Slog 2 and ISO

[Noah Yuan-Vogel]

It seems to me SLog2's strength is in the log, in that it's tonal response is very 'linear' through its range. I'd think if anything that means you don't have to follow sony's arbitrary middle grey guideline or worry so much about being able to recover or change exposure. You don't have to worry about skintones falling in some knee area that can never really be brought back. I've had quite good luck exposing Slog2 anywhere from normal to +1 and +2 stops to minimize the appearance of noise in low light shooting or just minimize noise in general by 'exposing to the right'.

I've even found in low lit scenes if you overexpose around 1.5 stops images look quite nice if you skip the LUT and just restore normal saturation (and perhaps set black level back to taste).

[Alister Chapman]

No, SLog is not linear, it's log. every stop you go up the exposure range you are recording with effectively 50% less data than the previous. In effect the entire gamma curve is a "knee".

Remember each successive stop of exposure contains double the brightness range of the previous, yet log records each stop with roughly the same amount of data. So the further up the range you go the less data you have about the total tonal range of each stop. In practice this equates to a reduction in contrast in each stop as you go brighter. If you do over expose SLog it doesn't necessarily look over exposed in the traditional sense. It will grade down to a natural level and may look OK in isolation. But compare the over exposed shot with a correctly exposed shot and suddenly you see a whole load of extra subtle details in skin tones and textures that are not there in the over exposed shot.

Normally log is designed to be used with 10 bit recording where you have an excess of data per stop compared to most viewing systems which are typically 8 bit. So when you shift your 10 bit values around by a couple of stops, the loss is not noticed by the viewer on his 8 bit screen. However with 8 bit recording you don't have that luxury, so getting the exposure correct becomes much more critical.

One issue is that if you do over expose SLog it doesn't necessarily look over exposed. It will grade down to a natural level and may look OK in isolation. But compare the over exposed shot with a correctly exposed shot and suddenly you see a whole load of extra subtle details in skin tones and textures that are not there in the over exposed shot.

[Noah Yuan-Vogel]

I kind of figured I'd get called out for that use of the word, but I thought some might understand my meaning without lengthy explanations.

When you are saying 'linear' is it in reference to linear light, non-log, or in it's more common, non-video use which describes an essentially 1 to 1 relationship (can be graphed as a straight line)? The irony is that 'linear' in its most common usage for video refers to something that least fits the word's actual definition. 'Linear' in its meaning as non-log gamma corrected video, often with knee and toe applied, is less 'linear' than linear-light and than log. Linear light data is directly proportional to the light reaching the sensor from a photographed scene measured in lux/fc. Log data is directly proportional to the light reaching the sensor from a photographed scene measured in stops (much closer to the was humans perceive variations in light). Linear in its meaning as non-log, usually gamma corrected and curve applied data, is not proportional to the light reaching the sensor at all due to the arbitrary curve applied.

Frankly I think the 'lack' of range given to certain parts of the sensor's DR by Slog or other image data storage schemes is often overblown. Remember 20 'shades per stop' is actually per stop per color channel, so in reality it's more like 20^3=8000 per stop for RGB/YUV pixels, plus actual image data is very well dithered by noise. Obviously if you push things hard in the right conditions you can break it, but I find a lot of people complain about 8bit or log based only on numbers they don't fully understand, not experience. 8bit is 256 'shades' per channel but for 3 channel RGB/YUV imagery, that's 16.7million 'shades' per pixel.

Yes you will always use your 8bits best if you get exactly the image you want in camera, but pushing around 'linear' isnt that different from pushing around log except that in practice you are more likely to have to push around log. For imagery thay I know will be graded anyway, i'd rather capture in log with the risk of losing a few steps/shades in the midtones than the risk 'linear' has of making slightly overexposed areas unrecoverable.

The whole curve is not a knee, that completely disregards the definition and purpose of a knee. You could easily say the same then about any recording scheme that is not linear light data, but linear light is very wasteful for image information that is intended to be perceived by humans.

[Alister Chapman]

I think your missing the point that gamma is a two way process. Gamma correction on the recording and then gamma correction on the display so that the response of the total end to end process should be linear.

If you add a log encoded gamma curve on the input side, then convert this back to linear via your output/display transformation, your final output will have a lot, lot less tonal information as you go up your exposure range. This is OK and does to some degree mimick the way we see the world, provided you are not shifting your levels around too much, but start shifting levels down and every stop down you go you halve the tonal range in each stop and this results in plasticky low contrast faces and skin tones.

Rec-709 is is pretty close to linear. The toe is completely linear, form 0 to 0.1 the response is 100% linear. Then it follows a simple 0.45x power function. The shape of the response curve follows a straight line that diverges from linear at a constant angle, it isn't actually a curve at all. The knee is optional, but the key thing is that the mid-range where faces and all your important data sits has a uniform straight line response relative to the light from the scene. In practice this means that an over exposed face (not in the knee), relative to the true linear light from the scene, contains only a little less tonal data as a correctly exposed one, as a result there is very little change in contrast or tonal range when you shift your levels around. This makes grading easy and the results highly predictable.
Log is not like this at all. Compared to the light coming from the scene log diverges further and further from the true linear light by ever greater amounts as you go up the curve, the response is a curve that diverges logarithmically from the true image. So a face exposed a stop brighter is recorded with 50% less tonal information (relative to linear light). As a result there is a narrow sweet spot within the curve where after converting to 709 or most other display gammas, skin tones etc will have more tonal range than a typical viewing system can show. This is all based on a 10 bit to 8 bit conversion. With 10 bit the consequences of this are minimal as viewing is normally done with 8 bit (or less) devices, so we don't notice the data reduction. But shift your levels down and every stop down you go you halve your tonal range and this hurts the final image, you are not just loosing a few shades but 50% of your shades. You can do this with 10 bit over a couple of stops, but 8 bit just doesn't have enough reserves to do this, move Slog 2 down one stop and you have about the same tonal range in the mid range than you would have with straight 709, move it down 2 stops and you have half of the tonal range of 709. This is why log always used to be reserved for 10 bit recording or higher.

