Reflections on HDR

29 July 2017 18:17 | Kevin Shaw (Administrator)

by Lou Levinson
After graciously being invited to be a CSI fellow, it seems that life circumstances forced me to have to pay attention to other things over the past year. Please accept my apologies, and allow me to stir up some trouble, if I may.

What I’d like to discuss is a subject that’s gained buzzword status in our endeavors on par with“Digital” and “4k* ”. Yep, that’s right, HDR. I’m sure you’ve all been inundated with this in one way or another. “Can you do this in 2020**? Your competition can”. 

So, what do we mean when we say HDR? I have to say that I have yet to hear a simple and coherent definition, but everyone seems to know it when they see it.  Do we mean high peak brightness? High average/diffuse white? Are we including wide color gamut? High frame rate? High spacial resolution? As far as I can tell, there are now five or six competing HDR “systems”.  We’re well on the way to the 19 different standards of the ATSC. I have asked folk at NAB, at HPA, at ASC Technology Council (new name) meetings. I’m allergic to SMPTE meetings, and no one wants to send me overseas to IBC or to ITU meetings, but I do have work associates for that. So, even though Ideal with this all the time, I’m confused.

If we take a step back to SDR, we see a fair amount of energy still being put into defining it and the environs one would master it in. Really? Let’s get real here. The only place you’ll find 100nit pictures*** is in a reference grading room. Let’s define it well, and be aware that it will be applicable for those archival projects when we want to see what was done in HD****, hopefully with Creative input. Everywhere else you’re likely to view them is going to be significantly higher.

Your phone is probably 400nits peak or more. For those that can do this, try looking at something you’ve graded at 100nits and 400nits side by side. Or even as single stimulus; in two different rooms one after the other. Tell me how your 100nit grade holds up as consumed by the Best Buy crowd. SDR does point to other benchmarks we need to exceed, such as 709 primaries.

So, you ask, what would I call HDR? Well, let’s take a clue from some Dolby research that says that even untrained observers, in single stimulus, can tell when the peak brightness has gone up by 2 stops. If we chose 100nits as SDR then entry level for HDR becomes 400nits. If we chose a more current 400nits as SDR, then 1600nits becomes an HDR entry point. This is not as arbitrary as it seems, as will be seen in a bit. So I would start to define HDR as 1000nits plus peak brightness. I’m taking some away here for realism, i.e., what we can come close to building as a reference display right this minute. Add wider, P3 primary, color gamut as a minimum. Use whatever transfer function, gamma, pq, hlg, the ballistic curve of a 16” shell fired from the USS Iowa, that you feel works. Package it however you want (you will anyway). Just use enough bits to be user transparent. And make sure there’s clear metadata that says what you’ve done. I would make a plea to keep it simple, but that barn has already burned down.

One thing of note I can pass on from my more recent imaging voyages is that there is a place, between 100nits and 400nits where dragons reside. What I mean by this is that on either side of that gulf, one makes different creative decisions about one’s visual storytelling. Period. I’ve yet to see any automathemagical, transform driven solution that crosses that place well, if at all.

Keep that in mind when those cost conscious clients want one master to rule them all. It’s a long walk to Mt Doom. Staying above 400 makes those automathemagical transforms work better, if not perfectly.

What I might have done if I were king (oh, thank god) is set the following HDR target:

  • 1000 nits peak bright
  • P3 primaries
  • D65 white
  • 1886 or 2084 transfer curve (sensible metadata)
  • 4096 minimum
  • 12 bits minimum

I would like to assign some homework if I may,  to be dropped on the CSI website in some manner to be determined by wiser heads than mine. I would like anyone who cares to, to submit your definition of HDR. The catch? You have a 147 character limit. [N.B. There is a thread in the CSI Forum to discuss this further - Kevin]

Next time, I might be talked into telling you why, almost certainly, none of your reference monitors exhibit ideal, reference behaviour.


Half Moon Bay, Ca
July 23, 2017

* UHDTV is not 4k, it’s 3840

** Can you actually capture 2020? Can you display it? I thought not. Nice package, though.

*** I actually use 102.7 nits as it equals 30 fl

**** I’ve seen folk put masters graded on a crt up on a dlp projector in 709 and say that’s the reference. In the words of Wolfgang Pauli “That’s not even wrong!” Crt’s were SMPTE C not 709, 30fl not 14fl, and blacks were significantly different, as well. I fear for our ability to know what came before now.

Colorist Society International is a registered 501(C)(6) non profit professional association © 2021

Powered by Wild Apricot Membership Software