How have high ISO bodies affected your habits

[jjj]

I've been a high-ISO guy since film days. Tri-X and HP5+ pushed to EI 1600, Delta 3200 used anywhere from EI 1250 to 12500, Fuji ISO 1600 color neg, etc. The earliest D-SLRs were actually a step down in this regard…the patterned noise above ISO 800 or so was pretty ugly, at least with the Canon 10D & 20D I used c. 2003–4.

I love film grain. I like non-patterened luminance noise too and frequently boost ISO to get more of it. Chroma noise I'm not so crazy about, but it can be easily mitigated. Not that you'd know anything about it from browsing LuLa forum topics, but there are plenty of folks who deliberately incorporate high levels of noise into their photographic styles. To me it's just the visual equivalent of overdriving a guitar amp.

-Dave-

I've mentioned doing similar myself, but the whole idea of wanting grain/noise seemed to really upset some people. 
Kodak Tri-X at 1600  and Kodak Recording film at 3200 in Acuspeed were combos I liked to use.

[Isaac]

but the whole idea of wanting grain/noise seemed to really upset some people.  

It's getting noise without wanting it that's upsetting to me, because I wanted the detail.

[mahleu]

For my media work, it's a game changer. It's nice to be able to go on a police raid at midnight into a very dark area and be able to shoot without flash, I have a flash on my second body for faster moving things.

[Isaac]

Question: I do not know much about your camera (a35)... is that one of those ISO-less cameras or...?

After looking through Guillermo Luijk's post and Dx0 Mark measurements (click "Dynamic Range"), I guess the answer is yes and no :-)

Between nominal ISO 800 through 6400 there's a 1 stop loss of DR for every extra stop of measured ISO -- so yes.

Between nominal ISO 100 through 800, there's less than 1 stop loss -- so no.

All other things being equal, I guess that means up to ISO 800 it's better to increase ISO.

Question #2: is there any reason to shoot 800+3 instead of 6400?

See Determining if a sensor is "ISOless"

[hjulenissen]

I wouldn't mind seeing one, but I do not care that much about a potential difference to do the test myself.

I have tried to do this test myself, but it is hard (for me) to get right. One would expect (from measurements, theory and claims from those who have studied the subject) that raising ISO has some advantages (shadow noise for Canon cameras) up to a point (e.g. ISO1600 or ISO3200), and some disadvantages (highlight clipping).

The wildcard is that your raw developer (Lightroom in my case) might not do +4Ev in the straightforward way.

-h

[hjulenissen]

I am now using auto ISO most of the time, allowing the camera to go up to 6400, and I've made quite a number of great shots at that speed. Portraits, cityscapes, both commercial and personal work. And mind you, I am using a lowly 60D, from a brand that is now a synonym for the worst sensor on the planet 

I believe that Canon receive critique for their DR at _low_ ISO. At high ISO settings, I believe that they are state-of-the-art.

-h

[Guillermo Luijk]

Do you have a comparison between ISO 6400 and ISO 800 +3EV in Lightroom?

I wouldn't mind seeing one, but I do not care that much about a potential difference to do the test myself.

The difference in the shadows will be none. The difference in the highlights when proper processing is applied can be up to 3 extra stops of highlight information (very interesting when strong light sources enter the frame such as the sun, lamps, reflections, shiny surfaces, fire,...).

In brief: ISO800+3EV in pp will always produce equal or better quality images than ISO6400 when shooting in RAW, never worse (same aperture and shutter is assumed).

[Isaac]

Thanks for making that clear.

[Slobodan Blagojevic]

The difference in the shadows will be none. The difference in the highlights when proper processing is applied can be up to 3 extra stops of highlight information (very interesting when strong light sources enter the frame such as the sun, lamps, reflections, shiny surfaces, fire,...).

In brief: ISO800+3EV in pp will always produce equal or better quality images than ISO6400 when shooting in RAW, never worse (same aperture and shutter is assumed).

Just to clarify: that holds true for ISO-less sensors only? I mean, my Canon files do not behave like that. Or, in other words, my Canon files at 6400 are visibly better in shadows and not worse in highlights than 800+3. Or, in the case of the attached example ISO 5000 vs. ISO 500 +3 ⅓.

[hjulenissen]

In brief: ISO800+3EV in pp will always produce equal or better quality images than ISO6400 when shooting in RAW, never worse (same aperture and shutter is assumed).

