Explosions In The Sky. Nikon D3 and 14-24mm at 14mm. f/2.8, 1/60, and ISO 12800.
Answers to the Pepsi Challenge of high ISO and push processing, and an introduction to unity gain. Earlier in the week I posed a simple question: Can you tell the difference between an image shot natively at ISO 12800 and one underexposed at ISO 6400 and pushed +1EV in post?
Between responses in the original thread, comments on Flickr, and votes through Twitter, the opinions broke down as follows:
Sample A was pushed: 6
Sample B was pushed: 10
Undecided/too close to tell: 8
By a very slight margin, the majority opinion was that sample A was more appealing due to slight differences in noise, contrast, and perceived sharpness, though most everyone agreed that the samples were almost identical.
The Answer
Sample A was the ISO 6400 image with +1EV in post. Sample B was the “native” ISO 12800 file.
The Explanation
So, how can an underexposed RAW file possess image quality that is the same (or arguably better) than an image properly exposed at a higher ISO? Shouldn’t software-enabled amplification of the signal result in inferior results?
At the heart of the issue is the way in which the camera at hand, the Nikon D3, achieves the sensitivity of ISO 12800, which is specified as an “extended” setting (“HI 1.0″) above the camera’s standard range of 200-6400. Specifically, unlike lower settings, ISO 12800 is achieved in-camera by multiplication, rather than analog amplification via hardware.
Thus, the ISO 6400 file and the ISO 12800 file were both essentially “pushed” in parallel means, albeit different points of the data chain, which accounts for the similarity in the final images.
An important concept here is the “unity gain ISO,” which describes threshold at which the A/D converter digitizes 1 electron into 1 data number of a digital file. Since 1 electron the minimum amount of information necessary in this process, the unity gain ISO represents the point after which digital multiplication must take over, analog amplification.
Shooting above the unity gain ISO, the signal-to-noise ratio decreases as the limit of real information collected by the sensor and digitized by the A/D converter is reached. Or in other words, above the unity gain the camera is going to be working to multiply more noise and less real, image-forming information.
Furthermore, raising the sensitivity above the unity gain ISO results in decreased dynamic range due to highlight clipping as a result of the multiplication of data.
What does this mean for the D3 and these samples?
From these results, we can assess that ISO 12800 is above the unity gain ISO of the Nikon D3 (which, depending on who you ask, is ISO 3200 or 6400). As such, with regard to image quality, there are few to no practical advantages of shooting at ISO 12800 over ISO 6400, since the former is a multiplied product of the same threshold of information.
All digital cameras and DSLRs possess have a unity gain ISO, and it will vary from model to model depending on the pixel pitch, signal amplification, and other factors of sensor efficiency. If you want to curl up with a lot of hardcore graphs and talk about quantum efficiency, you can check out an in-depth comparison of many DSLRs and a better explanation of unity gain at clarkvision.com.
For reference, here are the unity gain ISO ratings for some current cameras:
- Canon 5D Mark II: 1600
- Nikon D50: 1500
- Nikon D300: 1100
- Canon 50D: 900
But I like to photograph rock bands
OK, you’re wondering, “What does this mean in practical terms for real world shooting like concert photography?”
Knowing the unity gain ISO of your particular camera can translate into smarter choices of picking in-camera sensitivities, allowing you to maximize image quality and dynamic range. In terms of the latter, the drop going from ISO 6400 to ISO 12800 is roughly one-stop. Contrast, which many people picked up on in the test samples, may take a hit above the unity gain ISO, as can color saturation/fidelity.
While the goal of ultimate image quality and the pursuit of those iconic, Jimmy-Hendrix-burning-his-guitar-on-stage moments may not seem particularly aligned, knowing the limits and capabilities of your camera’s files can only benefit your output.
Of course, shooting in this manner necessitates post processing and conversion from RAW, so there are no benefits for JPG shooters. In addition, the ends will not justify the means for most photographers out there, given the extra work involved and the extremely similar results, as the previous samples indicate.
End Notes
There you have it – the long story of why an underexposed shot at IOS 6400 basically looks the same as an image at ISO 12800 with the D3.
In summary, the samples from the Pepsi Challenge look almost identical because both the ISO 6400 RAW file and the ISO 12800 RAW file are working with essentially the same info to begin with.
