This document has been prepared independently of Sony. It is based on my own findings having used the camera and tested various exposure levels and methods. LUT’s to accompany this article can be found here.
If you find this useful please consider buying me a coffee or a beer. I’m not paid to write these articles.
One of the really nice features of the Sony A7s is the ability to use different gamma curves and in particular the Sony S-Log2 gamma curve.
What are gamma curves?
All conventional cameras use gamma curves. The gamma curve is there to make the images captured easier to manage by making the file size smaller than it would be without a gamma curve. When TV was first developed the gamma curve in the camera made the signal small enough to be broadcast by a transmitter and then the gamma curve in the TV set (which is the inverse of the one in the camera) expanded the signal back to a normal viewing range. The current standard for broadcast TV is called “Recommendation BT-709”, often shortened to Rec-709. This gamma curve is based on standards developed over 60 years ago and camera technology has advanced a lot since then! Even so, almost every TV and monitor made today is made to the Rec-709 standard or something very similar. Many modern cameras can capture a brightness range, also known as dynamic range, that far exceed the Rec-709 standard.
The limitations of standard gammas.
As gamma effects the dark to light range of the image, it also effects the contrast of the image. Normal television gamma has a limited dynamic range (about 6 to 7 stops) and as a result also has a limited contrast range.
Normally the gamma curve used in the camera is designed to match the gamma curve used by the TV or monitor. This way the contrast range of the camera and the contrast range of the display will be matched. So the contrast on the TV screen will match the contrast of the scene being filmed and the picture will look “normal”. However the limited dynamic range may mean that very bright or very dark objects cannot be accurately reproduced as these may exceed the gammas dynamic range.
The over exposure typical of a restricted range gamma such as Rec-709 Â is commonly seen as bright clouds in the sky becoming white over exposed blobs or bright areas on faces becoming areas of flat white. Objects in shade or shadow areas of the scene are simply too dark to be seen. But between the overexposed areas and any under exposure the contrast looks natural and true to life.
Log Gamma.
Log gamma, such as Sony’s S-Log2, allows the camera to capture a much greater brightness range or dynamic range than is possible when shooting with conventional television gamma. Dynamic range is the range from light to dark that the camera can capture or the range that the monitor or TV can display within one image. It is the range from the deepest blacks to the brightest whites that can be captured or shown at the same time.
There are some things that need to be considered before you get too excited about the possibility of capturing this much greater dynamic range. The primary one being that if the camera is set to S-log2 and the TV or monitor is a normal Rec-709 TV (as most are) then there is no way the TV can correctly display the image being captured, the TV just doesn’t have the range to show everything that the camera with it’s high range log gamma can capture accurately.
Fixed Recording Range For Both Standard and Log Gamma.
The signal range and signal levels used to record a video signal are normally described in percent. Where black is 0% and the brightest thing that can be recorded is normally recorded at 100% to 109%. Most modern video cameras actually record the brightest objects at 109%. The important thing to remember though is that the recording range is fixed. Even when you change gamma curve the camera is still constrained by the zero to 109% recording range. The recording range does not change whether you are recording Rec-709 or S-log2. So log gamma’s like S-Log2 must squeeze a much bigger signal range into the same recording range as used by conventional Rec-709 recordings.
Recording S-Log2.
In order to record using S-log2 with the A7s you need to use a picture profile. The picture profiles give you several recording gamma options. For S-log2 you should use Picture Profile 7 which is already set up for S-log2 and S-Gamut by default (for information on gamuts see this article). In addition you should ALWAYS use the cameras native ISO which is 3200 ISO and it is normally preferable to use a preset white balance. Using any other ISO with S-log2 will not allow you to get the full benefit of the full 14 stops of dynamic range that S-log2 can deliver.
Grey Cards and White Cards.
Before I go further let me introduce you to grey and white cards in case you have not come across them before. Don’t panic you don’t have to own one, although I would recommend getting a grey card such as the Lastolite EzyBalance if you don’t have one. But it is useful to understand what they are.
The 90% White Card.
The 90% white card is a card or chart that reflects 90% of the light falling on it. This will be a card that looks very similar in brightness to a piece of ordinary white paper, it should be pure white, some printer papers are bleached or coloured very slightly blue to make them appear “brilliant white” (as you will see later in many cases it is possible to use an ordinary piece of white paper in place of a 90% white card for exposure).
The Grey Card.
The 18% grey card, also often called “middle grey” card, is a card that reflects 18% of the light falling on it. Obviously it will appear much darker than the white card. Visually to us humans an 18% grey card appears to be half way between white and black, hence it’s other name, “middle grey”.
Middle grey is important because the average brightness level of most typical scenes tends to be around the middle grey brightness value. Another key thing about middle grey is that because it falls in the middle of our exposure range it makes it a very handy reference level when measuring exposure as it is less likely to be effected by highlight compression than a 90% white card.
Exposing White and Middle Grey.
