> my ears pricked up when I saw the comment that almost nobody uses real log, and as you might imagine I wondered what that meant!
Me too. And I think it effectively doesn't matter whether anyone uses
"real log" or not. "Log" as we know it simply means that the storage
space available is being used to store dynamic range in roughly
perceptually equal steps. Whether the formula used is "true log"
simply doesn't matter, as I don't know that there's an advantage to
that over optimizing some of these curves to weight highlights or
shadows depending on where the camera's strengths lie.
Regardless of how the code values are distributed, if you look at any
of these "log" curves with a waveform monitor and a DSC Labs Xyla
chart you'll see a stair step pattern of stops with more spacing--and
more equal spacing--above middle gray than below, because most sensors
trail off in a non-linear fashion near black anyway and there are more
stops captured below middle gray than above, so something has to give.
Arri is roughly +7/-7 at ISO 800, but Sony cameras tend towards +6/-8
at base ISO. And, considering that middle gray is generally at 33% or
40%, that means the top 6 or 7 stops get 60-70% of the storage space
available, while the rest fit in that bottom 33-40%. That will
naturally result in some compression between stops.
As for the linearity of the range near black, that's fairly common for
the reason Ben mentions: the closer you get to black with a log curve
the closer you get to a point where math problems occur, so a linear
stretch is added. That happens in Rec 709 as well: if I recall
Poynton's description, if you follow the Rec 709 curve all the way
down it'll never reach 0% and just progress indefinitely, getting
closer and closer to 0% but never reaching it. As that's not desirable
a linear range is added to the bottom of the curve. A similar thing
happens to log curves.
You can see all this in Ben's LUTCalc app (on the web for free at
http://cameramanben.github.io/LUTCalc/ or on the Apple App store for
cheap). Click on "Preview" and then "Waveform" and you'll see a
display that looks a lot like a Xyla chart. Run through the various
log curves and you'll see exactly what's going on. There's a graph
that will also show a comparison between recorded and output gamma
curves so you can compare and contrast.
They certainly aren't all the same, but it's clear that the
intent--encoding data for storage in steps that are roughly
perceptually equal for the sake of both efficiency and
flexibility--is the same. A "log" curve is fairly easy to recognize.
As for whether any of these are "true log" or not... it simply doesn't
matter, and further confuses the issue by saying they are not. The
intent is the same: storing information in roughly perceptually equal
steps in a more efficient manner than can be done in linear light
alone. The term used for this kind of encoding has been "log" because
that's roughly how the formulas work, even if they aren't log for the
absolute full range or if they incorporate different log formulas at
different points in the curve.
Math is not my strong suit. I do understand that manufacturers have
tricks up their sleeves when designing log curves, and they are all
slightly different for a reason. But to say they aren't "real log"
confuses the matter, as I doubt there are many people here who look at
the formulas behind any of these curves and stage protests outside
camera manufacturers' offices chanting, "Hey hey, ho ho, 'log'
terminology has got to go!" The intent is the same even if the methods
vary, and ultimately it all comes out in the grade.
>That's the point, to me a "real" log is exactly that, even distribution across the stop range.
That doesn't seem to exist, and seems a bit pedantic given the common
usage of the term "log" to describe the intent of the curves, even if
it doesn't define the function absolutely perfectly. If we're going to
be this precise in our terminology then we should throw out "log"
altogether and call this something else, although that hardly seems
helpful.
"Non-linear perceptual encoding meant for efficient storage but not
direct viewing" might work. From now on let's call these
NLPEMFESBNDV--or Nulpemfes Bindev--curves, as that's a more accurate
description of how these curves work and has the added benefit of
rolling gently off the tongue.
Certainly the next time I'm viewing a manufacturer's camera
presentation and they mention "log" I'm going to throw a handful of
toothpicks at the presenter and yell loudly, "Log is for wood! Stop
lying to us! Perceptually equal steps... bullshit! Don't hide the
linear portion of the curve near black from me, you bastard!"
Mathematically-based lynchings can't be far off. A 50' roll of BNC,
two high rollers and an length of speed rail should do it. I think
I'll reserve a space at NAB 2017 between Sony, Arri and Canon and
construct a mass gallows. Anyone know where I can rent a popcorn
machine in Vegas?
-----
Art Adams | Director of Photography
San Francisco Bay Area | Silicon Valley
www.artadamsdp.com