Part 2 of LS300 review is now up

[Steve Mullen]

UNDERSTANDING LOG AND EXPOSURE LEVELS (ALSO OTHER GAMMAS). PLEASE READ AND UNDERSTAND.
ALISTERCHAPMAN

Please, please read this and try to understand how shooting with a high range gamma curve such as a cinegamma or hypergamma or log recording works. The principles are not well understood by many, even highly experienced DP’s and DIT’s get this so horribly wrong.

Why do so many get it all wrong? Because we are brought up used to looking at a monitor or viewfinder and seeing a picture that looks correct.

Why doesn’t the picture look right when we shoot log (or other extended range gamma)? It’s simply because the monitor does not have the right gamma curve (unless you have a log monitor), so there is a miss-match between the camera and monitor.

So what does this mean? DO NOT USE THE MONITOR TO JUDGE YOUR EXPOSURE unless you have a well calibrated Look Up Table between the camera and monitor!

...


A typical LCD monitor or TV set has a very limited contrast range and can only display about a 6 or 7 stop dynamic range. OLED’s are a bit better.

Thanks to the Rec-709 gamma curve in the monitor, when we send data bit 940 to the monitor we see what appears to be white. Send bit 64 and we see black, send bit 440 (approx) and we see a shade of grey that appears to be halfway between black and white, also known as middle grey.

Middle grey is approx 2.5 stops darker than white (as in a piece of white paper or similar) and if we go around 2.5 stops darker than middle grey we will see something very close to black. So we can see that using bits 64 to 940 we will get around a 5 stop dynamic range on the monitor with a bit of extra range from bit 940 to 1019, so overall there’s our typical 6 stop monitor range.

Now, what happens then if we have a camera with a much greater dynamic range than 6 stops? Well, the monitor can never show the cameras bigger range accurately as it can only ever show 6 stops, if we feed say 14 stops into the monitor the brightness range on the monitor will still only be 6 stops. So now the contrast of the picture is reduced as we are squeezing the cameras large contrast range into the monitors much smaller contrast range.

Now let's consider the camera. If I shoot a white card, I record it using bit 940, if I shoot a grey card I record it using bit 440, that way the white card looks white and the grey card looks grey on my monitor which uses those same levels for those same shades, then I have a little bit of extra space above 940 for a little extra dynamic range. Remember, near black to white is approx 5 stops of dynamic range.

But what if I want to extend my range beyond 5 stops? If white is bit 940 and my top limit is bit 1019, I really don’t have a lot of data space to record a load of extra range, so I have to do something else.

What do the camera manufacturers do to record a bigger dynamic range? They shift the data values used down. Taking SLog2 as an example, instead of using bit value 940 to represent white, they now use bit 600 (approx) and for middle grey, instead of bit 440 we now use bit 347. This now gives us a large amount of spare data from bit 600 to 1019 to record a greatly extended range beyond our original 5 stops.

...

Now the next common mistake is the thought that: “OK, my picture looks dark, so when I take it in to post production and raise the levels, it’s going to get noisy”. Well, this is to small degree true but it is not nearly as bad as many assume. The reason it’s not as bad as many assume is that you must remember that YOU WERE CORRECTLY EXPOSED. You are not trying to lift an under-exposed image. Remember what I said at the beginning: “The noise in a digital camera comes almost entirely from the sensor”.

So, with the same camera, if we expose any given gamma correctly then as the amount of light falling on the sensor is the same, the ratio of sensor noise to signal coming from the sensor does not change. So taking a face as an example, exposed correctly (ie. with middle grey at the correct level for the gamma curve in use) the amount of noise on that face will remain constant across all the different gamma curves. Do note however that some cameras may have different ISO ratings for different gammas and this might have a small impact on noise levels (but that’s the subject for a different article).

...

But what about a more aggressive gamma curve like SLog2 or any other log gamma. This is going to need some big level changes, surely this is going to get noisy. Again, no, not if you handle it correctly. You really should be using a dedicated grading tool for any log material as this will apply corrections that are designed for log and this will minimise any added noise. But the other thing to consider is that this is where you should be using a LUT or Look Up Table on your output to convert you data values from Log values to Rec-709 values.

By placing a LUT on the output of your project you shift your data levels from one range to another. Your grading is done to the original material in it’s original range so that you can retain that full range and then your LUT is used at the end of your grade (on the last node) to then convert your data values from log values to 709 values. When you do this you are simply moving your data values. So if the original input value for a part of the image is is bit 347, SLog2 middle grey for example. On your output you just use bit value 440 (709 middle grey) instead. Your just transposing data from one range to another and this does not add noise in the same way as adding gain does.

Now, looking at Log and the way it works. You should note that in order to squeeze 14 stops of dynamic range into our normal recording codec you use a lot of compression in the brighter stops. Remember, every time you add a stop of exposure, to record everything in that additional stop you should be recording the new stop with twice as much data as the previous. But that’s impossible with conventional recording, the amount of data required is simply too big. So log records every stop using roughly the same amount of data. This means that the brighter stops are very highly compressed, so it’s very important not to overexpose log to get the best results.

[Guy Smith]

Thank you for explaining how log works and the relationship between log and rec 709, Very Helpful!