Coming back to Rec-709 and conventional TV’s and monitors. If we want a piece of white paper to look bright and white on a TV we would record it and then show it at somewhere around 85% to 95% of the screens full brightness range. This doesn’t leave much room for things brighter than a white piece of paper! Things like clouds in the sky, a shiny car, a bright window or a direct light source such as a lamp or other light. Â In order to make it possible for S-log2 to record a much greater dynamic range the recording level for white and mid tones is shifted down. Instead of recording white at 85%-95%, when using S-log2 it is recommended by Sony that white is recorded at just 59%. Middle grey moves down too, instead of being recorded at 41%-42% (the normal level for Rec-709) it’s recorded at just 32%. By recording everything lower this means that there is a lot of extra space above white to record all those bright highlights in any scene that would be impossible to record with conventional gammas.
As S-Log2 normally shifts a lot of the recording levels downwards, if we show a scene shot with S-Log2 that has been exposed correctly on a conventional Rec-709 TV or monitor it will look dark due to the lower recording levels. In addition it will look flat with very low contrast as we are now squeezing a much bigger dynamic range into the limited Rec-709 display range.
This on screen reduction in contrast and the darker levels are actually perfectly normal when shooting using log gamma, this is how it is supposed to look on a normal monitor or TV. So don’t be alarmed if when shooting using S-Log2 your images look a little darker and flatter than perhaps you are used to when shooting with a standard gamma. You will adjust the S-Log2 footage in post production to restore the brightness and contrast later.
The post production adjustment of S-Log2 is very important and one of the keys to getting the very best finished images. The S-Log2 recording acts as a digital negative and by “processing” this digital negative in post production (normally referred to as “grading”) we manipulate the large 14 stop dynamic range of the captured image to fit within the limited display range of a Rec-709 TV in a pleasing manner. This may mean pulling up the mid range a bit, pulling down the highlights and bit and generally shifting the brightness and colour levels of different parts of the image around  (see PART 2 for more post production information).
SLog-2 and 10 bit or 8 bit data.
Originally Slog-2 was designed for use on high end digital cinema cameras such as Sony’s F65 camera. These cameras have the ability to record using 10 bit data. A 10 bit recording can have up to around 1000 shades of grey from black to white. The A7s however uses 8 bit recording which only has a maximum of 235 shades from black to white. Normally 8 bit recording is perfectly OK as most transmission and display standards are also 8 bit. Shoot with an 8 bit camera and then display that image directly via an 8 bit system and nothing is lost. However when you start to grade and manipulate the image the difference between 8 bit and 10 bit becomes more significant. If you start to shift levels around, perhaps stretching out some parts of the image then the increased tonal resolution of a 10 bit recording helps maintain the very highest image quality. Photographers that have shot using both jpeg and raw will know how much more flexibility the 12 bit (or more) raw files have compared to the 8 bit jpeg’s. However they will also know that 8 bit jpeg’s can be also adjusted, provided you don’t need to make very large adjustments.
Contrary to popular belief heavy grading 8 bit footage does not necessarily lead to banding in footage across smooth surfaces except in extreme cases. Banding is more commonly a result of compression artefacts such as macro blocking. This is especially common with very highly compressed codecs such as AVCHD. The 50Mbps XAVC-S codec used in the A7s is a very good codec, far superior to AVCHD and as a result compression artefacts are significantly reduced, so banding will be less of an issue than with other lower quality codecs. If your going to shoot using S-Log2, some grading will be necessary and as we only have 8 bit recordings we must take care to expose our material in such a way as to minimise how far we will need to push and pull the material.
Getting Your Exposure Right.
When S-Log2 was developed the engineers at Sony produced tables that specified the correct exposure levels for s-Log2 which are:
As you can see the nominal “correct” exposure for S-Log2 is a lot lower than the levels used for display on a typical Rec-709 TV or monitor. This is why correctly exposed s-log2 looks dark on a conventional TV. The implication of this is that when you grade your footage in post production you will have to shift the S-log2 levels up quite a long way. This may not be ideal with an 8 bit codec, so I decided to carefully test this to determine the optimum exposure level for the A7s.
Correct Exposure.
The panel of images below is from the A7s recording S-log2 and exposed at the Sony recommended “correct” 32% middle grey level. The correct exposure was determined using a grey card and an external waveform monitor connected to the cameras HDMI output. Then the S-log2 was corrected in post production to normal Rec-709 levels using a Look Up Table (LUT – more on LUT’s in part 2). You can also see the viewfinder display from the camera. If you click on the image below you can expand it to full size. Sorry about the shadow from the laundry line, I didn’t see this when I was shooting the test shots!
From this you can see just how dark and low contrast looking the original correctly exposed S-log2 is and how much more vibrant the corrected Rec-709 image is. I have also indicated where on the cameras histogram middle grey and white are. Note how much space there is to the right of white on the histogram. This is where the extra highlight or over exposure range of S-log2 can be recorded. When correctly exposed S-log2 has an exposure range of 6 stops above middle grey and 8 stops under.
Over Exposing or “Pushing” S-log2.
If we deliberately raise the exposure level above the Sony recommended levels (known as pushing the exposure), assuming you grade the image to the same final levels some interesting things happen.
For each stop we raise the exposure level you will have 1 stop (which is the same as 6db) less noise. So the final images will have half as much noise for each stop up you go. This is a result of exposing the image brighter and as a result not needing to raise the levels in post as far as you would if exposed at the normal level.
