Learning to not be afraid of high ISO shooting

[kimball]

100% with you as well on this. It's an entire system, not just the sensor. Although, with newer sensors, just about everything takes place within their circuitry, including analog and I think even digital amplification.

Yup, it's my layperson's understanding that's how it's implemented in about 99% of cameras.

Fuji did something interesting with their GFX 50 series medium format cameras though. Above ISO 1600 they stopped implementing ISO changes in hardware. Raw files shot above ISO 1600 had the exact same data as if they were shot at ISO 1600, along with a metadata tag instructing the raw processor to lighten the image by however many stops above 1600 the shot was taken at.

This sounds like fakery, but the benefit was no loss of camera DR as one increased their ISO beyond 1600. And because the sensor was also invariant in that range, there was no additional noise consequence beyond the usual increase in visible shot noise that comes with lower and lower light.

This is hopefully the future of how ISO will be implemented. Have a couple of hardware-based ISO values for the cases where hardware can actually reduce read noise, and then implement everything else in camera firmware, or in the case of raw files, in post processing software.

Raw video is already headed in that direction. Davinci Resolve even has an interface for changing ISO of raw video in post, after it was shot. This of course completely breaks the notion that ISO is a property of the camera, but it does have interesting and useful applications. Whether the ISO standard gets on board with this is another matter. And of course Lightroom has always had an Exposure slider, but that doesn't mean the slider actually changes the sensor exposure after the fact.

 

[Nimi]

Yup, it's my layperson's understanding that's how it's implemented in about 99% of cameras.

Fuji did something interesting with their GFX 50 series medium format cameras though. Above ISO 1600 they stopped implementing ISO changes in hardware. Raw files shot above ISO 1600 had the exact same data as if they were shot at ISO 1600, along with a metadata tag instructing the raw processor to lighten the image by however many stops above 1600 the shot was taken at.

This sounds like fakery, but the benefit was no loss of camera DR as one increased their ISO beyond 1600. And because the sensor was also invariant in that range, there was no additional noise consequence beyond the usual increase in visible shot noise that comes with lower and lower light.

This is hopefully the future of how ISO will be implemented. Have a couple of hardware-based ISO values for the cases where hardware can actually reduce read noise, and then implement everything else in camera firmware, or in the case of raw files, in post processing software.

Raw video is already headed in that direction. Davinci Resolve even has an interface for changing ISO of raw video in post, after it was shot. This of course completely breaks the notion that ISO is a property of the camera, but it does have interesting and useful applications. Whether the ISO standard gets on board with this is another matter. And of course Lightroom has always had an Exposure slider, but that doesn't mean the slider actually changes the sensor exposure after the fact.

Excellent description.

This discussion is even more important for video, where native ISOs are higher and shutter speed is very low. That's why we shoot Log and rely heavily on ND filters. The trend is for higher and higher DR, with cameras like Red V Raptor advertising 17 stops. That maximizes editability and grading in post.

 

[Steve]

Yup, it's my layperson's understanding that's how it's implemented in about 99% of cameras.

Fuji did something interesting with their GFX 50 series medium format cameras though. Above ISO 1600 they stopped implementing ISO changes in hardware. Raw files shot above ISO 1600 had the exact same data as if they were shot at ISO 1600, along with a metadata tag instructing the raw processor to lighten the image by however many stops above 1600 the shot was taken at.

This sounds like fakery, but the benefit was no loss of camera DR as one increased their ISO beyond 1600. And because the sensor was also invariant in that range, there was no additional noise consequence beyond the usual increase in visible shot noise that comes with lower and lower light.

This is hopefully the future of how ISO will be implemented. Have a couple of hardware-based ISO values for the cases where hardware can actually reduce read noise, and then implement everything else in camera firmware, or in the case of raw files, in post processing software.

Raw video is already headed in that direction. Davinci Resolve even has an interface for changing ISO of raw video in post, after it was shot. This of course completely breaks the notion that ISO is a property of the camera, but it does have interesting and useful applications. Whether the ISO standard gets on board with this is another matter. And of course Lightroom has always had an Exposure slider, but that doesn't mean the slider actually changes the sensor exposure after the fact.

That makes a lot of sense. I agree - I think as read noise drops incremental lower, how Fuji is doing it will likely become more common.

 

[kimball]

Excellent description.

This discussion is even more important for video, where native ISOs are higher and shutter speed is very low. That's why we shoot Log and rely heavily on ND filters. The trend is for higher and higher DR, with cameras like Red V Raptor advertising 17 stops. That maximizes editability and grading in post.

