Canon C-Log on the C300 compared to S-Log.

First let me say that as yet I have not used C-Log in anger, only seen it at a couple of hands on demo events and in downloaded clips.

From what I’ve seen C-Log and S-Log are two quite different things. S-Log on the F3 is a true Log curve where each stop of exposure is recorded using roughly the same amount of data and the available dynamic range is about 13.5 stops. It is inevitable that when you use a true log curve like this and play it back on an uncorrected Rec-709 (standard HD gamma) monitor that it will look very flat and very washed out. This is a result of the extreme gamma miss-match across the entire recording range. If you had a monitor that could display 13.5 stops (most only manage 7) and the monitor had a built in Log curve then the pictures would look normal.

What has too be considered is that S-Log is designed to be used with 10 bit recording where each stop gets roughly 70 data bits ( this roughly means 70 shades of grey for each stop).

Now lets consider the Canon C300. It has no 10 bit out, it’s only 8 bit. Assuming Canon’s sensor can handle 13.5 stops then using 8 bit would result in only 17 bits per stop and this really is not sufficient, especially for critical areas of the image like faces and skin tones. A standard gamma, without knee, like Rec-709 will typically have a 7 stop range, this is a deliberate design decision as this yields around 34 bits per stop. As we know already if you try to do a hard grade on 8 bit material you can run in to issues with banding, posterisation and stair stepping, so reducing the bits per stop still further (for example by cramming 13.5 stops into 8 bits) is not really desirable as while it can improve dynamic range, it will introduce a whole host of other issues.

Now for some years camera sensors have been able to exceed 7 stops of dynamic range. To get around the gamma limitation of 7 stops, most good quality cameras use something called the knee. The knee takes the top 15 to 20% of the recording range to record as much as 4 to 5 stops of highlights. So in the first 0 to 80% range you have 6 stops, plus another 4 to 5 stops in the last 20%, so the overall dynamic range of the camera will be 10 to 11 stops.

How can this work and still look natural? Well our own visual system is tuned to concentrate on the mid range, faces, foliage etc and to a large degree highlights are ignored. So recording in this way, compressing the highlights mimics they way we see the world, so doesn’t actually look terribly un-natural. OK, OK, I can hear you all screaming… yes it is un-natural, it looks like video! It looks like video because the knee is either on or off, the image is either compressed very heavily or not at all, there is no middle ground. It’s also hard to grade as mid tones and highlights have different amounts of squashing which can lead to some strange results.

So the knee is a step forward. It does work quite well for many applications as it preserves those 34 bits of data for the all important mid tones and as a result the pictures look normal, yet gives a reasonable amount of over exposure performance. Next came things like cine gammas and film style gammas.

These often share a very similar gamma curve to standard gammas for the first 60-70% of the recording range, so faces, skin, flora and fauna still have plenty of data allocated to them. Above 70% the image becomes compressed, but instead of the sudden onset of compression as with a knee, the compression starts very gently and gradually increases more and more until by the time you get close to 100% the compression is very strong indeed. This tends to look a lot more natural than gamma + knee, yet can still cope with a good over exposure range, but depending on the scene it can start to look a little flat as your overall captured range is biased towards highlights, so your captured image contains more bright range than low range so will possibly (but not always) look very slightly washed out. In my opinion, if shooting with cinegammas or similar you should really be grading your material for the best results.

Anyway, back to the Canon C300. From what I can tell, C-Log is an extension of the cinegamma type of gamma curve. It appears to have more in common with cinegammas than true S-log. It looks like the compression starts at around 60% and that there is a little more gain at the bottom of the curve to lift shadows a little. This earlier start to the compression will allow for a greater dynamic range but will mean fewer bits of data for skin tones etc. The raised lower end gain means you can afford to underexpose more if you need to. As the curve is not a full log curve it will look a lot more agreeable than S-Log on an uncorrected monitor, especially as the crucial mid tone area is largely unaffected by strong compression and thus a large gamma miss-match.

For the C300 this curve makes complete sense. It looks like a good match for the cameras 8 bit recording giving a decent dynamic range improvement, largely through highlight compression (spread over more recording range than a conventional knee or cinegamma), keeping mid tones reasonably intact and a little bit of shadow lift. Keeping the mid range fairly “normal” is a wise move that will still give good grading latitude without posterisation issues on mid range natural textures.

5 thoughts on “Canon C-Log on the C300 compared to S-Log.”

  1. I have a question: Even if the C300 supports log grading, doesn’t it being only 8 bit limit the flexibility of how much one can stretch the colors in post?

    Would this be much different from Technicolor Cinestyle on a DSLR?

    1. The C300’s Log looks to be tailored to work with 8 bit recording. That’s why it’s much less flat looking than S-Log. As it is closer to a conventional gamma, the amount you need to manipulate it in post production will be reduced, so you are less likely to run into issues. Equally as important as the bit depth is the codec. Mpeg 2 50Mbps 422 is far from ideal for heavy grading, but it is quite a lot better than anything that a Canon DSLR can record internally. Taking the C300’s output to something like a NanoFlash at 100 Mbps would dramatically improve how much you can grade the image, even though it’s still only 8 bit. One thing that will need very careful investigation is what happens when you take the C300 8 bit out and record it with a 10 bit recorder like the Samurai or Ki-Pro mini. Will the simple 8 to 10 bit conversion done by the recorder be beneficial or detrimental compared to recording 8 bit?
      8 Bit recordings will contain less actual luma steps and chroma shades than 10 bit. You can copy from 8 bit to 10 bit and use software to calculate values for the in-between data, which can be moderately accurate. If you then grade this new 10 bit then you will in most cases get results close to what a true 10 bit recording would provide. However you will need to be very careful with your workflow to get the true benefits of this. For example a simple transcode of the 8 bit to a 10 bit codec like ProRes or any other re-encoding prior to your high quality upscale to 10 bit may introduce banding that is then near impossible to eliminate.

        1. I wasn’t thinking straight when I answered your first comment. Of course when you record 8 bit with a 10 bit recorder all that happens is the two least significant bits become zero and thus your image is untouched and effectively remains as 8 bit data only contained in a 10 bit word.

          Too much alcohol this holiday season!!

Leave a Reply

Your email address will not be published. Required fields are marked *