Have you missed all of the drama surrounding Tony Northrup and his ISO claims? Let's take one final look.
Last week, Tony Northrup released an interesting video on ISO. In this video, Tony makes a few interesting claims; some were known by me, and one, in particular, was new to me. Since this video came out, the photography industry has been in a major debate about what Tony said and if it is partially incorrect. You can watch the entire video here:
Patrick showed me Tony's video, and I wanted to put two of the claims to the test. One claim was that ISO numbers may have been purposefully skewed by camera brands to make it appear that newer camera models have better ISO performance than older models. The second claim was that (in some cameras) you can get similar image quality from shooting a raw image at ISO 100 and raising the exposure in post versus choosing a higher ISO in camera. You can watch this video here:
Then, Dave McKeegan created a video claiming that Tony was spreading false information about ISO and that Fstoppers (Lee Morris) basically agreed with everything Tony said. You can watch this video here:
If you would like to go down the rabbit hole of ISO, Noise, and ISO invariance, I highly suggest reading this article on Photography Life. The only thing that I can point to that we may have gotten wrong by failing to mention it is that some noise can be produced by "Back-end read noise" (although I do mention it without knowing the technical name of it at 5:15 in our mythbusting video). This is the type of noise that Dave Mckeegan mentions in his video. Tony didn't mention this directly, and although I didn't mention it by name, I did point out some extra artifacts found on the image boosted in software rather than the camera.
Not all cameras are ISO invariant, but the ones that are or the ones that are very close may be capable of similar image quality when boosting exposures in post; we certainly proved that with the D850. The key here is that "back-end read noise" has gotten so clean in some cameras that it's hard to see a difference when comparing high-ISO photographs versus exposure-boosted photographs.
Well our curiosity got the better of us again, and we decided to make a second video. In this video, we test seven different raw processors to see which one could boost an underexposed raw file four stops and produce the best-looking file without noise reduction or sharpening enabled.
In the end, we found that Capture One may have produced the best-looking file without noise reduction or sharpening, but once you start adding those in, you can get almost identical looking files out of any of the programs.
So, what have we learned? Some, not all cameras are ISO invariant, which means in theory, you should be able to boost exposure in post and get identical noise to a correctly exposed image shot at a higher ISO. But, back-door read noise will introduce some noise when you do this. But, back-door read noise has become so insignificant in some cameras that it may not be worth worrying about.
So, should you walk around snapping black pictures and then raising the exposure in post? No, that would be stupid. Can we be done now?
Thanks for looking at the processors.
Hey, D850 is a dual gain sensor right? so that means ISO invariance ranges are split at >400 and <400.
http://www.photonstophotos.net/Charts/PDR.htm
See the hump in DR at ISO400.
Oh man check out the A7iii on that chart. I had heard that was dual gain and boy that's pretty stark. BRB setting my max Iso down to 640...
Sorry Lee, You should have just let it go. This is nothing of real significance to anyone who's just getting on with photography, this is pixel peeper stuff. Said it yourself, wasting more time on it.
You read the comments section didn't you! Golden rule of posting articles. Don't read the comments ;)
My honest first thought in all of this is was "Neat.... who the Eff cares?!" None of this makes anyone a better photographer. Pixel Peepers are making photography more complicated than it needs to be.
It might matter if they're wrong and the difference is clipping detail vs. not. I think that's an important distinction that has practical applications.
I guess... Just seems so trivial. They're all capable of taking beautiful photographs. Shoot more. Pixel peep less. Did B.B. King play the greatest guitar ever built or did he master the one he played with? The music sure is good.
I get what you're saying but we're not talking about the same thing. In your scenario, does it matter if B.B.'s amp is set to 6 or 7? Probably not, both are at a pretty similar volume level. But I'm pretty sure if he tried to record at "0", there's not much anybody could do to salvage that take, no matter how good he was.
Now what if somebody came along and said, "yeah, well, you actually can record with your volume turned all the way down, because the volume knob isn't doing anything". Would it affect somebody who already knows what they're doing? Probably not, because they would ignore it. But that doesn't justify spreading poor information to those who are trying to learn.
Fine detail in darker areas of an image IS important sometimes, this isn't just a pixel peeping issue. Sometimes the difference between capturing that detail and clipping it altogether is 1 stop on the camera. If you're controlling that stop with ISO, it matters.
What a weird hill everyone's choosing to die on.
"Fine detail in darker areas of an image IS important sometimes, this isn't just a pixel peeping issue. Sometimes the difference between capturing that detail and clipping it altogether is 1 stop on the camera. If you're controlling that stop with ISO, it matters."
That's what Pixel Peeping is.
Story telling > literally every technical aspect of the camera.
Get back to me when you're printing at 40x60 and your landscape is mush in the darker areas instead of having texture. You don't have to be a pixel peeper to see that. It's plain as day.
