On Potential HDR

It should be noted that the approach to "colour management" given in Filmic deviates from some of the "accepted" ideas around this area currently. It is up to the reader to interrogate the veracity of the claim.

As per the history behind Filmic, I have long believed that the nature of the final image includes the aesthetics of the unique attenuation of chroma present in the final film print. In most contemporary visions of the surface of "Colour Management", this attenuation is either an accident as in the case of per-channel image formation approaches, or engineered in more sophisticated systems.

The critical point is that the attenuation, and the resulting shifting in what an audience member sees, is not baked into the open domain tristimulus signal generated from either a render or even an expanded digital camera. This vital aspect of the image is seen as something... well you'd have to ask folks who think it's not part of the image to explain. I find the whole subject utterly disheartening.

On a technical level, I would consider the image formed via these mechanics. In the Filmic case, the actual mechanic of the formation is the per channel mechanic, bringing in the ever important crosstalk between the channels, and shifting the resulting image in front of us. To this end, if one were to want to bring the image over to a medium with a different set of qualia, it would be important to maintain the "intention" of this chroma interaction / attenuation to the medium.

As an example, I'd expect an image that was converted via some black and white processing to maintain the attenuating / cross talk characteristics. That is, it would be for example, a greyscale version of the post-formed image, not a greyscale representation of the open domain tristimulus in the render model.

Similarly, it seems logical to me to take the post-formed image tristimulus relative to the observed medium to a relative, closed domain uniform tristimulus range. This has the effect of maintaining the cross talk characteristics, and then reprojects the uniform / linear post formed image tristimulus into a medium such as EDR.

For the other outputs included in this repository, this is exactly the guiding principle behind the transformations. The post-formed image is linearized via a simplistic power function, and then re-encoded for the alternative displays. This asserts one side of the "colour managed" approach because the tristimulus values in the source post formed image are maintained across the mediums supported. More ideally, an appearance based definition of colour could be harnessed to assert that the appearance of the post formed image is maintained across the supported outputs.

Given the tremendous complexity of the latter appearance aspects, and the fact that it is my belief no such model even remotely works as of this writing, the former stimulus encoding constancy approach was taken. This leaves the appearance constancy subject to the authorial interactions.