ISO invariance, should I use low ISO?

snerkler

snerkler

Messages
23,617
Name
Toby
Edit My Images
No
More and more cameras seem to be ISO invariant these days meaning that (assuming aperture and shutter are fixed) you can underexpose your image and then brighten it in post with no penalty in noise compared to if you'd shot it at a higher ISO to get the exposure right in the first place, which then begs the question whether you'd be better doing this deliberately? The reason for my thoughts is dynamic range, if you can use 4-5 stops lower ISO that may make a noticeable difference in dynamic range and therefore yield better results after processing. I'm obviously talking about shooting in RAW.

Anyone else have any thoughts on this?
 
Wasn't there a thread on here about someone shooting at ISO100 for everything, then bringing it back in post.

I was always under the impression that increasing exposure increases noise (especially in the shadows) - but not sure about increasing the exposure to essentially get the same result.

IMO i don't mind shooting around 1 shot under exposed, but wouldn't really try lower.

Go do some testing @snerkler :D
 
There was a thread about leaving iso at 100 and adjusting in post. Not sure what the outcome of it was.
 
I think I'll give it a go and fine out, when we next have some decent weather that it :rolleyes: ;)
 
I'm not sure that there is no penalty. I think you can underexpose but if you go too far than shadows look absolutely horrible trying to bring it back in post processing if I remember.
 
snerkler said:
I think I'll give it a go and fine out, when we next have some decent weather that it :rolleyes: ;)
Click to expand...

Way to go (y)

There is some difference between theory and practise. While with a truly ISO-invariant sensor there would be no difference between raising ISO and brightening in post, in terms of overall exposure and shadow noise etc, the only real benefit to doing that is retaining highlight detail that might otherwise get blown. Alongside though, is the very real downside of potentially having a very dark LCD image to work with.

In practise, from my limited experience with ISO-invariant cameras, some compromise is the best way - basically raising ISO as high as necessary to keep the blinkies under control on bright highlights. That way you keep maximum dynamic range benefit, yet hopefully still retain a decent (if darker) LCD image that you can use to make a good assessment of other important image aspects.

Edit: basically, you still need to know where your highlight exposure is, and blinkies are the best way of doing that. If blinkies are not flashing anywhere, it's hard to know if you're not actually under-exposing and therefore increasing shadow noise more than necessary.
 
Last edited:
Just try it and see, apart from your time it cost nothing to experiment.
 
OK so my very remedial tests show that the D850 isn't truly ISO invariant, but it's not bad. Here's the results (I tried to get a lot of colour, as well as black into the scene)

Properly exposed at ISO 100

DSC_0103 by TDG-77, on Flickr

Underexposed by 5.5 stops

DSC_0107 by TDG-77, on Flickr

Recovered in post (LR only allows +5EV exposure bump so also brushed in a further 0.5ev)

DSC_0107-2 by TDG-77, on Flickr

Colours don't look bad imo. A bit of a colour shift but if you didn't have something to compare it to you wouldn't know ;)

And here's 1:1 crops to show that noise is quite apparent with such an exposure bump in post

Properly exposed

Screen Shot 2018-04-11 at 11.06.23 by TDG-77, on Flickr

5.5 stops underexposed and recovered

Screen Shot 2018-04-11 at 11.06.15 by TDG-77, on Flickr
 
I also tested the difference in 14 bit vs 12 bit and compressed vs uncompressed and tbh I couldn't really see a difference in any, maybe a tad in 12 bit compressed but not enough to be certain. I'm sure there are scenes where you will see a difference, but for everyday it does make me wonder whether I'd see a difference.

Following sequence is 14 bit lossless > 14 bit compressed > 12 bit lossless > 12 bit compressed

DSC_0103 by TDG-77, on Flickr

DSC_0104 by TDG-77, on Flickr

DSC_0105 by TDG-77, on Flickr

DSC_0106 by TDG-77, on Flickr



As I say, these are very unscientific but I was just curious and it's helped my curiosity ;)

Edit: File sizes for the above are:-
14 bit Lossless 53mb
14 bit compressed 43.8mb
12 bit lossless 41.5mb
12 bit compressed 34.8mb
 
Last edited:
(y)

So after doing this and your other research, what's your conclusion on this ISO-invariance thing?
 
Great posts @snerkler :)

Me, I would always expose to protect the highlights or bracket if necessary (with m43 cameras bracketing is sooooo easy handheld it just makes sense to do it especially when LR makes it soooo easy to merge them too).
 