Just because YOU don't care doesn't mean it doesn't matter. The point is that there are practical applications to this issue beyond comparing a little bit of noise when zoomed at 100%. It may not apply specifically to how you do things, but it's pretty arrogant to project your workflow onto the rest of the world.
Eh. Two sides of a coin I guess.
The point wasn't to argue that you shouldn't master your camera. It was that the debate being perpetuated by Mr. Northrup is trivial. Similar to bang for your buck on 50mm lenses. At some point the invested energy arguing over ISO is diminishing returns when the difference is so minute.
I agree with you sir and Mr. Michael Holst. Who the hell cares. One thing of note though....light coming into any camera has a measurement. However anyone wants to quantify that measurement....so be it. We are stuck with what has been given to us. Let's move on now.
I posted this on the last article but I was late to the game so it probably got buried in the comments. I think the problem with looking at the issue in the context of noise alone is misleading. I think we should be looking at detail (not noise) in clipped areas.
From a practical perspective (not arguing scientifically how the camera actually works), if you can take a shot where you properly expose at a high ISO and then the same shot, clipping detail (i.e. the left side of the histogram), by lowering the ISO only, according to Tony's logic we should be able to recover all of that detail in post.
The challenge with this test is that the dynamic range on cameras like the a7riii and D850 is larger than the native ISO range, so we have to rely on scenes that are unevenly lit to perform the test: i.e. where part of the scene is properly exposed at the higher ISO but part of the scene is dark enough that it falls out of the dynamic range of the camera when we lower the ISO.
Since I believe you would actually lose shadow detail in that example, it stands to reason that ISO is actually doing something more than these tests would lead you to believe.
I'm certainly open to being wrong here, as it's just a guess, but logically it seems to make sense.
I would suggest watching this video from Filmmaker IQ which illustrates very well how ISO and DR work together: https://youtu.be/2sshGdMgJxQ?t=1193
It is a good video but it doesn't really address my main point. Yes, it does suggest that dynamic range is affected by raising ISO (which would, in a different way, indicate that ISO is doing something more than just acting as an Exposure slider). But my point is still a very specific one: if you can clip detail (not talking about noise here) by lowering your ISO only, then it stands to reason that ISO has a very real effect on the overall exposure of the image. Using the "invariant" logic that's being presented in Tony's video, you shouldn't be able to do this. If ISO only affects the image once it's been captured (behaving the same way as the Exposure slider in Lightroom), then all of the detail that's been captured should remain the same regardless of what ISO you use, and therefore you should always be able to recover the same amount no matter how much you underexpose using a lower ISO. But I don't think this logic holds up because I think it is possible to lose detail by underexposing too much using a lower ISO. If that's the case, then it means ISO matters in the capture process.
While I do agree that a lot of the "comparing noise at different ISOs vs adjusting in post" arguments are largely about pixel peeping and being excessively argumentative, I think this particular point actually has real world implications. High contrast scenes being one. Just because there are darker parts of a scene doesn't mean you don't want detail there. Maybe you're shooting a long exposure at sunset with a 10 stop ND, using a small-ish aperture for DOF. At a certain point the difference between 1 stop of light using longer shutter speeds could be an extra 2 or 4 minutes rather than fractions of a second. But if you're just barely clipping your shadows and don't want to burn the extra minutes, raising your ISO a stop could save you. However, in Tony's example, this wouldn't work... because all of the "real" information is only captured through shutter speed and aperture.
This is a good argument, to add, on the opposite end of the histogram, I’ve found you can purposefully overexpose non-invariant cameras like the 5d3, D700 to get soft, creamy, film like skin. This isn’t the same as upping the exposure slider in post, like you said. This isn’t as possible in ISO invariant cameras, and it shows. People talk about canon skin tones, and it’s not just the colour, it’s the tone/curve and exposure *in-camera*
Wait—what?
Would it not stand to reason that a camera that is non-ISO invariant be able to be “pushed” like film?
Kinda the antithesis of capturing a file with an ISO invariant camera.
I can tell you from experience with Canon that you can expose for the shadows, maybe pushing a stop over your “proper” exposure (or instead of under to protect highlights) and the resulting file has a different look than a frame exposed to middle grey and then pushed in post- creamier skin midtones, softer shadows that don’t need recovery and don’t get grainy in the first place, almost like there’s a tone curve applied.
My take is that ISO invariant cameras aren’t truly “exposing” an image, just capturing the scene data and as long as you’re within 4 stops under, you have your file, your exposure choices in-camera don’t matter (within reason of course), where ISO variant cameras can be “exposed” to taste.
To base my opinion, my favourite Canon cameras were actually the 5D2/3 and 6D. I find my IV and Rs don’t have the same “magic” to a slightly overexposed in-camera image, maybe due to ISO-invariance and slightly different colour science.
