May 19, 2016
[In response to your observations about DPs you've spoken with who are surprised to hear that "there is no uninterpreted RAW.” And to your questions about what interpretation they are viewing when they display so-called "RAW" on a monitor.]
Before answering the essence of your questions, I’m going to do a pedantic preamble regarding the words being used. On the one hand it is indeed unabashed pedantry and not an actual answer to your questions. On the other hand, it’s worth clearing up the words because the confusion in the words contributes to confusion in the concepts.
This has to do with the word “RAW.” To my mind, the capitalized word (which represents something technically specific rather than just a general English-language adjective) means nothing other than “is still Bayer pattern and has not yet been converted to RGB for viewing.” Converting between Bayer and RGB has nothing whatsoever to do with what DPs believe they may be looking for when they view “RAW” (like, maybe evaluating exposures or checking highlight detail or something). Converting Bayer to RGB is purely geometry, not color/luma handling. It’s just scaling the three channels individually to different screen sizes.
For example, when you dabayer Alexa ArriRaw (nominally 2880x2160), you take the green channel which has 1440x2160 Bayer pixels and scale it to 2880x2160. And you take each the red and blue channels, which have 1440x1080 Bayer pixels each and scale them separately to 2880x2160, then you take the three newly scaled 2880x2160 channels together and you have an RGB image. This is just geometry, not color or exposure.
So, “RAW” in this sense has nothing to do with checking exposure:
- There is no such thing as a Bayer monitor so any image being viewed has necessarily been modified and is no longer in the very state that is indicated by the word “RAW”. It's meaningless to try to view “RAW” files in the strict sense— it literally can’t be done. On the one hand, modifying away from RAW in this way is trivial to what DP’s are usually intending to do when they view “RAW," but on the other hand the fact that it is trivial also shows why “RAW” is trivial and not of any essence.
- In the case of ArriRaw (which is at least analogously representative of other formats too), the actual data inside the files while they’re still RAW is Logarithmic, so when you debayer but stay in LogC, LogC still uses those unaltered logarithmic code values — the same one that were used inside of the RAW file. So, in that sense only, LogC is indeed “raw” (in lower case, because it is an adjective, not a file type).
- By #1 and #2 above, we can see it would be less misleading and more accurate to say you're "viewing log" rather than "viewing raw."
Okay, that’s my whole pedantic rigmarole and you may wonder why I’ve gone through all that to distinguish capital “RAW” from lower case “raw.” Well, now I’ll actually answer your question and you’ll see why:
We now see that the kind of “raw” that can actually be viewed has logarithmic density characteristics. But the log image is being viewed on a monitor that has a gamma-decode, not a logarithmic-decode. So, literally all you’re doing is showing it incorrectly: you're showing a mismatch of intent by taking a log-encoded image and decoding it as though it were a gamma-encoded image. Technically speaking, that's just a distortion, not some magic way to get closer to the essence of your image.
To make an analogy to photochemical acquisition: it's physically possible to put original camera negative in a projector and project it on a screen, but why would you do that -- it's not meant to be viewed that way. You would be able to understand less, not more, about your exposure by doing so than you would by making a print of the neg and putting that in the projector as intended.
Viewing a mismatched signal by feeding a log-encoded image into a monitor built for gamma encoded images tells you less not more about your exposures.
For example, when viewing log on a gamma monitor, the difference in overall apparent density as you close down the camera iris from properly exposed to two stops underexposed is difficult to discern perceptually because the log image on the gamma-decode monitor is so flat and washed out. So it is very hard to judge exposure. It’s also extremely difficult to judge if dark things are just a tiny bit dark or so dark that they’r almost gone, because they’re all so flat and close together.
To my mind, viewing with proper intent has some value (even if limited) whereas viewing with mismatched intent has almost no value and can actually be misleading and detrimental.
Also, any good LUT will not crush or clip anything that the camera actually got — it will just show the various intensities with their intended luminance on the screen. So, the whole idea of using "raw" to “see what’s in the highlights” does not distinguish it from viewing with a proper LUT and could actually be more misleading than helpful. In the example, a good LUT that showed proper intent might tell you that a bright highlight is still there but nearly gone whereas viewing with a mismatch would only tell you that it's still there but not how close it is to being gone, possibly instilling false confidence.
Moving on, you also asked if viewing “Raw” is “standard display prep result.” The answer is “no.” In the case of Alexa, the “standard display prep” would not be to view a mismatch by feeding LogC into a monitor that does gamma decode, but would be Arri’s standard LUT applied to LogC. Viewing LogC with Arri's LUT (in order to correctly send a gamma-encoded signal to a gamma-decoding monitor) is closer to my proposal of viewing an image in the intended way instead of through a mismatch, but is still problematic.
The reason it is problematic is that their standard LUT is very flat. It’s so flat that, to my taste, it’s actually still quasi-log.
If the Arri LUT really does represent your intent (to have a super flat, milky image), then you’re totally good to go by using their LUT. The problem is that many people that use it don’t notice how milky it is while on set and then in post are trying to get a more traditional look that actually has solid blacks in it. At that point, they have a problem because if they treated the on-set monitor as an evaluative guide, they may have (for example) exposed the dim side of someone’s face way dimmer than they intended.
If that’s the case, they now have a problem in post. That dim side of the face may be only a quarter of stop brighter than the darkest inky black thing in the shot. So now, either everything in the shot has to be milky for the face not to be that darker than intended or you have to violently rip the image apart by making one incoming value black while making an incrementally brighter value very bright.
In other words: you’ve lost all the subtly graduating tonal shades of shadow. In post you can now milk everything, you can crush everything, or you can rip it apart, but you can’t have a long gentle toe into black because of lighting ratios that were used on set — you just don’t have very many tones between the not-too-dim part of the face and deepest black. In this hypothetical example, the false accuracy of the LUT has boxed you into the corner by failing to be the evaluative tool you believed it to be.
So to recap it all in three short steps:
- True “RAW” is nonsense for evaluating exposures on set, because it refers only to geometry, not to how various intensities of light taken in by the lens are recorded as various code-values in the image file.
- Viewing log images (i.e. lower-case "raw" images) is not useful and probably detrimental because it literally achieves nothing other than viewing something incorrectly: a log-encoded system is decoded as though it were a gamma-encoded system. That's just a mismatch.
- Viewing through a transformation that is accurate for the show’s intent can indeed have value, but using an out-of-the-box LUT instead of something rigorously set up for your intent can be dangerous because it can instill false confidence while being deceptive for evaluating exposures.