A while back we posted a popular video series by Philip Bloom called New To DSLR Video? Here Are 8 Videos To Help. In one of the videos, Philip talks about how when shooting on Canon DSLRs (and presumably all DSLR cameras), certain ISO values actually may produce more noise than ISO values 1/3 and 2/3 stops HIGHER! Of course this made everyone a bit uneasy and sparked some interesting discussion. Well Andrew Schär made a video that demonstrates how ISO settings at multiples of 160 are actually better than the normal ones we use at intervals of 100. The question that immediately comes to my mind now is does this hold true for still shots as well or does it only show up when processing video? What do you guys think?
Related Articles
It rule works only in video, or in photo shooting too?
This is a very interesting find. Learning from film guys I automatically lock in at 100 or 400 but after seeing this I may have to play with my camera for a bit and see if the same find hold true for Nikon bodies.
I don't know if it holds true for stills, but I'm inclined to think that it is not.
The theory is that, in video, ISO multiples of 160 appear cleaner because the noise pattern is more easily downsampled into a cleaner looking 1080p signal and its subsequent encoding into H.264 compression.
So, although an ISO 160 shot may be more noisy on average than an ISO 100 shot, its noise distribution may be more advantageous for the signal processing and line skipping of a DSLR's video mode.
Ironies... so many obsessively shoot their images smooth... and then run actions with sophisticated old time grain structure. Silver Efex, Totally Rad, Ascough Silver...
:)
Noise is different from grain. Noise is digital, grain is film. Photographers never want noise while shooting, although sometimes they will postprocess their digital images to give it a grain effect.
Hi Viraj
I'm well aware and totally agree and officially there's a difference. And logically, it's always wise to capture clean and rough it up later... (and like most photographers, I crave all the latest super sensors and fancy noise reduction algorithms), but...
I still smile at the irony. Noise and grain are like kissing cousins.
I find it really interesting that ISO 125 looks worse than ISO 1250. That is a huge difference!
I found this article that talks a bit about it: http://shootintheshot.joshsilfen.com/2010/05/13/canon-hd-dslr-native-iso/
"
So, if the 160-multiple ISOs are not the native ones, why are they cleaner, and how are they derived? Well, it is correct that the 125-multiple ISOs are the noisiest because they are derived by a digital exposure push. ISO 125 is actually ISO 100 with a 1/3 stop digital exposure push, ISO 250 is ISO 200 with a 1/3 stop digital exposure push, etc. However, the 160-multiple ISOs are actually the cleanest not because they are "native", but because they are a result of a digital exposure pull. This pull brings down the exposure of the entire image, and hides much of the noise that would be visible at the next higher ISO. ISO 160 is the cleanest because it is the native ISO 200 with a 1/3 stop digital exposure pull, yielding even less noise. ISO 320 is actually ISO 400, with a 1/3 stop exposure pull, etc."
Interesting theory altough I would love some electronical engineer to poke at this theory... just to be sure!
I think the theory held true until it got above 1600, they all just looked rough after that.
That's also why in Video, Iso 320 and 640 have better Colors and the images appear to have more depth than other ISo settings.
Nice. I wonder if this applies to full-frame sensors like the 5DmkII's.
I guess I'm showing my digital camera ignorance here but why are all 1/3 stops not native? I mean aren't we just talking about electronic gain here? It seems like there should not be any real sense of pushing or pulling but rather simply "dialed in". When I turn the gain up on a digital guitar amp, it's not pushing it to say 7 and then digitally bringing it back down to 6.5. I guess I still don't understand how sensors work.
And if there is something true about native stops on a sensor then wouldn't this have to hold true with the normal still shots too? I like Thomas's explanation that it has something to do with 720 and 1080 resampling and compression into H.264. Anyone know for certain?
Patrick, Well it's one louder! http://www.youtube.com/watch?v=XuzpsO4ErOQ
Wow John. I was not expecting that comment. Props man, you made me laugh like crazy.
this only applies to video-- not stills. still photogs can ignore this video. it is, however, a good demonstration for video folks. as a videographer, i try to shoot at 160, 320, 640, 1250, 2500, and then i try not to go above that if possible. only exception is when its sunny i'll shoot at 100, because you don't really need to worry about grain until you get over 640.
I did some tests the other night after seeing on creative live where they spoke of multiples of 160. Seems it holds true for stills. I read somewhere else though that you loose about a third a stop in dynamic range from 160 to say 200.
Maybe I'm just too tired, but is it either me or does the red highlighting of the multiples in the video go haywire at ISO800 and upward? Last time I checked 800/160=5, and 1250/160=~7.8. Not exactly multiples of 160... It's not the point of the test really, in fact it's quite interesting how the noise levels compare as the apparent sensitivity is increased; but I don't think anyone has picked up on it at this point.
Does anybody know if this, or it's equivalent, applies to Nikon cameras? I've seen this idea floating around the net, but only see it related to Canon sensors.
I did a quick test with the Nikon D7000 here: http://www.davidmonnerat.com/2011/07/09/iso-testing-on-the-nikon-d7000-d.... In an e-mail exchange with Andrew, he indicated that this whole 160 ISO thing was with the Cinestyle profile he loaded on the camera, and it does things differently than the "normal" profile might. Since Nikon doesn't support that type of picture profile that the Canon does, it likely doesn't apply.
Wow, here's me going blindly thinking that the higher yo go the noisier you get ... hmm Should have seen this video before I went shooting the club last night. I was using ISO 2000 and 4000 grrrr. Never again, increments of 160 for me :)
Right off the bat, the test has been contaminated by the use of a color profile that is not native to the camera. The test needs to be redone using a default profile that comes stock with the camera. The technicolor cinema profile may be what is causing this perhaps? It's always good when conducting tests that you have a control test (test with default profile), and then the test with the variable (test with technicolor profile). Just to make sure the technicolor profile is not inducing the noise.