I have had several people ask me why there is a inflection in the curve Karl posted. So I thought I should talk a little bit about the issues of an HDR input to linear light output graph.
TV, movies, and photos, are all about what humans see. That is, linear light. HDR and SDR are purely engineering compromises to get the image from in front of the camera into your home with as much visual fidelity as possible. The Gamma in-between makes no difference as long as it provides the desired signal-to-noise ratio and it is isomorphic (one-to-one) as much as practical from camera input to TV/projector out. That is, what the camera sees is what you get (ignoring directors intent to change it for artistic reasons).
All computer created imagery does calculations in linear light. This is necessary to produce the correct images. HDR Intensity Mapping is done in linear light for the same reason. By converting to Linear light we can make adjustments to the transfer function and then convert the data to output in any Gamma we choose.
To evaluate what the HDR Intensity Mapping is doing to the image you need to look at it as linear light input to linear light output to make sense of it visually. Humans do not see in “HDR Gamma.” HDR plotted as the input warps the graph and can easily give the wrong idea of the transfer function. Of course, HDR in to linear light out graphs can be useful. It’s just that people may be misinterpreting what the graph means visually. Here is an example (attached).
The graph on the left is HDR input Gamma through HDR Intensity Mapping to linear light out. The curve on the right is linear light in to linear light out through HDR Intensity Mapping. Both use the same HDR IM parameters, but as you can see look totally different. The graph on the right gives an accurate visual of the transfer function and shows that the inflection point in the left curve is an artifact of having a non-linear input space.
In the right curve you can see how HDR Intensity Mapping is working for the chosen parameter values. The lower range has a steep but linear slope which is the nit-for-nit range. That is, in this range the image on the screen matches the source data as precisely as calibration allows. This slope is the HDR maximum of 10000 / Display_Max_Light. The graph reaches full output at 1.0 which in this case is above the maximum input level possible for the source’s reported Maximum Mastering Level. The upper region is linear, and there is a curve in the middle. Where the curve is vertically is set by Transition, and how smooth or sharp the curve is set by Shape. If we were to increase Clip above zero it would move the output intercept with 1.0 to the left.
I have considered changing the upper portion of the transfer function from linear to a curve. I am not convinced it is the right thing to do but I have not ruled it out. We will continue to listen to feedback and to work on improving HDR Intensity Mapping. However, I think it is already doing an excellent job – especially if you use it in conjunction with calibration – and doubling the range of the Transition parameter makes it even better.