Thanks. Wow, 17 stops of DR is insane! I'm guessing that's Engineering DR, where the noise floor is defined as SNR=1. I'd love to see that on some high-MP photo-centric cameras. I'm assuming it's achieved through a higher FWC (Full Well Capacity, the maximum lightness value). If so, that would lead to insanely low photographic ISOs. The Z7 with a base ISO of 64 has a max EDR of about 13.5. If it were 17, its base ISO would likely be 8 or less. Wild!

 

[John Woodworth]

...
Visible noise in photographs has two sources: 1) noise coming from the light itself ("shot noise"), and 2) noise introduced by the camera ("read noise").
...

Agreed. Plus, 3) Digital quantization noise. A 14-bit camera can't have much better than 14 stops of dynamic range.

Modern cameras have insanely low read noise. So low that, mathematically speaking, there's really not much more read noise improvement that can be squeezed out of cameras, at least not that will make a huge visual difference. In the past 5-10 years they've really come quite close to, asymptotically so, "as good as it's going to get."

Modern sensors do have electronic noise, but it can be subtracted out by using the masked pixels at the edge of the sensor. Part of the "secret sauce".

Unfortunately, nothing can be done about #1, shot noise in low light situations...No sensor tech can reduce shot noise in a photograph. High ISO does not cause shot noise. It merely correlates with it, because shot noise becomes more visible in photographs in low light shooting, where we are all much more likely to use a high ISO.

If you want to avoid capturing visible shot noise in your photos, you have to increase the amount of light hitting your sensor (i.e. increase exposure, in the formal sense of the word). Shoot with a wider aperture, and/or shoot with a slower shutter speed, and/or increase scene luminance. If you are at your widest aperture and slowest shutter speed you can tolerate, and can't increase your scene luminance, you've done all you can. Choosing a lower ISO will not reduce visible noise in your photo, and will actually increase it if your sensor is not ISO-invariant.

One way, maybe the only way, to increase dynamic range is to scale cameras up in size so they capture more light. That's why Full Frame cameras beat cell phones in low-lit night clubs.

Yes, you are right. The more photons you capture, proportionately lower is the statistical noise, and the better the ratio of signal to noise in the final image. Just don't blow the highlights.

 

[Nimi]

Agreed. Plus, 3) Digital quantization noise. A 14-bit camera can't have much better than 14 stops of dynamic range.

The Red V Raptor is full-frame, 8k, 16-bit, adeverised at 17+ stops DR. I should mention there is no standardized DR measurement for video. Cine-D has a system, which rates this camera at 13-14, and the Arri Alexa 35 at 16+. Regardless, this is where money is being invested and we'll see 20 stops in our lifetime.

 

[kimball]

Agreed. Plus, 3) Digital quantization noise. A 14-bit camera can't have much better than 14 stops of dynamic range.

Yeah, I have a bad habit of calling all camera-induced noise "read noise." There's also pattern noise, thermal noise, and PRNU. But the gist of what I was saying is the majority of visible noise comes from outside the camera when shooting in low light.

One way, maybe the only way, to increase dynamic range is to scale cameras up in size so they capture more light. That's why Full Frame cameras beat cell phones in low-lit night clubs.

DR and noise are two related but separate concepts (see my annoyingly long post about DR earlier in this thread).

Decreasing visible shot noise can be achieved by collecting more light, thus improving SNR. Larger sensors do this automatically (collect more light) given the same exposure as a small sensor.

But increasing DR doesn't necessarily need larger sensors. Technology that imcreases FWC--maximum recordable lightness-- while keeping the noise floor low will increase DR, independent of sensor size.

 

[kimball]

The Red V Raptor is full-frame, 8k, 16-bit, adeverised at 17+ stops DR. I should mention there is no standardized DR measurement for video. Cine-D has a system, which rates this camera at 13-14, and the Arri Alexa 35 at 16+. Regardless, this is where money is being invested and we'll see 20 stops in our lifetime.

Yes, storing N+C stops of DR in values N bits wide without gamma compression is legitimately possible, even though it doesn't sound like it. Apparently it works out when taking into account read noise. See Jack Hogan's blog posts here, especially the 2nd one:

https://www.strollswithmydog.com/dynamic-range-and-bit-depth/
https://www.strollswithmydog.com/how-many-bits-to-fully-encode-my-image/

 

[frdjohns]

I have taken images that I have thought would be totally unusable as the ISO was high.
Shot when the bird moved into the shade and I was not quick enough to change settings to suit the new subject environment.

