ISO Fiasco!

Odyssey-Price-Promo-Banner-01-21-2016 ISO Fiasco!

One thing that’s becoming very clear with the on-going FS5 discussions is that the use of ISO with video camera is confusing the hell out of people. ISO is an almost meaningless sensitivity measure for a video camera. Especially with a video camera that uses different types of gamma curves. The ISO rating is there so that when you use an external light meter you will get the correct exposure, but it doesn’t necessarily tell you a great deal about the cameras actual sensitivity.

Take a Sony S-log camera. If you set the camera to 0dB gain and expose white card or piece of paper using standard gamma correctly (white at 90%), then still at 0dB gain switch to S-log does the piece of paper get brighter? The answer is NO it does not. White will drop to about 70% depending on the log curve chosen. That should be telling you that the actual sensitivity of the camera is NOT changing. If you set the camera to ISO, the ISO indication will change from 1000 ISO to 3200 ISO when you switch between standard gamma and log on an FS5 for example, but these are both 0dB gain and the picture brightness does not significantly change when you switch between standard gamma and log (the picture does become flatter, don’t confuse this for “brighter”. Use a waveform or histogram and measure the levels of a fully exposed scene).

If in both cases, S-Log and Standard Gamma, the gain is 0dB, even thought the ISO rating is changing, the actual sensitivity of the camera is not changing, the only thing that’s changing is the shape of the gamma curve.

The ISO rating changes because for normal gammas you would expose white at 90% while for log it is exposed darker at 60%. So if using an external light meter you want the light meter to tell you to close the aperture relative to standard gammas so that white ends up at 60% (the lower white level makes space for more dynamic range above white).

When you change your external light meter from 1000 ISO to 3200 ISO the light meter will tell you to close the aperture by a stop and a half and that will give you that correct and now darker log exposure.

This is why I hate the use of ISO in video cameras because it is miss-leading people. The FS5 (and the same applies to the FS7, F5, F55 etc) isn’t actually any more sensitive to light in log than in standard gammas (video camera sensitivity is governed by the sensors efficiency at converting light into electrons and we cannot change the sensor). Neither does it have any more gain at 3200 ISO in log compared to 1000 ISO in standard gammas. Both are 0dB. That’s why when you shoot at 3200 ISO in Log you don’t get any more noise and you don’t get any additional NR artefacts…. because the camera isn’t actually becoming any more sensitive and it’s not adding any extra gain, it’s just a rating change to make sure an external light meter would give you the correct log exposure. If people actually went back and used the correct and appropriate terms for video cameras: ie: dB we wouldn’t have all this stupid confusion, it would be obvious that 0dB = 0dB.  The gain and thus sensitivity in standard gammas with 0dB is actually the same as in Log at 0dB. If people actually took the time to look at their monitors it would also be pretty obvious that the sensitivity does not change.

Sadly all this is resulting in some pretty ridiculous statements like: I use log in low light because it’s the only way I can get 3200 ISO without adding extra noise. The reality is that the only thing that’s really changing is the little number on the viewfinder overlay, in standard gamma it says 1000, in log it says 3200. But the sensitivity and noise levels are not changing. S-Log3 does give raised shadows, but those raised shadows will show more noise due to the raised shadow levels, but for a like-for-like brightness they are no more noisy than any other gamma. Using the wrong terminology or miss-understanding how a video camera works is resulting in very silly mistakes. Once upon a time video camera operators used to rely on monitors and waveform displays to get exposure and gain levels right. But this isn’t cool anymore because we are all now “Directors of Photography” and DP’s don’t use gain or shutter speeds they use ISO and angles. This is nonsense and the desire to sound hip and cool is resulting in poor or miss-guided camera operation.

29 thoughts on “ISO Fiasco!”

  1. In a college physics course I was impressed by the concept of “misconception” and how easy it is to come to a false conclusion. I most assuredly came to the wrong conclusion about the relationship between gain and ISO. Thanks for clearing up the confusion.

  2. I own the FS5 and have to admit I was disappointed with the lowlight ability—but I was coming at this camera with the mindset of the a7s, as it’s been years since I worked with a true video camera—my mind has been acclimated to ISO and not gain.

    A positive outcome of this debate (argument? has it been civil enough to be called a debate?) is giving me and others a newfound understanding of the relationship between ISO and gain levels in this camera. It’s helping me shoot smarter with the FS5, a camera I am still very happy with.