You will loose one stop of over exposure headroom, but gain one stop of under exposure headroom.
Bright highlights will be moved upwards into the most compressed part of the log gamma curve. This can result in a loss of texture in highlights.
Skin tones and mid tones move closer to normal Rec-709 levels, so less manipulation is need for this part of the image in post production.
This last point is important for the A7s with it’s 8 bit codec, so this is the area I looked at most closely. What happens to skin tones and textures when we raise the exposure?
Exposing at +1, +2 and +3 Stops.
Below are another 3 panels from the A7s, shot at +1 stop, +2 stops and +3 stops. Again you can click on the images if you wish to view them full size.
Looking at these results closely you can see that when you increase the exposure by 1 stop over the Sony specified correct level for S-log2 there is a very useful reduction in noise, not that the A7s is particularly noisy to start with, but you do get a noticeably cleaner image.
Below are 4 crops from the same images, after grading. I really recommend you view these images full size on a good quality monitor. Click on the image to view larger or full size.
The noise reduction at higher exposures compared to the base exposure is very clear to see if you look at the black edge of the colour checker chart (the coloured squares), although the difference between +2 and +3 stops is very small. You can also see further into the shadows in the +3 stop image compared to the base exposure. A more subtle but important effect is that as the exposure goes up the visible texture of the wooden clothes peg decreases. The grain can be clearly seen at the base level but by +3 stops it has vanished. This is caused by the highlights creeping into the more compressed part of the log gamma curve. The same thing is happening to the skin tones in the +3 stop image, there is some reduction of the most subtle textures.
From this we can see that for mid tones and skin tones you can afford to expose between 1 and 2 stops above the Sony recommended base level. More than 2 stops over and brighter skin tones and any other brighter textures start to be lost. The noise reduction gain by shooting between one and 2 stops over is certainly beneficial. The down side to this though is that we are reducing the over amount of exposure headroom.
Given everything I have seen with this 8 bit camera my recommendation is to shoot between the Sony recommended base S-log2 level and up to two stops over this level. I would try to avoid shooting more than 2 stops over as this is where you will start to see some loss of texture in brighter skin tones and brighter textures. Â Exactly where you set your exposure will depend on the highlights in the scene. If you are shooting a very bright scene you will possibly need to shoot at the Sony recommended level to get the very best over exposure headroom. If you are able to expose higher without compromising any highlights then you should aim to be up to 2 stops over base.
Determining The Correct Exposure.
The challenge of course is determining where your exposure actually is. Fortunately as we have seen, provided you in the right ball park, S-log2 is quite forgiving, so if you are a little bit over exposed it’s probably not going to hurt your images much. If you have a waveform monitor then you can use that to set your exposure according to the table below. If you don’t have proper white or grey cards you can use a piece of normal white paper. Although slightly less accurate this will get you very close to where you want to be. Do note that white paper tends to be a little brighter than a dedicated 90% reflectivity white card. If you don’t have any white paper then you can use skin tones, again a bit less accurate but you should end up in the right zone.
If you don’t have an external waveform monitor then you do still have some good options. Sadly although the camera does have zebras, these are not terribly useful for S-log2 as the lowest the zebras can go is 70%.
Light Meter: You could use a conventional photography light meter. If you do choose to use a light meter I would recommend checking the calibration of the light meter against the camera first.
Mark 1 Eyeball: You could simply eyeball the exposure looking at the viewfinder or rear screen but this is tricky when the image is very flat.
In Camera Metering: The cameras built in metering system, like the majority of DSLR’s is calibrated for middle grey. By default the camera uses multi-point metering to measure the average brightness of several points across the scene to determine the scenes average brightness and from there set the correct base S-log2 exposure.
Auto Exposure:
When you are using S-Log2, auto exposure in most cases will be very close to the correct base exposure if you use the default Multi-Zone exposure metering. The camera will take an average exposure reading for the scene and automatically adjust the exposure to the Sony recommended 32% middle grey exposure level based on this average. In the P, A and S modes you can then use the exposure compensation dial to offset the exposure should you wish. My recommendation would be to add +1 or +2 stops via the dial. Then observe the histogram to ensure that you don’t have any significant over exposure. If you do then reduce the exposure compensation. Lots of peaks to the far right of the histogram is an indication of over exposure.
Manual Exposure And Internal Metering.
If you are exposing manually you will see a small M.M. indication at the bottom of the LCD display with a +/- number. In the eyepiece viewfinder this appears as a scale that runs from -5 to +5, in S-log2 only the -2 to +2 part of the scale is used. In both cases this is how far the camera thinks you are away from the optimum exposure. + meaning the camera is over exposed, – meaning under.
In the image above we can see the M.M. indication is +0.3, in the eyepiece you would see a small arrow one bar to the right of “0” , indicating the cameras multi zone metering thinks the shot is just a little over exposed, even though the shot has been carefully exposed using a grey card and external waveform monitor. This error is probably due to the large amount of white in the shot, white shirt, white card, test charts with a lot of brighter than grey shades.  In practice an error of 0.3 of a stop is not going to cause any real issues, so even if this was exposed by setting  the exposure so that you have “M.M. 0.0” the exposure would be accurate enough. But it shows that multi point exposure averaging is easily confused.