Now with products such as the Topaz offerings I have found that high ISO does not in itself make images unusable.
A screenshot of a poor quality image loaded in Topaz Photo AI (Default Settings) and the preview of what it did with this image.
1/1600, ISO 25600, F 8.0, EC -.3
View attachment 48939

That's pretty impressive.

 

[guydagar]

Thanks for that. I would start to get antsy if ISO hit 1000. Not anymore though!

I totally relate. Coming from the film era when ASA/ISO 800 was the fastest color transparency film, Kodak Tri-X 400 was the fastest B&W film, and the resulting images were mostly too grainy, unless the graininess worked for a particular composition, which wasn't often. When I started shooting a DSLR, I was completely ignorant about sensors and how the cameras really worked—I just thought the sensor replaced film. And being digital, I also thought that ISO was just a measure of sensitivity, because after all it wasn't film so why would it be grainy? Well, shooting with an APS-C Canon camera for a while I discovered that ISO 800 wasn't very good and I became completely ISO-shy, avoiding higher ISOs always. With my first full-frame Canon 6D I discovered I could shoot at ISO 800 and still get some good pics. But beyond 800 was still off-limits.

As time went by and I switched to full-frame Nikons and became more adept with post-editing software, I discovered I could improve a lot of those ISO 800 shots and even shoot as high as ISO 1000. Still high ISO-shy, though. When I bought first a D3s and then a D5, I quickly found, somewhat by accident, not paying attention to the ISO whilst shooting in manual mode with auto-ISO, that noise at ISO 12500, even 16000 could work the noise was easily corrected in post. Now, I don't intentionally shoot at high ISOs—I try to keep it reasonable; I shoot in manual mode with auto-ISO and generally choose my shutter speed and aperture and let the ISO fall where it may—I don't pay too much attention to it; shutter speed is more critical; I have auto-ISO set to a limit of 12500, most of the time.

But, to be fair, I have for many years concentrated on landscapes and seascapes, portraiture and products, and always some, but not so much on wildlife and birds-in-flight in, especially in low-light. I always tried to stay at base ISO on a tripod, which is primarily why I upgraded from a D800E to a D810, for that ISO 64, lack of noise and wide dynamic range.

Now, higher ISOs have opened up a whole new world, even with my Z7 and especially with my Z6 II. But my D6 reigns supreme for those really high ISOs—with it I have no fear of high ISO. The camera/sensor makes a difference. And I agree, Topaz Photo AI is a real help. As is Lightroom Enhance, or DxO PureRAW, or On1 NoNoise—they all work extremely well. Of course, you need to start with proper exposure and an image in focus. Software is not magic. Or so it seems.

capephoto.net

 

[lwan]

Maybe its just me but DxO Photolab 6 seems yet even better at noise reduction. I prefer to "shoot first and ask questions later" i.e. don't worry about ISO.

Same here. If the shot is not recoverable, so be it. At least it was attempted. But with a max auto ISO set to 6400 or even 12800 on 20-24mpx cameras, it has become rare to be unable to get something worthy even if that implies some downsizing.

 

[bjanes]

This is generally the correct advice. Let the ISO fall where it falls once you can't increase sensor exposure anymore.

However, if you wish to protect highlights in an otherwise low-light scene, there's very good reason to shoot at a much lower ISO and then raise your image lightness in post (assuming you're employing an ISO-invariant sensor). There will be no noise penalty in doing so.

Interestingly, in non-ISO-invariant sensors, you actually pay a noise penalty by shooting at lower ISOs and lightening the photo in post. Electronic read noise is higher in such sensors at lower ISOs. As ISO is raised, read noise is reduced. This is shown quite clearly in Bill Claff's "Read Noise vs ISO Setting" chart, which in this example shows read noise on the Nikon D6 dropping dramatically as ISO is increased, whereas on the D7000, which has an invariant sensor, read noise is practically constant regardless of ISO:

Input-referred Read Noise versus ISO Setting

Kimbal makes some excellent points. In low light settings, it is necessary to use higher ISO but it is most important to give the maximum amount of exposure (shutter speed and aperture with existing illumination) while protecting the highlights. The histograms and blinkies are helpful in avoiding clipping but they are conservative on most cameras and looking at the raw histogram in Rawdigger often shows that giving 1/2 or even 1 EV of additional exposure could have been given while preserving the highlights. Since these camera warnings are derived from the jPEG preview embedded in the raw file, it is useful to use a flat or neutral picture control and know the characteristics of your sensor from prior testing.