    That being said, I’ve yet to have a shoot negatively affected by any of the image artifacts caused by high gain and high compression. Perhaps I’ve been lucky in that I simply haven’t screwed anything up yet! Reading your information about gamma curves has been a huge help in learning how to use this camera properly, and has improved my a7s footage as well.

    The FS5 is an imperfect camera. But it is capable of stunning images, I believe, when used correctly in the appropriate circumstances. I do hope Sony supports it with firmware updates for a long while to come. I love it’s small size and flexibility, and I am very, very eager for the raw firmware upgrade to have the option for less compression when it’s necessary. I’m also eager for companies like Atomos to produce smaller 4K recorders to help in that.

    1. Yes it’s an imperfect camera, but I have yet to find a camera that is perfect. I’m sure there will be many firmware updates in this cameras future. Of course at some point we will get raw and that will bypass all these “issues”.

      1. Of course! I think that’s what I meant to say about the imperfection—using any camera excellently necessitates knowing its limitations very, very well. This discussion about gain and noise is very enlightening on that end—it exposes the weaknesses of the camera so you know how to use it better.

        1. While 12 bit cannot record 14 stop linear using conventional binary it can be done with 12 bit floating point math which is almost certainly what Sony do. In the normal speed modes the FS7 does deliver 14 stops of linear raw. The problem is that floating point rounds data up, so noise for example instead of having 4 or 8 shades/values only has 2 values, small and large, so this makes the noise very coarse and grainy and the same with shadow areas.

          ACES uses floating point to get a range of over 20 stops using 16 bit. It is also likely that the F5/F55 use floating point to put a bit more data in the shadows, even though these are already 16 bit.

          1. Much of what is in that thread is incorrect because no one considers the use of floating point math, which is how Sony get 14 stops in to 12 bits. 12 Bit linear has fewer steps in the bottom 3 or 4 stops (don’t forget though that there are 8 stops below middle grey) but a lot more data in the brighter stops than S-log. So pros and cons to 12 Bit linear. It can be incredible when exposed properly or with nice bright scenes, but is not so good in low light or dark scenes. The difference between 12 bit and 16 bit linear raw is all in the shadows and it’s quite a big difference.

  3. Hello. Thanks for this article. I appreciate the primer.
    However, there are a couple of typos in the sixth paragraph. Where you say “… isn’t actually any more sensitive to light log than in standard gammas”, you’re missing an “in” between “light” and “log”.
    And in “Neither does it have any more gain at 3200 ISO in log compared to 100 ISO in standard gammas”, I think you mean “1000” rather than “100”.

  4. Hey alister, I got to use a fs7 for the last two days and could not find the histogram, tried both cine ei mode and custom s and q and normal mode but could not find the option or what I thought was the option in the vf menu would not let me select it. I’d prefer to have waveform as that’s what I’m used too but at least histogram is better than nothing, without it other than my zebras I couldn’t tell what was decent exposure. Shooting in flat light made it even harder. Do I have to use a screen or external vf like graticle to get exposure tools? Thanks I had fun but like you said I felt pretty clueless on exposure

    1. FS5 or FS7? The FS7 has a waveform, but until the next the V3 firmware comes out later this month the Waveform is only available in a very limited number of modes. For the F5 and FS7 I have a Zacuto Gratical and that makes using these cameras so much easier.

    2. Another way to get beautiful histograms and waveforms on FS5 or FS7 is to use SmallHD 502. The Sidefinder attachment is pretty useless, so I agree the Gratical is far superior solution for EVF. But if you want a small on-camera monitor that won’t compromise the small-camera aesthetic, it’s a great solution. Also supports LUTs, so you can use it with FS5 (which doesn’t support LUTs) to gauge exposure correctly when overexposing SLOG.

      One drawback with the 502 is that it does not provide numerical overlays on the histogram or waveform, which makes it hard to tell exactly where the data is falling.

  5. Thanks Alister, this is a very informative article.

    I think the reason many young videographers are using ISO instead of gain has more to do with the fact that many who started videography in the last 8 years learned on DSLR’s. As you know DSLR’s rate gain in ISO. I don’t think it has anything to do with people trying to sound hip or pretending to be DP’s. It’s just how they’ve learned.

    Totally agree that decibels is the best way to describe gain.

    1. And frankly ISO is also pretty meaningless in a DSLR, gain is much more meaningful, lets face it, very few people use external light meters. Would you be able to tell me the native sensitivity ISO (where you get the best DR) of a DSLR if I handed you one that you’d never used before?

