asked to judge which was the graded content when the content was shown side by side. There was an option of ‘not sure’. 560 participants were used, typically in groups of 5-8.
The result was a ‘bell curve’ with 145 respondents getting 50% correct – essentially the same as a random choice. Only 3 of the 560 respondents correctly identified all the clips. On the other hand 12 people got none right. 8.6% of scores were declared as ‘not sure’.
Some film makers that started a project with an SDR version and then created a Dolby Vision version later and used an SDR trim pass just to create metadata found that the SDR version was improved by this process. They saw better shadow and highlight detail as well as getting better colour vibrancy and saturation.
What We Think
If you are involved in grading, there are some interesting anecdotes about automatic SDR producing better SDR versions automatically from an HDR master, rather than manually creating SDR.
Of course, if Dolby had not been pretty confident of this kind of result, it wouldn’t have gone to such lengths. Still, it’s good to see some data around this kind of topic. (BR)