The Nikon Z9 is ISO invariant above ISO 500 (and most modern cameras become ISO invariant beyond a certain ISO). As Kimbal states you can protect the highlights by using the exosure settings (shutter speed and aperture) determined at a higher ISO and use these settings while setting the camera at a lower ISO to take the shot and brighten in post. Indeed this is what the Fuji GFX 50x does; above ISO1250 it stops applying gain and records the boost to be applied to brighteng of the image in post. See here for posts by heavy weights Jim Kasson and Bill Claff that document this.

Much is made about read noise at high ISO, but with the Z9 and many other high end sensors the input referred read noise is hardly above one electron at high ISO (see Bill Claff). You can't reduce read noise any further. You can pretty well characterize the noise perfformance of a sensor using the sensor full well electrons, clipping limit in electrons at higher ISO and the input referred noise (see Roger Clark).

I have done this for the Z9 using data from Jim and Bill Claff in the aticle referenced above. Total
noise is the sum in quadrature of the shot noise and read noise (sqrt [ShotN^2 + ReadN^2]).

You can only see EXIF info for this image if you are logged in.

The Z9 is below state of the art below ISO 500 (probably because of limitations imposed by the stacked sensor, but is state of the art above high gain at ISO 500 and above. Shot noise is predominat at all ISOs and exposure is needed to maintain a high SNR.

 

[DavidT]

I often have to shoot rodeo or sports north of 20K and with the a1 and topaz I have no issue unless I have to crop a lot which fortunately I don't as I have the correct glass for the job. Modern camera sensors and software are simply amazing.

 

[bjanes]

Not at all. Most modern raw developers employ highlight protection. -2 EC by way of reduced ISO will have 2 more stops of captured highlights (assuming such highlights exist in the scene), and those highlights will remain available in raw processors. a +2 stop equivalent adjustment in a raw processor does not mean those highlights will be necessarily pushed into clipping.

As I indicated in my original post, you have the option but no obligation to brighten the image if that would clip highlights, but there are better ways to protect the highlights than merely dialing in negative exposure.

Bill

 

[fcotterill]

As I indicated in my original post, you have the option but no obligation to brighten the image if that would clip highlights, but there are better ways to protect the highlights than merely dialing in negative exposure.

Bill

As I understand your interesting post #62, the picture control generated as the jpeg preview "from" the raw file, can indeed give one a fairly reliable sketch of blown vs conserved highlights? EDIT to add: I've always been skeptical of the reliability of the histogram along the 'right-edge' of the chart, particularly images in which the red tones have more of a tendency to blow out in some scenes.

So is it worth while comparing a diversity of picture controls? Or does the general rule apply to use a flat aka neutral picture control? This being with the Z9 and D6 sensors, thanks?

 

[marc]

A great discussion!

I quickly learnt to embrace High ISO with the release of the D3 back in 2007, the phrase "game changer" gets tossed about far too often in the photography world these days,
but back then that phrase really meant something, pushing images beyond 6400 really became possible, and since then right up to my (now sold) D5 levels with full frame wildlife portraits could now be pushed to 25600.
Here's one example I captured with my D5 & Nikkor 400 f/2.8E FL @ ISO 25600 1/400s f/2.8 F/Frame

A young sub adult male Leopard watching his soon to be dinner of a young Red Letchwe in the Okavango, Botswana.

You can only see EXIF info for this image if you are logged in.

 

[fcotterill]

Agreed, invaluable information has accumulated in this thread. The D4 sensor was my first leap/appreciation of the low light abilities of full frame Nikon cameras. This was in a Df I bought Used as Used prices of a D4 was beyond my reach at the time. The Df is surprisingly capable for wildlife (but perhaps it helps to have started off with a FM2 years earlier).

Post 2020, Used D5 prices had dropped sufficiently, and what a solid investment; although the D6 is a significantly advance and well worth keeping close.

I also have lovely images off a Z6 sensor in lowlight, but the D5 (and D6) sensors are certainly in their own league
A lioness with whom I crossed paths in June, in a pride hunting before sunrise on an overcast morning north of the Olifants River, KNP. Nikon D5, 180-400 f4E TC14 @560mm, 1/640, f5.6, ISO30000. Moderate vs Minimal processing in LR, both 1:1 crop; so no use of the superior noise editing apps

You can only see EXIF info for this image if you are logged in.

You can only see EXIF info for this image if you are logged in.