Heavily recovered ISO-invariant files, while maybe not noisy, have weird tonal transitions on skin, blocked up smudgy shadows, colour shifts, etc. Again, IMO, because the file is being numerically transposed in post to a brighter level, not the actual “exposure” being “pushed” at the time of capture.
I get that you can pop skintones in post many different ways and it can look creamy and bright, but there’s a certain je ne sais quoi about a 5D3 + 85L portrait that is purposefully over exposed in camera at golden hour that I believe no system can touch for that type of look. Obviously landscape photographers are laughing right now, but I find a camera more appealing when it “has soul” over being a computer box, again, for my purposes.
Note: These photos were slightly overexposed in camera and then brought *down* in post a bit to retain dress detail, so not the best example of my theory but there was no need to recover shadow detail and the tonal transitions and shadows are clean.
Bender out.
Another from the same event, exposed a bit over. If I shot this under and recovered, it wouldn’t have the same softness. This isn’t a preset, hasn’t been manipulated other than a third of a stop pull down. It’s an exposure choice in-camera. If you push an underexposed ISO-invariant file to this exposure, the shadows will be mostly noise free, but not smooth and clean in a transitional tones way. It’s subtle but it’s a thing. You ever see the shadowed underside of a bride’s arm or the hand of a groom in the small of her back that has been grossly recovered? It looks like ashy burnt leather. But pull down the highlights of a barely overexposed canon file and you keep your scene’s dynamic range and get clean shadows and tones.
Sure, 15 stops is nice, pointing a camera at any subject in highlight weighted metering and getting a usable file is nice, but shooting lazily or in some cases poorly and relying on recovery instead of exposing for your subject and trying to contain the scene’s DR through fill flash or other methods and gaining a subjectively nicer file and lessening my workload by getting it right in camera is how I shoot. ETTR instead of the left in my case.
From a completely non-scientific, but practical point of view I have tried to do exactly this. I often underexpose on purpose I have found that in general it's easier to bring back pleasing looking details from shadows than it is from highlights. I say "pleasing looking" because it's subjective. I have taken many "black" images and created stunning photos. I have also taken largely whited-out photos that were non-recoverable garbage. I don't know "why" this is, but I just know that it "is."
I think the sticking point here is that there's a difference between "underexposed" and "clipped". A picture can be underexposed while still having all of the information contained within the histogram. At that point, it's still recoverable through post processing. But if there's a point in the scene that falls darker than what the histogram is able to include, you've *actually* lost detail there. The question I'm posing is whether lowering your ISO alone can affect the final outcome of the image once you cross that threshold. My vote is yes, but using Tony's logic it shouldn't matter. I realize I'm approaching this a slightly different way than what the main discussion has been (noise at different ISOs), but I think it paints a more accurate picture of how ISO affects an image during the capture process.
You just explained it very well and I think you might have touched on a determining factor in this discussion. I had not thought of clipping!
What you described is exactly the same in audio levels. You need enough audio level from your source signal to overcome the inherent actual, audible noise (just as you need a minimal amount of light to expose a sensor) of the environment the audio source is in and even the internal noise made by the gear used to capture it, but not so much audio signal that you distort your pre-amps (or anything else along the signal chain) resulting in loud, staticky distortion.
In the early days of digital audio this was an even more serious consideration as the boxes that converted analog audio signals into digital signals were low "resolution" compared to today. When you clipped a 44.1k 16bit signal it just turned to absolute noise. Un-recoverable... as opposed to analog distortion which, although can become almost pure un-listenable static at some point, it does so gradually. The solution in audio was higher resolution audio and better convertors with lower internal noise just as in photography the answer has been higher resolution files capturing more information along with "light convertors" or "sensors" that could render the light (even when there is too much or not enough) in an acceptable way... but it has its limits. Clipping!
On modern, high end cameras the space in between the clipped area is vastly usable all across the range between a noise floor of blackness and the whiteness of overly-intense light. Dynamic range! Just as in photography, this is also called dynamic range in audio. It's amazing how many underlying principals between the light and sound are so similar.
"It's amazing how many underlying principals between the light and sound are so similar."
Yep, it's all signal processing, whatever that signal is!
So far it has to do about "waves" for both audio and light, fun, I keep my cameras and my audio gear by the same shelf 🙃
The Filmmaker IQ video on Dynamic Range explains it very well, I find.
Just saw this and watched the video... holy crap that was a great video! Thanks for tip!
Jesus Lee, time for a new headshot? ;)
Ruthless :)
What? It doesn't look like me anymore?
You look fantastic! Living your best life! ;)
Ignore me, I’m just jealous of the glorious bouffant!
Actully, it doesn't! :-D Guys who have hair and don't grow it may come to regret that later in life LOL...
All results from the softwares were expected. But I was surprised with your conclusion at the end.