      If what your suggesting is that the young photographers turned DP’s can’t be bothered to learn the correct terminology or are not being taught correctly then I would completely agree. But it’s not an issue just with the ex DSLR users. I’ve seen many a seasoned old school TV cameraman turn to ISO and abandon gain. Part of the problem is that producers expect you to use film terms, again mostly because it sound cool and they don’t know any better.

  6. I’m glad you said this because the industry use of “ISO” vs “gain” is truly getting stupid. It’s confusing hundreds of thousands of people and it makes matters worse.

    It’s such an antique idea that really apply to the chemical sensitivity of film stock. Here we all are tryin to translate and apply it to analog image sensors that don’t change their sensitivity to light one bit when gain or “ISO” is increased. Using the phrase “more sensitive ISO” is completely wrong. “Circuit amplification” of the light you already collected from a fixed ISO is a better way of saying it.

    I have been wondering this for a very long time.

    Thanks for saying this. You should probably repost it once a month so people don’t forget. It is a very important concept.

  7. Excellent explanation, as always. I know you are sick of hearing about FS5 “problems” but can you take a look at this and give your insight as to what is happening? I have also noticed similar aliasing between high contrast objects in 1080p but without the strange color issue.
    This occurs only in normal 1080p mode, not in UHD or center scan modes, and it effects both internally and externally recorded footage. When I record the same scene in UHD and then down convert it, the aliasing is not present. It seems like the FS5 is line-skipping when it reads out the sensor rather than super-sampling the entire 4K frame. I would expect that in the high frame rate modes but not in normal, 24p.
    Any insight?

    1. To be honest I’m not sure what is going on in your example. It looks a lot like CA, but it also looks quite digital and I’m a bit surprised by the quite jaggie appearance of vertical and horizontal edges as I don’t see this with my FS5. The test shots I did this week feature tree branches against a very bright and clipped in places sky yet there is no sign of anything like this.

      Similar effects have been seen with other bayer sensors it may be sensor bloom or CA from the micro lenses on the sensor, take a look at the tree example here: http://hugin.sourceforge.net/docs/manual/Chromatic_aberration.html.

      One question did you stick with the preset white balance and use S-Gamut/S-Gamut Cine or did you use a different color matrix or offset white balance. Sometimes you get color clipping artefacts if you offset the white balance or use the wrong colorspace.

      1. That’s my Vimeo post. I was using PP9, only thing I changed was color mode, which was set to S-GAMUT3.CINE/5500K. I’ve only had the issue appear in a few of shots, well at least that I’ve noticed. That clip being the clearest example. I was able to reproduce it by viewing thin tree top branches against an overcast sky. It appears when using other gammas and color modes.

        I’ve passed on the details (including the original clip) to the company who sold me the camera and they say they have passed those details on to Sony.

        As pointed out by someone on the vimeo page, it does seem very similar to an issue that was found (and fixed/reduced by Canon) in the C300. See this story: http://nofilmschool.com/2013/02/canon-firmware-update-c300-color-fringing

        Lets hope Sony can do something to reduce it in a firmware update.

        1. Yes, I remember the same issue on the C300. I believe it is a form of sensor bloom as it only occurs against a bright background, many other cameras suffer from it. Adjusting the de-bayer algorithm may be able to reduce it or even hide it entirely so hopefully we will see a firmware update to minimise it.

  8. All this internet hate because this camera makes weird fuzzies when it’s focused on a poorly composed shot waving back and forth on the edge of a computer monitor is insane. It’s amazing how quickly fanboys can flare up over something and fuel the fire. None of the real world examples of this thing actually shooting meaningful content (you, behiri, bloom) show anything other than beautiful images and neat little stories. I wish I had a platform to shout from the rooftop “Everyone shut up and shoot something real, outside of your office, or some obscure, out of focus something!”

    It’s so refreshing to hear you talk about ISO vs Gain. Consumer tech has brought on a generation of shooters who don’t remember that you never, unless it’s a “must get” situation go to 12db of gain. My gain presets are typically 0, 3, and 6. If I need to go higher than that, something else has to change. I’m currently shooting a show for BBC Worldwide on C300’s in Alaska and the preferred method of controlling exposure is by changing ISO via the handle dial rather than aperture. It drives me nuts, and I’ll often find myself or other people shooting at ISO 10,000 with too much ND on because it’s just not the way I’m used to doing things. They prefer it though because the steps between exposures are smaller than with the EOS iris. Which leads me to my next point…

    I cannot wait to shoot reality with the little FS5. I have been telling the show’s DP since it’s announcement we should be shooting on FS7/FS5 combo. Then they can keep their shallow depth of field look and have smooth, perfect exposure control with this variND. What a brilliant idea!

