Just to respond to a couple of points Geoff Boyle mentioned earlier:
'3 different log formats because Clog is the original format that's kinda obsolete now but is there for compatibility reasons, Clog2 because it's the right log for the camera and gives the widest dynamic range, Clog3 because there are so many people out there who can't deal with a real log workflow but think they should be shooting log, it's a hybrid gamma/log.
..
As I've said in the past, there is REAL log, which almost nobody uses, and sorta log which everyone else uses.'
A few personal observations as I have worked my way through manufacturer documentation, marketing bumpf and white papers:
C-Log is a pure log curve. It uses a single equation across the range which is log. However, the parameters chosen don't get close to the 'loggy' even distribution per stop until above mid gray. The consequence is that there is as much distribution as possible in midtowns and highlights to counteract the severe limitation of 8-bit, and the poor distribution in the shadows has the effect of keeping them nearer to what the end result (i.e. for a Rec709 monitor) is likely to be, with the added advantage that the darker, 8-bit shadows will also tend to keep you less aware of shadow noise. In other words, tuned to 8-bit, relatively unscary looking to a non-colourist and not a million miles away from a usable picture. Ideal as an introduction to log for the masses and did so well for C300.
C-Log works in extended range, so taps out at about 5 1/3 stops headroom with nowhere left to go. The C300mkII introduced 10-bit recording, and talk of 15 stops dynamic range. As camera people, headroom is where it is at with DR, so Canon had to fundamentally ditch C-Log in order to record more DR, and certainly to allow decent manipulation of shadow stops as colourists would expect from the likes of Arri LogC and (by then) Sony S-Log3.
C-Log2 is mathematically, again, basically pure log. The one caveat there is that if anyone remembers logarithms from school, log(0) is -infinity, so to avoid that possibility and presumably also to 'even' the spread of the darkest stops within what is still a limited range of values at 10-bit, the log curve transitions to a linear portion somewhere several stops down in the shadows. The rest is still pure log, the transition is clearly defined and the whole thing is easily reversible. To my mind it is still pure log.
It should be pointed out that Rec709 and sRGB are 'piecewise' gamma curves which combine a power curve with a linear region in the shadows.
C-Log2 also has much better spread in the shadows, so rather more for a colourist to play with, but milkier with more chance of seeing noise to the casual fiddler. Response-wise it is much more like LogC and S-Log3.
Now many non-colourists have been shooting with the C300mkI in 8-bit C-Log, so C-Log2 will have been a shocker to them. In response, Canon have come up with C-Log3. With the maths again (I've got this from looking at Canon's ACES IDT, haven't seen a white paper yet) C-Log3 has the linear bit in the shadows transitioning to a pure log region which maps as closely as possible to C-Log up to mid gray and then that transitions to a different set of log parameters to fit in more headroom, i.e. the slope of the straight line that log tends towards on a C-Log3 vs Stop chart is shallower than on a C-Log vs Stop chart, but below 18% gray everything is very similar. I believe that because the curve has two distinct sets of log parameters this is why Canon call it a 'hybrid log'. Mathematically it would also be legit to call it a log curve.
Sony S-Log1 takes a similar approach to C-Log with relatively unscary shadows, a pure log definition across the range and about 5 1/2 stops headroom. S-Log2 is mathematically essentially S-Log1 underexposed by half a stop allowing for an extra half stop of headroom. The only other difference is the addition of a linear region in the shadows presumably for the reasons I already gave.
Panasonic V-Log follows the S-Log3 / C-Log2 / LogC path of pure log with fairly even distribution coupled with a linear region in the deep shadows.
S-Log3 breaks that approach and introduces a much more even spread in the shadows. As with S-Log2 and C-Log2 it is a combination of a pure log curve with a linear region in the deep shadows. All still very well defined.
Arri LogC is complicated from the back end, but making things really simple for the user. C-Log baked in ISO or gain changes, Sony's S-Logs all record the entire dynamic range identically regardless of (CineEI) ISO, with metadata supposedly telling post software how to cope later (Hmmmm) but Arri LogC has a basic definition (log curve with linear deep shadows again) but the parameters change depending upon the ISO selected. When you see a response chart for LogC, it is generally given at 'native' ISO of 800.
The reason for the changing parameters is so that when you load pictures into post software, shadows through to mid gray will always look the same regardless of ISO (as opposed to 'over' or 'under' exposed Sony and more like C-Log), but when you shift the ISO LogC effectively shallows its highlight slope so that nothing is clipped and no DR lost (as is the case with Canon's C-Log). It actually works the other way as the ISO goes down, maximising the spread of data and making sure that white clip on the recording stays close to where it is at native (another thing which I know has caused consternation on Sony cameras).
To me, all that seems very clever on Arri's part, and since the end goal is generally just the easiest route to the nicest pictures it makes for a very comfortable experience both as camera person, editor and (presumably) colourist. The only hassle comes for whoever has to knock up the code for something like ACES where everything ends up at a common base. Fortunately that was knocked up by Arri themselves, so phew!
RED is just introducing Log3G10 which I have only recently been filled in on, which is an interesting development shifting away from the pure Cineon of REDLogFilm, so since I'm on a roll... ;-) This is defined by a pure log curve tuned to set 18% gray at 33% with ten stops of headroom by 100%. The log(0) issue is avoided by incorporating a 'mirror' before values get there (actually at linear zero i.e. camera black).
ACES has their own log, ACEScc (and the ACESproxy versions designed for monitoring). The only way in which this is more 'loggy' than other curves is that the parameters are set such that the even distribution per stop line is approached far earlier in the shadows than any of the recording logs. This values each stop equally, the effective 'goal' of log, but the downside is when you move away from floating point to limited bit depth integers (i.e. the 'proxy' versions) you have to define a hard cutoff for black. ACESproxy seems to do that around seven stops below mid gray, so not really a great loss for now. ACEScc is defined by two sets of log parameters and a hard cutoff deep, deep, deep in the shadows. Arguably in the sense of even distribution per stop across the (usable) board this could be considered 'real' log, but personally I think that that would be an unhelpful definition.
Anyway, sorry for the long, long post, but 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!
Ben
Ben Turley
Lighting Cameraman / DoP