 

[bleirer]

As I understand your interesting post #62, the picture control generated as the jpeg preview "from" the raw file, can indeed give one a fairly reliable sketch of blown vs conserved highlights? EDIT to add: I've always been skeptical of the reliability of the histogram along the 'right-edge' of the chart, particularly images in which the red tones have more of a tendency to blow out in some scenes.

So is it worth while comparing a diversity of picture controls? Or does the general rule apply to use a flat aka neutral picture control? This being with the Z9 and D6 sensors, thanks?

I'm not that Bill but from my testing (and what Bill said) with rawdigger on my camera (Canon) with the camera set to faithful when I find the highest non-blinkie exposure I still can add 2/3 of a stop and be confident. After adding the 2/3 enough times I got used to how the blinkies looked at that exposure and could eyeball it without having to find the non-blinkie exposure every time. But at first I set the camera to give a hold preview to take some test shots. On my camera and I assume Nikon is the same, I don't get the preview until I release the bbf, so it doesn't interfere with shooting bursts.

Bottom line is experiment with consistent settings to get to know the gear.

 

[John Navitsky]

I'm not that Bill but from my testing (and what Bill said) with rawdigger on my camera (Canon) with the canera set to faithful when I find the highest non-blinkie exposure I still can add 2/3 of a stop and be confident. After adding the 2/3 enough times I got used to how the blinkies looked at that exposure and could eyeball it without having to find the non-blinkie exposure every time. But at first I set the camera to give a hold preview to take some test shots. On my camera and I assume Nikon is the same, I don't get the preview until I release the bbf, so it doesn't interfere with shooting bursts.

Bottom line is experiment with consistent settings to get to know the gear.

agree. i liked this video by Dave Morrow where he points out learning about your gear to find those limits:

 

[bjanes]

As I understand your interesting post #62, the picture control generated as the jpeg preview "from" the raw file, can indeed give one a fairly reliable sketch of blown vs conserved highlights? EDIT to add: I've always been skeptical of the reliability of the histogram along the 'right-edge' of the chart, particularly images in which the red tones have more of a tendency to blow out in some scenes.

So is it worth while comparing a diversity of picture controls? Or does the general rule apply to use a flat aka neutral picture control? This being with the Z9 and D6 sensors, thanks?

IMO picture controls are mainly for JPEG work. If you shoot raw, a flat of neutral picture control may give a better starting point for giving a more accurate camera histogram that you use to evaluate exposure. You can replicate the effects of the picture control in post. ACR/LRC have camera matching profiles for this purpose.

To investigate this, I took shots looking from the front door of my house using the Z9 with the flat, neutral, standard, vivid, and landscape picture controls using the aperture priority mode. Exposure was the same for all shots (1/20 s, f'/8). The raw histograms were the same for all shots and a representative raw histogram is shown below. The raw channels were nearly two stops below clipping and this could be a subject for further discussion.

For processing in LR, +1 EV with the exposure slider gave an image with good brightness. I don't have images of the camera histograms but the embedded histogram in FastRawViewer can show the histogram of the embedded JPG preview imbedded in the raw file which should be the same as the camera histogram.

Here is a composite of the flat image with the raw histogram on the left and the embedded histogram on the right. The raw histogram is similar to that of raw digger and the embedded histogram is well below clipping.

You can only see EXIF info for this image if you are logged in.

And here is a composite with the standard picture control image. The embedded RGB histogram is near clipping.

You can only see EXIF info for this image if you are logged in.

Cheers,

Bill

 

[fcotterill]

Interesting. Thanks for sharing your test examples. I must experiment.
Up to recently, I'd ignored Picture controls, as confined to 100% RAW files; and I was arguing with a Nikon guy about this some months back who advised exploiting picture control settings for RAW as well as jpg. The embedded histogram was not the justification in this case.

 

[Tom Reynolds]

I forgot to drop the shutter speed from 1/3200 since I was hoping for an Acorn Woodpecker in flight. ISO 8000

You can only see EXIF info for this image if you are logged in.

But I did ultimately get one in flight.

You can only see EXIF info for this image if you are logged in.

 

[Stephen Berger]

Thanks so much everyone for the input and the wealth of info. I can't say I understand all of the technical nuances but for sure the gist is, shoot away at high ISO's!

This female belted kingfisher is from this morning at 5000 ISO (and then run through Topaz's 'low light' setting). Not so high by today's standards but something I wouldn't have tried two weeks ago...

You can only see EXIF info for this image if you are logged in.

 

[Spirit Vision Photography]

I have not had to deal with high ISO yet. But, which program or software is recommended to reduce ISO noise? I'm just learning Lightroom now.

 

[John Navitsky]

i like topaz denoise