The scene above is a fairly normal scene, not excessively bright, not particularly dark. If shooting a snow scene for example the cameras multi point averaging would almost certainly result in an under exposed shot as the camera attempts to bring the bright snow in the scene down to the average middle grey level. If shooting a well lit face against a very dark background then the averaging might try to bring the background up and the shot may end up overexposed.
If you want really accurate exposure then you should put the cameras metering system into the spot metering mode where instead of taking an average of various points across the scene the camera will just measure the exposure at the very center of the image.
You can then use a grey card to very accurately set the exposure. Simply place the circular shaped symbol at the center of the viewfinder display over a grey card and set the exposure so that M.M is 0.0 for the correct S-Log2 base exposure. To expose 1 stop over with a grey card, set M.M. +1.0 and two stops over M.M. +2.0 (not flashing, flashing indicates more than +2 stops).
One small issue with this is that the camera will only display a M.M. range of -2.0 to +2.0 stops. Provided you don’t want to go more than 2 stops over base then you will be fine with a grey card.
Using White Instead of Grey:
If you don’t have a grey card then you can use a 90% reflectivity white target. As white is 2 stops brighter than middle grey when S-Log2 is correctly exposed the 90% white should indicate M.M +2.0.
Once you have established the correct exposure you can then open the iris by 1 or two stops to increase the exposure. Or halve the shutter speed to gain a one stop brighter exposure. Each time you halve the shutter speed your exposure becomes one stop brighter, so divide the shutter speed by 4 to gain a 2 stop increase in exposure. As always you should observe the histogram to check for any over exposure. White peaks at the far right of the histogram or disappearing completely off the right of the histogram is an indication of over-exposure. In this case reduce your exposure back down towards the base exposure level (M.M 0.0 with a a grey card).
Exposure Summary:
I recommend using an exposure between the “correct” base S-Log2 exposure level of middle grey at 32% and two stops over this. I would not recommend going more than 2 stops over over base.
In the P, A and S auto exposure modes, when using the default multi-zone metering the camera will set the base S-log2 exposure based on the average scene brightness. For most typical scenes this average should be very close to middle grey. This exposure can then be increased (brightened) by up to 2 stops using the exposure compensation dial.
In manual exposure the “M.M.” number displayed at the bottom of the viewfinder display is how far you are from the correct base S-log2 exposure. M.M. +2.0 indicates +2 stops over base. If using multi zone metering (the cameras default) this exposure will be based on the scenes average brightness.
If you set the metering to “Spot” you can use a grey card centred in the image to determine the correct base exposure and up to 2 stops of over exposure via the M.M. indication when shooting manually.
In Part 2:
In part two I will take a look at grading the S-log2 from the A7s and how to get the very best from the S-log2 images by using Look Up Tables (LUT’s).
I welcome feedback on my articles. If you have any feedback please let me know. I will make this available as a PDF for download once part 2 is completed.
What a great tutorial!
I have found +3 stops over middle to be great, especially in tandem with a Zebra of 100. I’m glad to see that it tops at 94 IRE at 90% white. Also, the sensor is capable of full swing all the way to 100+ IRE (assuming 109 IRE) so I think this method works great for S-Log2.
So far, my observation as been that +3 stops is safe. Further testing is necessary, though.
Thank you once again for an excellent tutorial.
At +3 stops your really pushing things. Skin tones will be recorded slightly higher and using a narrower range than for normal Rec-709. Each stop of 8 bit S-Log2 only has around 25 shades per stop once you go above middle grey (Rec-709 has around 35-40 shades in the normal skin tone range). Each stop you go up in exposure the actual scene has double the brightness range of the previous. So with the limited 8 bit data when you start over exposing skin tones and then pulling them down in the grade you start reducing your effective tonal range. You can see this in the test shots.
As the main noise reduction benefit occurs at +1 stop, is less of a benefit at +2 stops and really of almost no benefit at all at +3 stops there is no reason or benefit that I can see from shooting 3 stops over. You are reducing your tonal resolution in mid and skin tones, reducing your over exposure headroom to near 709 levels for no obvious gain over shooting at 2 stops over.
At +2 stops the skin tone shift in post production when going to Rec-709 is at it’s smallest so this is where you should see the best skin tones from the 8 bit recordings. Darker skin tones and mid tones will have the smallest shift at +1 stop, so I believe the sweet spot is between +1 and +2 stops, depending on the highlight range of the scene you are shooting.
There is no doubt that +2 has the best-looking skin tone.
Could it be that the +3 needs a ‘tweaked’ LUT? What else could explain the color shift?
Are you sure about your over/under numbers? My understanding is ‘correct’ is closer to +7/-7. Cine4 is known to have about 5.3 stops over and slog2 has a rated ISO about 1-2/3 stops over cine4, giving it that much more overexposure.
Absolutely sure. S-Log2 Base gives +6 stops over and 8 stops under. If the values were incorrect then the LUT’s wouldn’t work, plus the information can be found from any S-log2 plot in any Sony published S-log2 or S-log3 white paper.
Thanks very much for this very informative article.