I've found color critical shifts occur even with ETTR (collecting more light and adjusting in post). It seems that due to the CFA (bayer filter) and demosaicing algorithms when you deviate from a "normal" exposure you get these color shifts, and the farther you deviate, the worse they are.

I use ISO invariance pretty much the same way as with any other exposure... set the EC to retain the desired highlights and let the exposure be under. Since there is no benefit to staying at base ISO, and because there are numerous negatives, I only underexpose by the necessary offset.
 
sk66 said:
I've found color critical shifts occur even with ETTR (collecting more light and adjusting in post). It seems that due to the CFA (bayer filter) and demosaicing algorithms when you deviate from a "normal" exposure you get these color shifts, and the farther you deviate, the worse they are.

I use ISO invariance pretty much the same way as with any other exposure... set the EC to retain the desired highlights and let the exposure be under. Since there is no benefit to staying at base ISO, and because there are numerous negatives, I only underexpose by the necessary offset.
Click to expand...

'Critical color shifts'? You're basically saying that exposure level changes colours?! Care to elaborate?
 
HoppyUK said:
'Critical color shifts'? You're basically saying that exposure level changes colours?! Care to elaborate?
Click to expand...
Yes, but I have to say that different cameras show this effect to different degrees... it looks like the D850 with its massive oversampling/fine pixel pitch behaves better (minimal shift even with massive exposure changes).

Colors in a digital image are varying levels of RGB filtered electrons collected, but it's not linear like a simple exposure shift would be. This is due (I believe) to the CFA, photon shot noise/randomness, and demosaicing algorithms... i.e. if shot noise and the CFA prevents a pixel from collecting the same amount/ratio of blue spectrum (filtered) light, then the demosaicing/color calculation of that pixel and surrounding pixels will shift. And since it is not linear, a simple global WB adjustment cannot correct it... it can correct one color, but it will also cause a greater error in other colors.

I tested this with my Nikon1 and D800 using a color checker and a +2 ETTR shift.
 
Last edited:
HoppyUK said:
(y)

So after doing this and your other research, what's your conclusion on this ISO-invariance thing?
Click to expand...
That it's not truly ISO invariant and that you're better off exposing correctly, or just under to protect highlights. I wondered whether the higher DR at base might make the final image better, but clearly there's a colour shift and noise. I don't know what it would be like 2 or 3 stops under and then rescued, but I've got all the answers that I need without having to do that, I shall continue to expose as I always have (y)

Nawty said:
Great posts @snerkler :)

Me, I would always expose to protect the highlights or bracket if necessary (with m43 cameras bracketing is sooooo easy handheld it just makes sense to do it especially when LR makes it soooo easy to merge them too).
Click to expand...
It depends on the scene to how much I'd be willing to protect highlights. If I have to underexpose 4 or 5 stops to protect highlights then that might be too much. Plus sometimes if you protect the highlights too much and lift the shadows a lot it can start to look a little unnatural to my eyes. YMMV.

sk66 said:
I've found color critical shifts occur even with ETTR (collecting more light and adjusting in post). It seems that due to the CFA (bayer filter) and demosaicing algorithms when you deviate from a "normal" exposure you get these color shifts, and the farther you deviate, the worse they are.

I use ISO invariance pretty much the same way as with any other exposure... set the EC to retain the desired highlights and let the exposure be under. Since there is no benefit to staying at base ISO, and because there are numerous negatives, I only underexpose by the necessary offset.
Click to expand...
Pretty much what I do (within reason) (y)
 
snerkler said:
Anyone else have any thoughts on this?
Click to expand...

I don't know how invariant my cameras are but I doubt it'd change what I do. For right or wrong I ETTR if I can and possibly back it off post capture. I think the advantage of invariance for me would not be in shooting at lower ISO's and boosting the image post capture, I think I'd continue to ETTR and maybe boost the shadows with more confidence. I think boosting the shadows post capture and getting better results would be the advantage for me.
 
The way that I look at it is (from a landscape photography point of view), it just doesn't sound like something I'd want to do!

For example, shooting at 100 ISO, boosting 5 stops means your shadows are at 3200 ISO. The lower the ISO, the better the dynamic range, so the better the image quality, therefore better print quality. This is why I will always shoot my landscapes at base ISO with ND grads and should I forget them, bracket and merge in post.

Obviously it depends what you shoot / how you work etc. but just because you can, doesn't mean you should!
 
Unlocker said:
The way that I look at it is (from a landscape photography point of view), it just doesn't sound like something I'd want to do!