At the very least, the next time you think you’ve underexposed a high ISO shot, you may just have done yourself a favor.
This entry was posted on Wednesday, January 28th, 2009 at 7:17 pm and is filed under Photography Gear and tagged with concert photography, d3, high iso, iso, nikon, unity gain, unity gain iso. You can follow any responses to this entry through the RSS 2.0 feed.
Welcome to ishootshows.com, your source for music photography featuring daily concert coverage, portrait features, and photo advice from professional music photographer Todd Owyoung.
Support this site by purchasing your photo equipment through my affiliate links to B&H Photo Video or Amazon.com
Learn more about how you can support the features of ishootshows.com.
Shooting Caitlin Rose and band before the Old 97s was a fantastic warm up, and a nice way to dust of…
Here’s a little preview from a recent shoot with my friends Option Control toward the end of l…
Today Adobe has released the beta version of Lightroom4, the successor to their popular image manage…
You’ve seen it before – weird, colored blobs in your photographs radiating outward from some…
About that time when I said about the megapixel war being over with the release of the Nikon D4 and …
ZING!
Now where can I claim my pepsi? ;)
Ha, I’ll buy you a cold when when I make it Down Under!
Yes, let the record show that Joel called it first, and correctly. Thanks for playing.
Well, Joel might have been first, so he can have the Pepsi… but that’s fine, because I don’t like Pepsi. I called it second though, so can I claim the Coke? ;)
Really interesting, Todd. And really technical. Wow.
Sure, you can have a Coke when I’m Chi-Town for the Fall Out Boy tour. Or Pitchfork, or Lollapalooza.
Thanks for the comment, Tasha, hope this post was interesting to you.
2009 Goal: learn and completely understand all the technical terms in this post
Hey Jessie, that’s one of my goals, too! ;)
Here’s a great explanation of some of the more technical details of unity gain and the way ISO works with digital sensors – last post on this page:
http://cs.astronomy.com/asycs/forums/t/35652.aspx
@toddowyoung bookah! Called it first :D ZING! Re: http://snurl.com/awwdj
Mind-blowing and eye-glazing at the same time. Impressive.
Hey Mike, thanks for the comment, hope the post was interesting! I certainly learned a lot in researching and writing it.
My brain hurts!
Todd, are you now changing the way you shoot? Would you rather have your camera not operate beyond the D3‘s unity gain iso rating preferring instead to underexpose by one stop where needed, then reclaim that one stop in post? If so you could simply apply that extra stop as an auto step in say NX2 so you lose little in time.
Hey Paul,
Just from shooting and processing many, many high ISO files from the D3, I’ve long since had a gut feeling that ISO 3200 is about the limit that I like to shoot at. From there, ISO 3200 pushed one stop was essentially no worse than shooting at ISO 6400, and cleaner in some ways, and so there have been shows where I’ve consciously tried to stay around 3200 as my limit.
Another factor of this is that shooting at the unity gain with the intent of pushing in post allows one to shoot more cautiously in some lighting with less worry about clipping highlights.
Overall, I think that will definitely consider ISO 3200 as a loose limit, though it’s ultimately going to depend on the event and the turn-around needed on the images.
thanks for your comment!
Thank you for the explanation and links on unity gain Todd. :)
I had assumed that the reason both shots looked nearly identical was due to the max native ISO on the camera being around or below 6400 (and thus both shots had really been pushed the same amount, albeit at different points in the signal chain), but the concept of unity gain is quite new to me, and rather enlightening as well.
Hey João,
You’re right. I think that this concept of unity gain is one that a lot of shooters arrive at, though the exact mechanics of it aren’t necessarily known. In pushing various high ISO images, I’ve often thought, “Well this doesn’t look any worse than if I’d shot at ISO X!” And now we know. Thanks for the comment!
Just tried it out in and found my results were slightly different which, Todd, may make this process not workable for you in poor lighting conditions.
In reasonably well lit areas the images produced were almost similar. An iso 800 -1 ev shot taken on a D300 then bumped +1 ev in post looked almost identical to a native iso 1600 in the same lighting conditions.