Thanks for a very useful article. I’ve been very impressed by the footage I’ve seen from this camera around the web. Flat/Log profiles have been around for a long time for 8 bit cameras. There is plethora of them available for the 5d3. Why is slog2 on the Sony better than the various log gammas available for other 8 bit cameras?
Because it’s a real log profile, not some half baked black gamma/knee affair that often doesn’t truly tap into the sensors full output range.
Awesome article, Alister! Very helpful, since I just got the A7s and am trying to learn its capabilities. One question about shutter speed, I see in your outdoor shots it is 1/6400, while indoors 1/125. I thought one aims for a shutter speed double the frame rate? Or is there some other factor?
Thanks, looking forward to part 2.
doubling the frame rate is about the minimal frame rate to avoid flickering or ghosting for video. These were just pictures.
I ended up with those shutter speeds as that allowed me to start at f16 for the exposure range tests. So I could go to f22 and to f5.6 or even f4. I wouldn’t use 1/6400 normally, but the camera is so damn sensitive. Waiting for my 10 stop ND’s to arrive.
Thanks, Alister. You also mention time-lapse in one of your articles about the A7s, but I didn’t see it in the menus. Did you download a PlayMemories app, or some other way?
Awesome article!
Love ya work…
I am experimenting cutting between my fs700 with slog2 and my a7s with slog 2
Would all the info above also apply to an fs700?
Thanks again
Yes, I have been testing the same thing on the FS700 and my findings are the same. The LUT’s work with the FS700 too.
Hi Alister, GREAT Articel ! This is what alot of us have been waiting for. Your invited to a coffee m8t ;).
Just so I get this clear. If I go for a +2 overexposure and set my Zebra to 70% I should be fine regarding in correct exposure on Faces right ? Since I come from an ENG Background I am very used to shooting with Zebra set to 70%. This would make my life much easier finding the right exposure with the A7S.
Cheers Gil.
Yes you should be OK, but your zebras should only be on the darker skin tones. Then ideally you want to use one of my LUT’s in post to restore the correct levels as part of the grading process. This will be covered in part 2.
How can i donate ? I think i am missing an email address.
Hmm, I’ve just reset the buy-now link. It should be working now.
Enjoy ur Coffee 😉 Do you think you can make an Article on the different Color Modes inside the S-Log2 Profile and what differences they make ? (S-Gamut, Cinema, Pro, ITU709,…) Really enjoyed ur Article. Keep up the good work !
Thanks for the coffee!
I’ll probably end up writing a whole series of articles on the A7s covering all the different gamma and colour space options. But it takes a lot of time, so it may be a little while before they are all done.There is an article on colourspace here: http://www.xdcam-user.com/2014/05/what-is-a-gamut-or-color-space-and-why-do-i-need-to-know-about-it/
Although written for the F5/F55 the principles are exactly the same for the A7s.
Thx. Ill have a look into it. Totaly understand that these things take alot of time ! No Pressure here 🙂
Thanks mate!
One sorta weird nooby question-is there anything in your exposure and camera setup ‘ritual’ that is different using the fs700 and the a7s in s-log2?
Do you do white balance before exposure? same for different frame rates etc etc….any tips for setup with this combo?
No not really. FS700 and A7s both behave the same.
White balance depends on your post workflow. If you are going to use LUT’s (recommended) you should use preset white at either 5600 or 2300 (sunshine or tungsten).
This was a great read. thank you for taking the time to educated us. i look forward to part 2.
Geez your a good bloke!
One coffee coming your way…
A couple of other questions though-what did you do about ND? Any advice on this front?
And do you know of any good guides on shooting sony slog2 in the more general sense? Want to try and learn all I possibly can…
Lots to say here…
First, the a7S’s S-Log2 gamma using S-Gamut color is very accurate against Sony’s ACES input device transform (the same one for the F65). After transformation into the ACES color space, the image is highly linear, and you can do exposure compensation over a range of many stops with nearly perfect results. Here’s a little test I did that shows it: http://www.youtube.com/watch?v=41grTkW4lUs If you want to really understand what the camera’s sensor is capturing and not be confounded by the peculiarities of S-Log2, transforming into ACES is the way to do it.
Second, the S-Log2 gamma uses more codes per stop towards higher stops, not less. Here’s a chart that shows it nicely: http://community.sony.com/t5/F5-F55/S-Log3-vs-S-Log2/m-p/286167#M12596 So if you are properly transforming the S-Log2 levels to linear, there’s absolutely nothing to lose by exposing higher, so long as you don’t start to clip in the red, green, or blue channels in a way that the clipping would show up in your final rendered output. I’ve verified this with testing. Higher exposures are absolutely better, with less noise and less banding right up until clipping starts, and even then the only problem is the clipping in the clipped areas. Skin tones and color accuracy do not suffer with higher exposures. If anything they are slightly better. +3.0 on the exposure meter (if it actually went that high) is great, and clipping that would show up in the rendered output (after exposure compensation in ACES) starts somewhere beyond +3.0 most of the time. As a rule I try to expose between +2.0 and +3.0. +3.0 has a noticeable improvement in noise performance over +2.0, but I’m really not sweating the small amount of noise at +2.0 or even +1.0. This camera is really great for how I can vary the exposure over a wide range and still get great results after performing exposure compensation in ACES.