As a test we shot with an F5 and FS700 both in SLog2 at a recent workshop and deliberately over exposed both by a couple of stops, then graded the material comparing the +2 stops footage with our correctly exposed reference. With the 10 bit recordings from the F5 it was hard to see any difference, the +2 stop clips was perhaps marginally less contrasty. But the FS700 footage was a different matter. You could clearly see the degradation of the FS700 material. The skin tones were noticeably less pleasing lacking the subtle textures that make faces look real. 709 looked better.

The higher end of SLog2 from around CV 800 uses less data per stop than a typical 709 knee.

[Sami Sanpakkila]

Originally Posted by Alister Chapman

As a test we shot with an F5 and FS700 both in SLog2 at a recent workshop and deliberately over exposed both by a couple of stops, then graded the material comparing the +2 stops footage with our correctly exposed reference. With the 10 bit recordings from the F5 it was hard to see any difference, the +2 stop clips was perhaps marginally less contrasty. But the FS700 footage was a different matter. You could clearly see the degradation of the FS700 material. The skin tones were noticeably less pleasing lacking the subtle textures that make faces look real. 709 looked better.

With the Odyssey 7Q using 4K2HD my understanding is that it's 10bit?

[Alister Chapman]

With the Odyssey in 4K raw to ProRes HQ mode, yes it is 10 bit and this makes a big difference giving you that extra data that does allow you to move your exposure around a little.

[Colin Elves]

FWIW, working on the FS700 with the O7Q I tend to expose the Raw at ISO 800 (using a light meter) and bring it down in Resolve using the camera raw exposure slider (or, if I'm moving too fast to use the light meter I'll apply the LUT and deliberately over expose using false colour) because I find the final image too noisy otherwise. (don't ask me where I got ISO 800 from - it just works for me. ISO 640 would probably make more sense as that's the 'base' ISO Sony decided to give the camera when working in Cine 4, which makes me assume they feel that's a level that they felt the camera could perform well at: retaining good headroom, without adding too much noise to the mids).

Shooting slog2 4KtoHD I tend to just apply the 709 LUT and expose using the false colour - plopping skin tone highlights around 70 IRE/what looks good, pulling off the LUT to check if any of the image is clipping. Partly because the 4KtoHD debayer and downscale seems to hide a lot of the noise (which may be due to the way CD do it - according to Mitch Gross it's a bit more involved than a simple debayer and scale) and partly because if I'm working in 4KtoHD, odds are it's a fast moving project so I don't have time to use my light meter. If I do have time I'd expose again at 800 and bring it down.

Colin Elves
DP, London
Colin Elves - Director of Photography and Steadicam Operator, London, UK

[Sami Sanpakkila]

Hi Colin

I'm shooting 4K2HD with the Odyssey7Q due to file size of raw being too much for us to handle. We're shooting an indie feature and this was the best way to do it.

I'm noticing that the images are somewhat noisy as you mention. Although it's a nice film like noise and I don't mind it that much I'm interested in how would you bring the exposure down in post with the 4K2HD files?

I'm using Premiere Pro CC myself and am not that familiar with Resolve, but next year when we color grade I'm planning to use Resolve.

I just bought a Sekonic 608 cine light meter as well and will use that on set from now on.

[Colin Elves]

Originally Posted by Sami Sanpakkila

Hi Colin

I'm shooting 4K2HD with the Odyssey7Q due to file size of raw being too much for us to handle. We're shooting an indie feature and this was the best way to do it.

I'm noticing that the images are somewhat noisy as you mention. Although it's a nice film like noise and I don't mind it that much I'm interested in how would you bring the exposure down in post with the 4K2HD files?

I'm using Premiere Pro CC myself and am not that familiar with Resolve, but next year when we color grade I'm planning to use Resolve.

I just bought a Sekonic 608 cine light meter as well and will use that on set from now on.

I just use the gain controls (i.e. on the lift, gamma, gain control) - I'm not sure what that would be in Premiere pro as I don't use it.

You have to be a bit more careful with slog2 than the Raw though, as I think it can mess with skin tones if you overexpose too much (you don't want to push them too far up into the top of the curve to where the gradient increases a lot). Probably best to rate it at ISO 1000 or 1250 if recording slog2. Or shoot a couple of skin tone tests and see how it fares.

[Dave Allen]

What exposure percentages would be advised when shooting in FS700 RAW on the 7Q? I normally apply the 7Q LUT when shooting which adds a bit of blacks to the monitor and I just adjust based upon monitor appearance, but I need to start using the exposure tool son the Odyssey.

[Colin Elves]

Generally, if I'm shooting raw, I expose using a light meter set at ISO 800.

I tend to avoid using the LUT as exposing it 'correctly' using the LUT means a very noisy final product (sure, lots of DR, but the cost is noise).

If I do use the LUT, because I'm shooting docco style, or the director wants to see what it will look like, I set it 'correct' (i.e. Zebras at 70, or false colour showing yellow on skin tone highlights) then open two stops from there for the recording, (or one if I'm struggling due to high contrast or low light).

Working without the LUT I'll expose the slog2 image as if it were Rec709 (i.e. Zebras at 70/Yellow on false colour) - again, this has the effect of over exposing by a couple of stops (the 'correct' setting should be skin tones at 52 - green on the false colour).