Comment by unicornporn | Hacker News Reader

Comment by unicornporn | Hacker News Reader

Parent Full thread

Since darktable introduced the scene referred workflow I would argue it's an extremely steep learning curve for anyone coming from display referred software (Lightroom, Capture One etc).

Scene referred is very much a cine-thing and I know of no other still image software that uses it. dt v3.6 made scene referred default.

Aurélien has been one of the driving forces behind the switch to scene referred and his hour long explanations of the new modules can be found on his YouTube channel[1]. Boris has a YouTube channel[2] which is quite the opposite. You watch him edit in silence and you will pick up the craft.

darktable is made up of modules. The Filmic RGB module takes a central part of the editing workflow in scene referred. This is where to go if you need help getting started: https://discuss.pixls.us/t/lets-learn-filmic-rgb-your-one-st...

RawTherapee has a more traditional, display referred, approach. If you want local adjustments (which you will want), you need to use nightlies until 5.9 is released.

[1] https://www.youtube.com/channel/UCmsSn3fujI81EKEr4NLxrcg

[2] https://www.youtube.com/channel/UCMbDlOwmmQnkRmcb2_5WERg

For anyone caring just enough to wonder what scene referred and display referred are talking about…

scene referred does its filtering and mapping work on the pixel values in a linear space, presumably linear to "photons observed" but I could be wrong there.

display referred does its filtering and mapping on pixel values which have already been through a nonlinear function to make them appropriate for display.

The scene referred people say they don't tend make silly skin tones and avoid "rat piss" sunsets as well as reducing artifacts like halos around blurs. (I presume that means the blurs are something like Gaussian blurs performing linear operations on the nonlinear data.)

This is all I know.

> scene referred does its filtering and mapping work on the pixel values in a linear space, presumably linear to "photons observed" but I could be wrong there.

You are correct. If you see Y light in real life, and that corresponds to a pixel value of 1000, 2Y light should correspond to 2000.

> display referred does its filtering and mapping on pixel values which have already been through a nonlinear function to make them appropriate for display.

The nonlinear function is usually 0 brightness to 100% brightness of the media (for instance, pure white on a printed piece of paper or pure white on a display). Scene referred photos technically have no bound (there is clipping, but you can just expose less and increase the exposure in editing to effectively get the same thing)

Of note is that you have to change a scene referred color space into display referred color space anyways...

I have to admit I still don’t get Filmic. I have a basic grasp of it but I am much more comfortable with shadow and highlight adjustments. I have watched pretty much all tutorials I could find but I am struggling big time. I have no idea how other people feel about it but I am starting to wonder if the scene refereed workflow is more an intellectual exercise and less a practical tool.

Filmic mimics the nonlinear dynamic range response on film medium to push the contrast and dramatic look present on the shot.

It also highlights how old film struggles when compared to much more capable CMOS we have today.

I use filmic rather frequently to get the look / emotion response for a shot, however it's not always suitable. Sometimes the lighting is not favorable for it, sometimes the shot I have at hand is not well suited for that because of other reasons.

Shadow, highlights, exposure, etc. are corrective tools while Filmic tends to be an artistic one.

My other favorite tool is contrast equalizer.

None of what you said makes any sense. For people like me, here is an explanation of scene referred workflow

"The scene-referred workflow places an emphasis on performing image processing in the linear scene-referred part of the pixelpipe. This helps to reduce artifacts and color shifts that can result from processing non-linear pixel values and, by decoupling the image processing from the characteristics of a specific display, it makes it easier to adapt your work in the future to new display media, such as high dynamic range displays."

In case anyone else is looking for a tl;dr on "scene-referred" this link has a very simple diagram. https://www.filmriot.com/blog/scene-referred-vs-display-refe...

gist is whether you do your color grading in the camera's original color space (scene referred) or after transforming to display color space (display referred)

> you do your color grading in the camera's original color space

you do your color grading in linear color space (or something that roughly maps to linear color space) is what that actually means. Linear meaning, the exposure of a pixel is supposed to be proportional to the data number, and the exposure is theoretically infinite (since you can have anywhere between 0 and infinite light). For display color space (for photography this is more in line with "print media color space") there is a maximum brightness and minimum brightness, and you have all the numbers to map brightnesses between the two.

The math is a lot cleaner with linear color spaces.

Reply on news.ycombinator.com