If you’re not using ACES, then you are using a LUT or some RGB level mapping that is very particular for how the image is exposed, and a higher exposure won’t necessarily be better. There’s really only one correct exposure if you’re not using ACES, and this is probably what you’re seeing when you say that skin tones suffer at higher exposures. It’s not a limitation of the captured image. It’s a limitation of the way you are processing the video. Every LUT that maps the recorded image to some display-referred color space like BT.709 expects a certain exposure as input. The BT.709 color space is not linear and thus not a good space for performing exposure compensation, nor is S-Log2. To do exposure compensation correctly, it must be done in the linear domain. S-Log2 itself is a weird curve that is more akin to an extended BT.709 gamma curve than to a true log curve. When selectively mapping S-Log2 RGB levels into BT.709 levels, you’ll get very different results depending on how the image was exposed and thus on which range you need to select. You have some leeway here, but really only one exposure will be best, and still not as good as when using ACES or when recording directly to BT.709 with a correct exposure.
Let me say it again: when you are not using ACES, there is only one correct exposure, and higher is not necessarily better. The correct exposure depends on which LUT (or level mapping scheme) you are using. It’s a limitation of your LUT, not of the camera’s recorded image.
Picture Profile 7 defaults to S-Log2 gamma and S-Gamut color. These are the correct settings if you use the ACES S-Log2 IDT or another Sony-provided LUT intended for S-Log2. But it is the wrong color space if you are selectively mapping a range of S-Log2 RGB levels to BT.709 levels. That’s because S-Gamut is shifted in CIE xy space relative to BT.709’s gamut, and you cannot correct such a shifting with standard color correction tools. Such shifting requires a LUT, or a matrix acting on linear values. So if you are trying to use standard color correction tools to convert S-Log2 to BT.709, you should shoot with S-Log2 gamma and BT.709 color, not S-Gamut color. The picture profile should be customized to change the color mode to something other than S-Gamut.
In S-Log2, ISO 3200 is not always preferable. We all know that higher ISOs are better, and I’ve told you that exposing higher is better for S-Log2. So what do you do when your exposure is maximized but the image is still too dark at ISO 3200? Do you leave the image in a middle or lower part of the S-Log2 range, or do you increase the ISO setting to shift the exposure into a higher part of the S-Log2 range? Noise performance and dynamic range will be the same either way, but you may face banding, especially in the shadows, if you don’t boost the ISO. That’s because S-Log2 has fewer level codes per stop at lower stops, and being just 8 bits on the S-Log2, the number of codes available in lower stops is pretty small. Of course if you can boost the exposure to expose into the upper part of the S-Log2 range AND keep an ISO setting of 3200, that’s best. But if your exposure is already maxed out, exposing into the upper part of the S-Log2 range is more important than keeping the ISO setting low. Boost the ISO as necessary to expose into the upper part of the S-Log2 range.
Finally, a note about zebras. Zebras show the level of the output Y-channel (luma). On the a7S, a zebra setting of 100 corresponds to 100% luma in the output, and 100+ corresponds to a luma greater than 100% in the output. But that’s not the max luma value for S-Log2: S-Log2 goes up to 109%. So you have about one stop of additional highlight range above where zebras at 100+ start to show. But don’t forget that you can clip in the R, G, or B channel and not clip the Y (luma) channel. Looking at zebras is good for making sure that you don’t clip on a sunlit white object, but it won’t help so much to make sure you aren’t clipping in one or two channels of R, G, or B on saturated colors and skin tones.
First of all, as stated in the article, LUT’s were used to do the transformations. There is a link at the head of the article to download the LUT’s and these LUT’s incorporate exposure compensation for both gamma and gamut. Included in the set are +1, +2 and +3 stop LUT’s, but it’s easy enough to produce LUT’s for +4 stops or whatever compensation you need. Using these LUT’s to correctly transform the S-log2/S-gamut to Rec-709 will be covered in part 2.
Yes, of course you could use ACES, but I doubt that many of my readers are using ACES with the A7s. Those that know ACES probably won’t need to read the article.
Of course you can expose at different levels when not using ACES. All you need is the correct LUT.
Yes you are correct that in terms of actual bits used, S-log2 uses more or less the same number of bits per stop for the higher stops. But you appear to be ignoring the fact that each time you go up a stop in exposure, each time you open the iris one stop or light the scene one stop brighter you are doubling the amount of light falling on the sensor, doubling the contrast range. So open the iris an extra stop and the contrast range doubles, but the amount of data being used to record that only increases by a constant amount. For the scene to be recorded accurately you would have to double the amount of data being used, but this is not happening, the data is only increasing by a small amount, so you have less data per stop relative to the scene you are shooting. So over expose to the point where you are shifting levels down in range in post and you will be sacrificing information about the scene.
If log really did offer the ability to over expose as much as you want without any consequence then there would be no need for linear recording or linear transforms. Why do you think that most log cameras are at least 10bit or better still 12 bit? The extra bits provide the buffer needed to allow for the lossy nature of log, but with the A7s we don’t have that luxury.
