Understanding Log

Video recordings with log curves, are not raw data.

Log is video image, not RAW format.
Log recording formats were originally mainly found in professional high-end camera systems, but this function is now also available internally or via external docking recorders in more and more smaller, mid-range prosumer devices. In connection with the recording with log curves, questions arise again and again regarding the recording of RAW and the uncompressed recording of video signals. This article is intended to help you navigate safely through these terms from now on.
Video recordings with log curves, in the following referred to as "log", are not raw data, i. e. no RAW. Since Log is video and RAW is not a video image (RAW is only a data stream), Log and RAW are by no means the same. However, Log and RAW have similar application goals. In both cases, two different strategies are used to get as much information as possible from the sensor and store it in order to produce a detailed and balanced image afterwards.
Each camera manufacturer has developed its own log curves. ARRI calls it LogC, Sony calls it SLog, Canon calls it Canon Log. That means, each of them are specially created curves which are tuned to the respective sensors installed in the cameras The aim is to transfer the dynamic range coming from the sensor into the available image signal in the best possible way. Everyone is inclined to get the best image quality with the smallest possible amount of data. This can be achieved with codecs or by using a reduced bit depth compared to the raw sensor signal. This means that sensors internally quantify the light captured by them with 12 or more bits. However, in order to sensibly reduce data volumes in the output formats, 8- or 10-bit video signals are usually used.
With 8 bits, 256 gradations (28 = 256) from absolute black to super white are available for the video signal. 10 bits would provide 1024 possible gradations, which definitely is much more, but the described context applies here as well.
For the sake of simplification, a hypothetical dynamic range of 8 f-stops is assumed. The following table shows how an exposure range divided into 8 equal parts is distributed across the available gradations of an 8-bit signal if a linear relationship is used as a basis:
If one looks at the distribution of the gradations in relation to the individual "f-stops", it becomes clear that a total of 128 gradations are used from f-stop 7 to f-stop 8 alone - half of the available bit depth! And for the entire darker part of the signal, in which all information from the shadows of an image lies, only 16 gradations are left. In this case, no noteworthy detail reproduction in the color correction could be expected.
If a log curve is used for recording, the brightness values in the signal are redistributed in a way that the dark and medium exposure values are shifted to the upper third of the signal, where more gradations (code values) are available.
This relationship becomes even clearer when one realises that the natural perception of brightness is not subject to a linear relationship, either. One feels a difference in brightness between 100% and 200% just as much as a difference in brightness between 200% and 400%, 400% and 800%, and so on.
The subjective perception of brightness must be taken into account. Therefore, the brightness values are distributed in the signal in such a way that the difference between the one and the next value always corresponds to a subjectively equal increase in brightness. And that is exactly what a log curve is designed for.
One effect of this approach is that you get a comparatively large dynamic range coming from the sensor into a relatively "narrow" signal without having too much subjective loss of quality. Another effect is that the resulting video image appears extremely low-contrast and colorless. Log recordings are therefore extremely unsuitable for evaluating the image from an aesthetic point of view. Despite the log curve, checking the correct exposure is also still necessary A logarithmically recorded image must never be used for evaluation without measuring instruments (waveform/vectorscope/histogram) or on an uncorrected monitor. In order to make a log image attractive and assessable for the eye, either a color correction or the application of a LUT (Look Up Table) to the material is always necessary.  The outputs of certain cameras or preview monitors can apply LUTs directly and live to log material - these LUTs usually convert to Rec709, the standard HD color space and contrast.