What I understand and knew was, every software will have their own philosophy on editing files so results will vary. Software inside the camera has its own way of processing it.
The subtle differences are purely based on that. This makes what Tony was talking about still valid. Without needing to go through all technical details, it is safe to say, iso markings on cameras is not what it used to be with the film era. There are no different size of photon particles.
In my opinion what makes a better difference is the size of the pixels like sony S line proposes. That’s more similar to the photon size in film.
So, I still can see Tony is making a good point. And Lee again seconds that.
I'm going to make a response video and keep this going forever because there's nothing more important than technical minutia
https://media.giphy.com/media/3og0IDV8KQpDZaMfVS/giphy.gif
Unless you've become PhD computational scientist, I think it's best to just let it be Tony. You've been proven wrong. Just make an apology video and we'll move on.
@Tony we can have a private discussion if you want any explanations or start by reading my comment below.
This is your sort of thing so everyone expects you to keep kicking the can down the road.
There's one major problem here, and you basically arrived at it at the very end of the article as a sarcastic quip, but failed to realize that it pokes a sink-hole in your conclusion.
"So, should you walk around snapping black pictures and then raising the exposure in post? No, that would be stupid. Can we be done now? "
If indeed ISO invariance leads you to the conclusion that ISO is akin to moving the exposure slider in post, then this joke-like supposition, should actually be possible given an exposure slider that goes higher than 4 or 5 stops (I believe Capture One can do this). But we know this is absurd. Taking a black photo, and jacking the exposure up 7 stops will make your image unusable. But turning up your ISO 7 stops to bring light and colour into the frame, will likely not. I'm happy to be proven wrong here, because it means I can just set the ISO to 400 and worry about whether there's anything in my actual photos later in post.
The one simple fact errodes the entirety of the Northrup argument (and by association, FStoppers "tests"). Let's just all agree that unless we're becoming computational photographers, there's more here than the Northrups are qualified to even offer passing commentary on.
ISO isn't fake, it just doesn't mean the same thing it once did with film.
This reminds me, I have some ISO 12 film in my fridge I need to try.
Good man.
lol, youtubers.
btw, capture one noise reduction is phenomenal. for some reason looks better than adobe feature...
That's because it's raised to 50 by default.
lol :) i mean it looks better when used. also your "default" is not my "default".
Yes the drama yet again by our dear Toni Notorious.
Simply ignore it. He makes some really good videos, and yet he cannot resist to post with a stunning regularity some grand claims which simply only entice the irrational reactions. Not serving the hobby or profession, sorry. I could make an entire list of these controversial videos, but ... wait, I will rather go out and photograph some interesting things. If you believe that ISO is a total fake, I have some swamp land in Florida for a sell to you.
There are crazy difference on A7III. So does not work for all cameras for sure.
Edited in C1, 4 stops +.
So wait, it's better to get it right in camera rather than fix it in post? This was SO worth all that time to find out... ;)
As a camera developer I know that Tony is wrong, in most claims in this video. Even if some of the statments by them selfs has merit the video in its entirety is misleading and over simplified. As mentioned the base sensitivity of a sensor does not need to correspond to a ISO of 100. It could be more or less and gain most be discussed separately in digital and analog. Its up to the manufacturer to handle the capabilities of the sensor where different manufacturers have different capabilities. Analog gain can for some sensors be set for each ISO step or even continuously where each approx 6dB is one stop of exposure. This is much preferable over direct digital gain since the then the noise floor is linear for digital gain but not for analog. So if the sensor have the the restraints of having either 0 analog gain or for example 24db analog gain (4 stops of exposure) its just dual gain sensor while many sensors might be able to set each 6 dB or even 1dB. If the sensor reach its maximum analog gain but the camera manufacturer wants to reach higher ISO most of them start doing digital gain or digital gain in between analog gain stops for higher ISO resolution. And then we have not talked about ADC oversampling.... What I am most allergic to is the universal claims he does with out sufficient data and saying one thing but the data are different settings and circumstances. FilmmakerIQ in general is very good but its limited to dual ISO but he does not make any claims otherwise.
Just out of curiosity since i dont own a D850, if i shot an image at ISO 100 6-7 stops under accidentally, and an image at ISO 1600 2-3 stops under and matched them, which would look better?
See that is the real test nobody seems to be doing. Yes we can see that ISO 100 had great flexibilty, but does ISO 1600? Becuse if it does not, then ISO in that camera is indeed irrelevant. But if it does, then open up those programs again Lee, you are not done.
FYI here is a trick in Lightroom to get to 10 stops exposure slider, Make a DNG HDR raw out of 3 images nearly identically exposed, on that HDR image Lightroom will allow you 10 stops in either direction.
If the camera is truly iso invariant then there is no difference in the iso 100 or iso 1600 file if the other camera settings are the same. The 1600 file is just “boosted”