For example, shooting at 100 ISO, boosting 5 stops means your shadows are at 3200 ISO. The lower the ISO, the better the dynamic range, so the better the image quality, therefore better print quality. This is why I will always shoot my landscapes at base ISO with ND grads and should I forget them, bracket and merge in post.

Obviously it depends what you shoot / how you work etc. but just because you can, doesn't mean you should!
Click to expand...
Yep, 99% of landscapes will always be at base ISO (y)
 
Unlocker said:
The lower the ISO, the better the dynamic range, so the better the image quality, therefore better print quality.
Click to expand...
*When at least some pixels are filled to capacity*.

I felt the clarification necessary because just saying DR is always better at min ISO leads to incorrect assumptions in terms of ISO invariance. I.e. in the underexposed example above, less DR was actually recorded.
 
Last edited:
At the time of the earlier thread about ISO invariant sensors/cameras I searched for any ref to that characteristic on the Canon 5D3

If I recall on the DP review of this body it mentioned as being ISO invariant but from a minimum ISO of 400

So gives me pause to wonder how to utilise this and under what conditions it might be beneficial.......as ISO400 is for most situations noise free.
 
Box Brownie said:
At the time of the earlier thread about ISO invariant sensors/cameras I searched for any ref to that characteristic on the Canon 5D3

If I recall on the DP review of this body it mentioned as being ISO invariant but from a minimum ISO of 400

So gives me pause to wonder how to utilise this and under what conditions it might be beneficial.......as ISO400 is for most situations noise free.
Click to expand...
I’ve never seen the 5D3 being mentioned as invariant tbh, everything I’ve seen says not. You get a lot of purple colour shift, noise and banding when the 5D3 files are pushed hard.
 
snerkler said:
I’ve never seen the 5D3 being mentioned as invariant tbh, everything I’ve seen says not. You get a lot of purple colour shift, noise and banding when the 5D3 files are pushed hard.
Click to expand...

On smartphone now, will try to find the page and link it for others to see what can be made of it???
 
I've read that for astro shots setting iso to 400 on the D850 works well.... you get the benefits of the wider dynamic range and can decide how much to push the exposure in post rather than set iso at say 6400 and then have to deal with the noise... not done much astro but it's on my list this year so will certainly be trying it

Simon
 
Ah! it was the 5D4 not the 5D3 that I read what I mentioned in post #23

On this page https://www.dpreview.com/news/32297...c-dynamic-range-improvements-to-the-5d-line/2 here the relevant part

The 5D Mark IV isn't entirely ISO-invariant: pushing an ISO 200 underexposed by 5 stops by 5 EV in post-processing yields slightly higher noise levels than a native ISO 6400 exposure. An ISO 100 exposure pushed 6 stops fares even worse. However, above ISO 400, the camera does, for the most part, exhibit ISO invariance, meaning that you could underexpose a traditional ISO 6400 exposure by 4 EV by shooting it at ISO 400 (while maintaining the shutter speed and aperture for ISO 6400), and then raise exposure 4 EV in post. This technique would afford you 4 EV of highlight headroom, with little to no noise cost, relative to shooting at ISO 6400. This is precisely the technique I used to shoot the galloping horse on the first page.

Compared to its predecessor, the 5D Mark III, this is a massive improvement. With the Mark III, one could not simply underexpose to protect highlights without paying a significant noise cost when boosting shadows in post, meaning you had to make your choices about which tones you wished to presere on the spot, at the time of the exposure. Even compared to the 5DS, the Mark IV shows vast improvements. It even slightly edges out the 1D X II. In a nutshell, the 5D Mark IV will be far more capable at dealing with high contrast scenes without the need for workarounds.

oops! :(
 
That DP review must be wrong. You've already beeen told that Canons DR is rubbish.
 
sk66 said:
*When at least some pixels are filled to capacity*.

I felt the clarification necessary because just saying DR is always better at min ISO leads to incorrect assumptions in terms of ISO invariance. I.e. in the underexposed example above, less DR was actually recorded.
Click to expand...

Understood, that's why I mentioned shooting with grads to get a better balanced exposure in camera, less to do in post. Obviously not so easy to do if not shooting landscape!
 
Hi Toby @snerkler , great to see you are doing some empirical testing of your own to not just understand theory but to test it's effects.

Certainly not a criticism, but maybe you have missed the point of ISO invariance slightly here. All you have done is taken a picture at ISO100 and pushed it 5 stops or so and compared it to one also taken at ISO100 but exposed correctly in camera, correct?