Using a higher ISO than 3200 will reduce your dynamic range as the clipping/saturation point of the sensor is mapped to the top of the S-log2 gamma curve. Raise the gain and you will push this clipping point beyond what can be recorded so you can never take full advantage of the full range of the sensor. Any gain in the shadows is likely to be negligible as gain will also bring up the noise floor. If you are shooting a dark scene when dynamic range is not an issue, then yes you could raise the ISO, but you could also add gain in post. Either way you will bring up the noise as much as the signal. It is MUCH better to alter the exposure by adding light or opening the aperture rather than altering the gain/ISO of the camera. If you are shooting in such a dark environment that you feel there is the need to add gain, then frankly you don’t want to be using S-log2, you would be much better off using a narrower range gamma with even more bits per stop than S-log2, possibly Rec-709 will give you a better result.
The primary reason for shooting with S-log2 is to gain a greater dynamic range.
In your article you didn’t say which LUT you used. It is pointless to talk about the optimal exposure if you don’t say which LUT.
I really don’t understand what you are saying about contrast. Contrast is relative brightness. If you increase the exposure by 1 stop, the bright parts become twice as bright and the dark parts become twice as bright. The contrast is unchanged. It doesn’t require any more data to record that image.
S-Log2 has progressively more levels per stop at higher stops, not about the same. You don’t lose a thing by exposing higher. You gain, actually, until it clips.
Log color spaces have more bits per sample so they can capture a large dynamic range without banding. It’s got nothing to do with any “lossy nature” of log. In fact a logarithmic scale is the optimal scale for encoding brightness levels, since human perception of brightness is logarithmic. A logarithmic scale makes best use of the limited number of codes available in an 8-bit or 10-bit or however many bit sample, regardless of the size of the dynamic range to be recorded.
Raising the ISO reduces the dynamic range because the shadows become noisier, not whatever you said. Of course it’s better to add light than to increase the ISO setting. But if adding light is not an option, increasing the ISO setting shifts the image into a higher part of the S-Log2 curve that has more precision, reducing banding (especially in the shadows). Naturally it doesn’t help the noise performance and you still need to be careful of clipping.
You are completely missing the point that while we perceive the world in a log way, a camera sensor does not and when you start manipulating levels, if you have not started off with an accurate representation of what the scene TRUELY looks like you will loose picture information. Log is perceptually lossless when you capture a scene 1:1 as the highlight roll off or effective data reduction (relative to true scene linear) mimics the way we perceive the world. But when you start raising those levels artificially so you can then reduce them later this no longer holds true.
If you really don’t think that changing aperture (or changing the amount of light in a scene, pretty much the same effect) doesn’t have an effect on dynamic range then you need to go back to photography 101. Of course both bright and dark objects get brighter if you open the aperture or add light. But in addition you will see further into the shadows as objects previously in the noise floor become visible. Open the aperture and Black will still be black, brighter objects become brighter, the gap between black and brightest expands, contrast increases as the brightest objects become brighter relative to the darkest. Do you not see this change in contrast on your monitor when you open and close the aperture?
According to your argument, that opening the aperture makes no difference to contrast or dynamic range, why do we even need cameras that can deal with big ranges? If a scene is too bright why not just stop down and close the aperture to deal with it? After all, according to you the dynamic range will be the same, according to you, exposure makes no difference to contrast range???? Obviously this is NOT the case. We all know that when trying to deal with a bright scene closing the aperture will reduce the brightness but also decrease the dynamic range as the low key parts of the shot are lost or squashed into the shadows where they can no longer be seen. And of course the opposite is equally true, opening the aperture increases the dynamic range and every time you open up by a stop you DOUBLE the dynamic range, a 100% increase, meanwhile the data increases by only around 10% if we are talking of the top 4 stops of S-Log2. I don’t understand why this is so hard to understand.
Go up one stop the scene range doubles, data increases by only approx 10%.
+1 stop = 100% recorded with just 10% more data.
+2 stops = 200% recorded with just 20% more data.
+3 stops = 400% recorded with just 30% more data.
+4 stops = 800% recorded with just 40% more data…………………
And you don’t think log is “lossy”????????? This is why there is a night and day difference between how much you can manipulate linear footage and log footage in post.
Of course there is every point in talking about exposure without stating which LUT is used because the LUT does NOT effect your exposure. Even if you over expose by a couple of stops and then use a LUT without exposure compensation to convert to your chosen gamut you can still correct your exposure in post. The whole S-Gamut primaries shift is a red herring as once you have applied your LUT you are no longer using those skewed primaries, you will now be in your new colour space. It may not be quite as easy to grade an over exposed shot after converting with a base exposure LUT, but it is still absolutely possible. Your whole notion of ACES having only one single usable correct exposure is totally flawed because the whole point of ACES is to move everything into linear space where it doesn’t matter where your exposure is as everything is allocated the appropriate amount of data for the brightness captured. Take over exposed S-Log and convert it to ACES and EVERYTHING you have exposed is converted to linear, with the same correct transform for the colour space, from black to white. Once you are in that linear space you can easily reduce the gain to get your desired output levels. You don’t need a different IDT for every exposure level, the transform from gamma/gamut to linear is exactly the same in every case the ONLY difference is the output level which can very easily be adjusted in the ACES linear domain.