However, employing the same technique but with much reduced ambient lighting caused the bumped file to contain significantly more noise especially in the shadows than the native iso 1600 file.
And so for me, it’s good to know and keep the technique in the ‘mental bag of tricks’ for a last stand approach.
Hey Paul,
Thanks for the follow up, it’s great that you’ve had a chance to test this and share your results. As shot at ISO 800, you’re still a bit below the unity gain of around ISO 1100 (I’ve also seen reports that it’s closer to 1250) on the D300, so that could have something to do with it.
Either way, maybe I will try to do some more extensive tests on this and get back with some results. Thanks again for sharing your experience!
Concert photographer extraodinaire @toddowyoung writes on the possible advantages of push processing vs high iso: http://tinyurl.com/adjauz
Here’s a link to an interview did with representative from DXO about the next version of their RAW converter:
http://www.imaging-resource.com/NEWS/1224662427.html
It’s topical to this discussion, as around 8:18, they start talking about pushing an ISO 6400 image from the D700 to ISO 51200, up three stops from ISO 6400. At 10:00, they very briefly hint at the reason for pushing from 6400 instead of just +1EV from ISO 25600.
The interviewer seems to perk up at the idea that a lower ISO would be better for pushing, but unfortunately they don’t go into too much detail.
Todd,
Thank you for this post. You’ve actually touched on something that I’ve been trying to figure out since buying my Canon 50D — grain amounts in high ISO vs. push processed images. I looked at the article at clarkvision.com and, according to the graph, it seems that the 20D, 40D, and even the XT all have a higher Unity Gain than the 50D. Perhaps I’m misunderstanding Unity Gain, but it would seem that the newer 50D with the reportedly improved sensor and capabilities would outperform the older models at higher ISO speeds.
In my experience, the 50D definitely produces a cleaner image at ISO 1600 than the 20D did. But it’s disappointing to find that it “shouldn’t” produce a clean image at 2000 or above, especially considering the majority of my photography is low-light sports and concert photography.
Am I misunderstanding the graphs? Is it suggesting that those lower on the Unity Gain scale (1000 ISO, for example) produce grainer, poorer quality images than those higher up (at, say, ISO 2000)?
Thank you,
Ally
Hey Ally,
Good issue you raise. I think that it’s important to note that unity gain isn’t the definitive factor for performance at high ISO, but only one limit of sensor efficiency at a very specific point in the range of “sensitivity.”
There are other elements at play for the overall performance with regard to noise, such that a camera with a lower unity gain might ultimately outperform one with a higher rating, which is what you’re seeing in the 50D vs 20D.
“At the very least, the next time you think you’ve underexposed a high ISO shot, you may just have done yourself a favor.”
Absolutely!
This is very interesting, and one less thing to consider before, during or after those precious moments.
Thanks
Brian
Hey Brian,
Thanks for joining in the discussion, glad you found the post interesting! You do bring up an interesting point, and I think you’re right in a sense – that at a certain point (the gain unity ISO), you can just “set it and forget it” if the conditions are right.
For example, if the show is definitely going to require sensitivities above the unity gain ISO, I suppose one could plan for underexposure within a range of say, 2 stops, and stay locked at the unity gain ISO. Exposure compensation would be necessary in post, but the image quality shouldn’t suffer to a huge degree, if at all – in theory.
Nice job Todd.
Thanks, Chris.
Hey Todd,
Me again. I’ve done some further none scientific testing using the D300 at iso 1250 which is slightly higher than the site suggests is optimum for the D300. They suggest iso 1100 is where it taps out but given that I can’t get the D300 to exactly iso 1100 I had to make do.
Here’s a link for two NEF files both shot with the Sigma 30mm f1.4 lens using matrix to meter. First shot, file BXF8309 meter read correct exposure and the second file, BXF8311 with -1 ev. Apart from a slight change in contrast on the 2nd file, they are as close as I could get to making the same image.
Open the files, I use NX2 btw, and viewed full frame they look real similar but the devil is really in the detail. The room was lit with two small free standing lights and reasonably dark but the noise is very much more apparent in the pushed file than in the straight 1250iso shot.
http://www.mediafire.com/?znymnw5jeli
http://www.mediafire.com/?ttftzmudyen
Interested in your thoughts.