    I’ve owned an FS700 for about three years and exposure to weather is starting to rear it’s ugly head with the poor little buggar so I dove in and bought an FS5 to replace it. I think it’s pretty much a perfect replacement, slightly upgraded tech, massively upgraded design, but still easy to pay off and have a solid little grab and go camera system.

  9. Hi Alister,
    thanks for all your articles and videos which are really helpful.
    After watching your two hours videos it was quite clear for me the difference between log and hypergamma. And why you suggested to overexpose log of 1.5 stops in order to record more information for the shadows.
    But now with this last article I am a little confused again.
    To resume if I understand well in FS5 we have 1000iso=0dB in Hypergamma and 3200 ISO=0dB in Slog.
    So now you say in this new post that if I use a light meter and I switch from Hypergamma to Slogin order to expose correctly, I will close 1,5 stop because I am going from 1000 Iso to 3200 Iso.
    But you said that using log it’s better to OPEN of 1,5 stop, right?
    So If I follow the advice of this new post AND the advice of your video, I will First close of 1,5 stop to expose ‘correctly’ and then Open back of 1,5 to have more details in the shadows.
    So finally I will find back exactly the same value of stop that I would have used in hypergamma.
    Is this right?
    Many thanks!

    1. Yes, that’s right.

      The light meter would tell you to close the aperture by 1.5 stops, that would give you the Sony recommended exposure where white is only 60%.

      Then to get the best from log you want to open the aperture between 1 and 2 stops, so yes you end up back where you started, although white will still only be at about 70%.

      If you think about it, if you just went by the light meter a higher ISO means closing the aperture, so you have less light falling on the sensor. What happens when there is less light on the sensor? Normally a noisier picture. But when we open the aperture back up again the light falling on the sensor is the same, so no additional noise. So now we have the extra dynamic range of log and no more noise so after grading it is no noisier than of we had shot it conventionally, but generally end up with a better looking image.

  10. Here are two more examples of very pronounced artifacts in FS5 footage.

    1) Another example of aliasing in 1080 mode. At 2:30 it can be seen around the edge of the driver’s window, at the bottom of the door and in the window frames on the building in the background. Are you sure that the FS5 is doing a full sensor readout in 1080 mode? It obviously employs line skipping in the higher frame 1080 modes so I’m wondering if maybe this applies to all of them. That would explain the aliasing and strange processing issues we have seen in 1080.

    2) Extreme banding in 4K mode. It’s not a complex scene so I’m not sure why the camera needs to apply such heavy compression. It was shot in SLOG3, perhaps the shooter didn’t expose it correctly and the reduced data in the mid tones combined with heavy compression caused this ugly banding? Just a guess.

    1. In super slow mo, just like the FS7, F5, F55 etc the sensor is read at a reduced resolution so it will suffer from some aliasing. On the FS7 version 3 firmware will get around this by allowing you to use a center crop of the sensor. The F5/F55 already have center crop bit can also accept a 2K low pass filter that eliminates the aliasing. In normal 1080 the camera reads the sensor in 4K and the down conversion is done in camera so aliasing is extremely low. Normally in 1080 the FS5 produces a first class image. However unfortunately a few tests that have revealed some artifacts at high gain levels or in 4K where it must be remembered the codec is only 8 bit and very highly compressed have been blown out of all proportion.

      S-log3 and an ultra compressed 8 bit codec really is a far from ideal combination. To make it work you must get the exposure just right. Get it right and it can work, but it isn’t easy. Again though, this is to be expected. In addition LUT choice is crucial to avoid banding. As the poster of the video you have linked to states the 10 bit HD S-Log3 does not have these issues. Why is the 4K so compressed? Because the camera uses SDXC cards.

      It now appears to be contest to see who can break the camera and it’s codec because somehow producing flawed material makes you a better camera operator than producing virtually flawless material.

  11. Hi Alistair,
    What is the base ISO (= 0dB) in cinegamma of the A7S and A7SII (if it is different) ? Thanks for all the information you provide.

Leave a Reply

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