The maximum dynamic range of a camera is determined in most cases by two things. One: The sensor noise floor and Two: The sensors saturation or clip point. When you add or remove gain these do not change, they are ALWAYS the same. In addition signal to noise ratio of the sensor remains fixed. Gain does not alter ratios, only levels.
With something like S-log2 the darkest thing that can be meaningfully pulled out of the sensor noise floor is mapped to the bottom of the log curve and the brightest thing or the saturation point of the sensor is mapped to the top end (109%) of the curve. This is how you transfer the maximum dynamic range from the sensor to the recording.
When you raise the gain/ISO BOTH the noise and smallest signals are amplified by the same amount and thus recorded at a brighter level. But you don’t gain any picture information as how far you can see into the shadows is limited by the sensor noise floor. You can’t suddenly see below the sensor noise floor. The signal to noise ratio does not change so there is no change to the dynamic range in the shadows, the amount of picture, relative to noise is EXACTLY the same. The ONLY change is in recording level. Brightness will increase, but we don’t see anything extra due to the matching increase in noise level. Do not confuse a brightness change with a dynamic range change.
When you raise the gain any signal coming off the sensor at or close to the saturation point is amplified. So, what was recorded previously at 109% now goes higher than 109% and can no longer be recorded as it is beyond the recording range of the log curve, so you LOOSE the ability to record highlights that the sensor is actually capable of capturing because the gain makes the total signal range larger than can be recorded by the gamma curve. For every 6db you raise the gain you will loose 6db (or 1 stop) from the TOP of your recording range because you push what was previously at 109% beyond 109%.
See: http://www.xdcam-user.com/2013/12/why-gain-is-bad-for-your-dynamic-range/
Also: http://provideocoalition.com/aadams/story/no_pain_no_gain/
This is why EVERY wide dynamic range video camera that I know of uses a fixed gain/ISO figure when capturing the widest possible dynamic range. This is why Arri/Red/F55 etc use exposure indexing to alter exposure mid point and NOT a shift in gain applied to the recording.
If you are shooting in low light and you add gain to the recording to boost the brightness, most of the benefit of adding gain to supposedly raise the recording into a fatter (more data) part of S-Log2 curve is lost because you must remember that gain not only raises levels but expands them. If lets say the darkest signal is at 20mV and the brightest at 100mV at 0db gain then that’s an 80mV swing to be recorded. Add 6db of gain (x2) and those signals become 40mV and 200mV, so now the swing is 160mV, the swing has also doubled. So to record this we will need twice as many bits. Shifting the levels up the curve will give you more bits, but not double, The net result is that adding gain gives very little, if any benefit to quality of the recording.
In a low light scenario where you feel gain is necessary it is so much better to use a different gamma curve more suited to this kind of scene, you would be crazy to use S-log2 unless you really need the dynamic range, in which case adding gain would undermine the whole reason for using S-log2.
Hi Alister – thanks for this terrific article. I was wondering what you thought of this thread and approach:
http://www.dvxuser.com/V6/showthread.php?327893-A7s-S-log2-too-noisy-Then-underexpose-cine4
Hmmm, not sure, looks good on the surface, but I don’t know how the base dynamic range etc is established as the numbers are all wrong. Sony’s Cinegammas are 460% curves (11.5 stop) so they cannot record more DR than than 11.5 stops, that’s impossible. S-log2 is a 14 stop curve. Middle grey is also placed at the wrong point in the S-Log2 curve (it should be +6 stops/ -8 stops) so I am very suspect of how the base exposure settings are bing established, including some of the ISO’s mentioned. Normally using negative gain is a really bad thing to do with any kind of high dynamic range log curve. I’d also like to know how the S-log2 was handled in post, because the DR difference between S-Log2 and Cine 4 is really quite big and I don’t understand why this has not been seen.
Very interesting; any chance you could join one of those threads, or may I direct people here/quote your comment? (they’re aware of this article).
This is, btw, the thread where he does the tests establishing those numbers – quite a few people are not sure about the numbers, but he seems to be:
http://www.dvxuser.com/V6/showthread.php?327893-A7s-S-log2-too-noisy-Then-underexpose-cine4
The numbers are wrong.
A really wonderful tutorial. Informative, generous and genuinely useful for a number reasons. It really is a nice gesture to share this valuable knowledge. Thank you. This will make a difference to how I use my camera and shoot log. Do you want sugar in that coffee?
Thanks for the coffee, your welcome.
Hey Alister,
Just bought you a cocktail and wanted to say thank you for sharing all of your knowledge! I’ve been following your posts since I bought my ex-1 back in 2008-09 I’ve learned so much from you and I’m truly grateful! Enjoy your drink 🙂
Glad you find it useful. Thanks for the cocktail.
In 2008, Alister, you helped me a great deal with my new EX3.
Now this very clear explanation of the Slog2 of my A7S is a huge help to me. I already used it as a B camera in a documentary for night shots and low light scenes using the 3200 ISO, and my editor is very pleased with the results. After reading this excellent article of yours, I will make a better use in the next projects. Thanks again Alistair for your precious knowledge sharing with the community of story tellers 🙂