To add to Slobodans comment:
Your statements seems to rely on the camera being so-called "ISO-less" (or at least ISO-less in that particular range) AND your raw editor behaving in a certain way (lineary multiplying raw camera channels).

Practice trumps theory. If this works for you, then good. If not, then the theory must be restricted.

-h

[Guillermo Luijk]

Correct. Since most sensors behave ISOless from ISO1600-3200, that holds true for (nearly, the A7S seems a counterexample) any camera from ISO1600-3200.

I.e. ISO1600-3200 produces equal or better images than any superior ISO, never worse (assuming the same aperture and shutter).

Practice trumps theory. If this works for you, then good. If not, then the theory must be restricted.

It doesn't work for me since I mostly shoot RAW+JPEG and expose with JPEG in mind (well not in mind, in the EVF ).

[Jim Kasson]

Correct. Since most sensors behave ISOless from ISO1600-3200, that holds true for (nearly, the A7S seems a counterexample) any camera from ISO1600-3200.

The D810 is another counterexample, although this may be due to in-camera raw processing:



Test protocol: Image pairs, subtracted and divided by 1.414 for standard deviation, added and divided by 2 for mean. FWC and RN sum of squared error fitted by ISO, using Nelder-Mead downhill simplex. Points under SNR = 2 discarded.

Jim

[Slobodan Blagojevic]

Jim, what are we seeing in that graph? This time, English please

[Jim Kasson]

Jim, what are we seeing in that graph? This time, English please

It's the noise floor of a raw image (all 4 raw channels plotted) vs in-camera ISO setting. You can see that something happens between ISO 2500 and ISO 3200 that causes a discontinuity in the curve, with the background noise dropping suddenly at that point.

Therefore, if you're  going to boost the dark parts of the picture a lot in post, and you're trying to decide between setting the camera to ISO 2000, 2500 or 3200, you might want to pick 3200. Or not, depending on how Nikon is accomplishing this.

Is that clear? Is it at least in English?

Jim

[Slobodan Blagojevic]

I am confused. Whatever the graph represents, it seems to be the best at ISO 64, from which it gradually declines (getting worse) and then suddenly drops at 3200 (i.e., getting much worse). So, my layman conclusion is not to set the camera to 3200 or more, but to stick to up to 2500. That appears to be the opposite of what you are saying. I am sure I must be wrong, but where?

[Fine_Art]

The graph is noise. Less noise is better, yes?

Is that the Nikon black point chop off that annoys some astronomy types? I have read in some forums Nikon clips the black. I have also read that there is a firmware fix for people that want that data. Of course for regular photography the clipping is usually fine.

[Slobodan Blagojevic]

The graph is noise. Less noise is better, yes?..

Less is better, of course, but the graph seems to show that the worst noise is at ISO 64, which, again, seems counterintuitive.

[Jim Kasson]

Less is better, of course, but the graph seems to show that the worst noise is at ISO 64, which, again, seems counterintuitive.

This is noise measured at the sensor. If it would help, I can compute the curves for noise measured after the in-camera amplifier whose gain the ISO knob controls. That would be the noise that you'd see in the raw file. Would you like me to do that?

Jim

[Jim Kasson]

The graph is noise. Less noise is better, yes?

For most people, most of the time, yes.

Is that the Nikon black point chop off that annoys some astronomy types? I have read in some forums Nikon clips the black. I have also read that there is a firmware fix for people that want that data. Of course for regular photography the clipping is usually fine.

The D810 has an entirely different black point algorithm than its predecessors.

http://blog.kasson.com/?p=6783

My way of computing the read noise gets around the in-camera black point fiddles almost entirely.

Jim

 

[Jim Kasson]

Less is better, of course, but the graph seems to show that the worst noise is at ISO 64, which, again, seems counterintuitive.

Considering the noise as measured at the sensor is appropriate if you're trying to decide whether to crank up the ISO or push in post, since you're going to take the same amount of gain one way or the other. The graph shows that you're just about as well off pushing in post from say, 800 to 2000 and 3200 to 6400. Pushing in post has advantages in highlight recovery, especially with modern versions of Lightroom.

A caveat: however important it is in some circumstances, read noise is not the entire noise story, and the technique that I used cancels out the component of read noise that is the same shot-to-shot.

Jim