Hey Paul,
Thanks for the links, I just downloaded the the two NEFs and opened them in NX2. I see exactly what you mean about the shadow noise, especially when turning off noise reduction in the advanced camera settings menu. There’s certainly a difference in the levels of noise, not only in the shadows, but the upper tones as well.
But looking at the EXIF of these shots, NX2 indicates that these were both shot at ISO 1250 – is that right? If so, that would make your experiment a slightly different test than what I conducted. A more parallel test would be if you shot at ISO 2500 as the control, and then to push an ISO 1250 shot +1EV for a noise comparison. That said, RAW files from different cameras will respond differently to pushing and pulling exposure.
Arg! Bum!! You’re right Todd, totally missed that I should have shot at 2500 and a -1 ev 1250 shot so not a fair comparison.
It’s 1.47am here and off to brew a drink and shoot two more files…post back in a bit.
Mindful that I don’t want to become a pest on your site.
Hey Paul,
I’d love to see the results when you get them uploaded. Keep us posted!
Hey Todd,
Here you go…making the adjustments to record an 2500iso shot to compare to a pushed iso1250 shot made all the difference. Here’s the shots…
1. iso1250 pushed 1 stop http://www.mediafire.com/?z4momnnzkyz
2. iso 2500 http://www.mediafire.com/?ckmhmyzyjwm
Looking closely at the files there is definately colour noise as to be expected but it looks to me, under closer scrutiny, that the pushed shot retains a little more detail and house very slightly less noise. Splitting hairs frankly re the noise statement before but certainly worthy of more experimenting; I would also expect that the power of the D3 and D700 would yield better results given the ability of the FX sensor.
Hey Paul,
Great, thanks for doing this test and uploading the shots. Now this is much close to what I’d expect to see!
I opened the files and I do agree, the pushed file is doing slightly better in the midtones for noise. There’s less random color noise and few hot-pixels, and the noise pattern seems a little finer, as if there’s less variation between the highs and lows of the luminance noise. The ISO 2500 seems to have a slight bit more of a contrasty look, which is somewhat different than what I saw with the D3‘s files.
The shadows of both images look very, very similar to me.
Overall, I do prefer the ISO 1250 shot pushed. I’ll have to do more experiments with the D3 that have more of a tonal range, as you’ve shown here.
Thanks very much for sharing your results, Paul!
Everything depends on converter when you do this kind of pushing. NX2 is not bad in this area but certainly limiting. If you do pushing with RPP, especially by using compressed exposure you’ll see why less ISO is often better. On those examples from D300 f.e. in case of pushed ISO 1250 lamp (+1EV, compressed) still looks like a lamp and on 2500 it’s just a bright spot. And don’t stop on your way down with ISO – you may find that RPP allows you stay much lower and maintaining virtually the same picture quality after proper push.
Andrey, I’m a PC user so RPP, assuming this link is the correct app you refer to http://www.raw-photo-processor.com/RPP/Overview.html, won’t help me.
Dude, I’m completely lost on 90% of the terminology here, but I think I get the gist of it. Thanks for your work and information here… it’s a pretty neat idea, even though I don’t understand the definition directly, haha. But it does provide an interesting perspective and take on cameras and their ISO controls. Also, I think it contributes more to your site with regards to tutorials, etc. Thanks for the info, and nice job. Hittin the UK soon?? Let me know for sure.
Hey Chris,
Nice to hear from you, glad you found this post interesting. I’ll definitely hit you up when/if I’m coming to the UK this Spring.
Contrary to what’s implied here, ISO3200 is a real analog amplification step on the 5D2, as is ISO1600 on the 50D.
This is a very interesting post. I wish you would add shots taken at 3200, 1600, 800, and 400 – all raised to equivalent exposure.
Also, this test is a little unrealistic in terms of lighting conditions for “real world” use of such high ISO’s.
Couldn’t you give us a test setup with some shadows in medium or dim light?
Shoot ISO 6400 and push, or shoot ISO 12800? http://ishootshows.com/2009/01/28/push-processing-and-unity-gain/
Good article, Todd.
Once you have maxed out EV parameters, there is an excellent explanation of why push works and up to what limit here:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3a.html
Cheers
Very cool, thanks for the link, Jack.