The recording and notes from meeting #183 are now available.
Below are just some of my thoughts as to discussions in the March 5th, 2025 meeting.
I appreciate and would like to emphasize Doug’s comments as to having an 18% grey look the same in SDR and HDR. Maybe not necessarily at 10 nits or whatever standard would be used, but identical (or very close.)
I always consider HDR values above 100 nits to be the brighter colors that become clipped, crushed or compressed in SDR. For the maximum it seems comfortable having 300-350 for dark room and 1000 otherwise. But I would try for a reasonably close match of SDR/HDR for less than 48 nits dark room and 100 (or 108) nits otherwise. Also understanding that there will likely be more discrimination of dark values in HDR (denoting more detail,) I would still expect the SDR and HDR to be close otherwise in mid to lower values. I also expect skin tone to look fairly close between SDR and HDR (accounting for skin ‘wetness’ and highlights differing.)
And the above thoughts are for a starting place (something standard) because things can always be shaped as desired. But would be valuable to have a place to have SDR and HDR ‘match’ somewhat.
This issue was discussed for more than a year during the group’s extensive work on nailing down a system tone scale for ACES 2.0. It was discussed that some users would need a way to pin mid-gray at the same point for SDR and HDR, but was ultimately decided that the default behavior should be for mid-gray to elevate slightly as peak luminance increased, based on direct user feedback and data from where real productions tend to place things.
The SSTS in v1 HDR transforms allowed for this but had its own other issues that made it difficult to adjust consistently (at least without further work to define its behavior beyond the few presets provided). The v2 tone scale is much simpler to adapt as it just takes a single input parameter - peak luminance.
However, I do hear the possible need for users to have a way to have the SDR and HDR to look more similar in the mid-tones. We discussed some possible ways to achieve this, by either adding an additional ODT or providing an LMT. It was also noted the importance of having some documentation (with graphs) explaining the intended behavior and how to utilize the accessible parameters to adjust the behavior to use cases beyond those provided in the current set of default transforms.
Right now there are other priorities, but I will be working on making a proposal for this and documenting some of the trade-offs of different approaches once I have time to do so.
thanks, and thanks for the reminder about the elevation of mid with peak lum
I just reviewed the meeting recording where there was discussion around matching SDR middle grey and HDR middle grey. I understand the rational of having middle grey change based on max luminance and it sounds like the change is not too drastic, even when comparing 100 nit max luminance to 10,000 nit max luminance. I agree that detailing the expected behaviour with some example graphs would be helpful. The comments from @alexfry about why this is desired behaviour were good to hear and would be good to have written down.
During the meeting, I believe @alexfry asked if there any users who were actively finding this behaviour to be a problem. I actually made a post on a similar topic of using externally controlled reference luminance to dynamically match the apparent brightness of host operating system and its SDR content. I felt it was worth linking this thread here because it touches on some of the real-world cases where it is important to match HDR apparent mid-grey brightness to SDR apparent mid-grey brightness on an HDR display.
Common use cases would be games that are played in windowed mode on a desktop computer with an HDR monitor or mobile games where the user is switching back and forth between an HDR game and operating system/SDR apps, all of which should use the same reference luminance reported by the operating system. Using the same reference luminance should mean that middle grey ends up at approximately the same apparent brightness. (I don’t meant to say that it should be the exactly the same as per reasons discussed in the meeting.)
The ACES DRT is a common DRT which needs to address an average of many use cases.
Dynamic tone mapping was never the scope of this Working Group and also I don’t think you want to keep an arbitrary point on the tone scale fixed as you change the peak luminance :-).
I don’t think it makes sense to match a graphics white luminance in a pictorial rendering, if you switch between different scopes your eyes readapt anyway so equal luminance will not mean same appearance.
4 Likes