To really test ISO invariance what you should be doing is shooting the scene correctly exposed at ISO800 for example, then (in manual) without changing the shutter speed or aperture drop the ISO to 100. Naturally this will result in an image 3 stops darker than the ISO800 one. Then, in your RAW converter push the ISO100 image by 3 stops to compare with the in camera pushed ISO800 one. With an invariant sensor there will be no loss of quality.

That is what they mean by ISO invariant, or ISO less sensors. For me the advantage is if you allow the camera to push the exposure using ISO it does it broadly across all of the tones, if you do it in RAW conversion you can apply finesse and pull the white point back. This gives you more headroom of the highlights and more dynamic range to play with. But, you get a darker JPEG thumbnail preview image, for relatively small gains in IQ. Although if you accidentally blew the highlights at ISO800 you would have been grateful for the underexposed ISO100 shot that you can push from the comfort of your computer in a controlled manor!
 
Craig_85 said:
Hi Toby @snerkler , great to see you are doing some empirical testing of your own to not just understand theory but to test it's effects.

Certainly not a criticism, but maybe you have missed the point of ISO invariance slightly here. All you have done is taken a picture at ISO100 and pushed it 5 stops or so and compared it to one also taken at ISO100 but exposed correctly in camera, correct?

To really test ISO invariance what you should be doing is shooting the scene correctly exposed at ISO800 for example, then (in manual) without changing the shutter speed or aperture drop the ISO to 100. Naturally this will result in an image 3 stops darker than the ISO800 one. Then, in your RAW converter push the ISO100 image by 3 stops to compare with the in camera pushed ISO800 one. With an invariant sensor there will be no loss of quality.

That is what they mean by ISO invariant, or ISO less sensors. For me the advantage is if you allow the camera to push the exposure using ISO it does it broadly across all of the tones, if you do it in RAW conversion you can apply finesse and pull the white point back. This gives you more headroom of the highlights and more dynamic range to play with. But, you get a darker JPEG thumbnail preview image, for relatively small gains in IQ. Although if you accidentally blew the highlights at ISO800 you would have been grateful for the underexposed ISO100 shot that you can push from the comfort of your computer in a controlled manor!
Click to expand...
Yes I see your point, made me realise I was a bit of a muppet as all I was doing was assessing recovery from an underexposed shot :facepalm: I may do some further tests if I get another rainy day on my day off. I really want to push the 12 bit stuff as well as I’m considering using that for sport to keep file size down. Shooting at 9fps it’ll not take long to fill a card at 14 bit lossless. My 14 bit lossless files have ranged from 51-62mb.
 
snerkler said:
Yes I see your point, made me realise I was a bit of a muppet as all I was doing was assessing recovery from an underexposed shot :facepalm: I may do some further tests if I get another rainy day on my day off. I really want to push the 12 bit stuff as well as I’m considering using that for sport to keep file size down. Shooting at 9fps it’ll not take long to fill a card at 14 bit lossless. My 14 bit lossless files have ranged from 51-62mb.
Click to expand...

Toby, you'll never be happy with the compromise of 12 bit data, when you will have that devil on your shoulder says 'do you think the 14 bit would have been better', you strive for perfection, its just in your nature. Just buy extra or bigger capacity cards, you know its the right thing to do :)
 
Mr Perceptive said:
Toby, you'll never be happy with the compromise of 12 bit data, when you will have that devil on your shoulder says 'do you think the 14 bit would have been better', you strive for perfection, its just in your nature. Just buy extra or bigger capacity cards, you know its the right thing to do :)
Click to expand...
Lol, yes you're right for most things. However, the reason for me looking into 12 bit is that I'm shooting the London Marathon next weekend for Arthritis Research and the photos are only used for the web and small leaflets and from my preliminary tests you're just not going to see the difference. I've even thought about just shooting jpeg tbh.
 
snerkler said:
Lol, yes you're right for most things. However, the reason for me looking into 12 bit is that I'm shooting the London Marathon next weekend for Arthritis Research and the photos are only used for the web and small leaflets and from my preliminary tests you're just not going to see the difference. I've even thought about just shooting jpeg tbh.
Click to expand...

Shooting 12 bit will give you a bigger buffer too, that might clinch it.
 
Unlocker said:
Shooting 12 bit will give you a bigger buffer too, that might clinch it.
Click to expand...
TBH that was another reason. Although to be honest 51 shots is overkill for me. Last year I didn't run into the buffer limit on the D750 too often, only 2 or 3 times IIRC. That being said, I'll be shooting at 9fps this year rather than 6.5fps so maybe I will need all that buffer after all ;)
 
You must log in or register to reply here.
Back
Top