ISO Full vs. 1/3 Stops (Noise & DR Test) for Still Photography

[mackguyver]

The topic of the which ISO is best came about several years ago with the 5D II and Technicolor's release of their CineStyle Picture Style. Numerous blogs and tests were done for video, but it seems that this is still an unanswered question in terms of stills. There have been several debates over the camera pushing and pulling data, reduced DR, etc.

I believe I ignited the latest round of this discussion when reviewing some distant telephoto shots, so I'll see what I can do to settle it.

Over the next few days, I'll get out some of my IT-8 targets & ColorChecker Passport and take lots of shots under controlled lighting with my 5DIII and TS-E 24 II and/or 300 2.8 IS II at numerous ISOs and light levels. I'll post the results / files and ask for peer review. I don't have any analytical tools, so will have to go by IQ and file sizes, I suppose, but I'm still thinking about it...

If you have any suggestions before I start testing, please let me know.

 

[Marsu42]

Re: ISO 100 vs. 160

under controlled lighting with my 5DIII and TS-E 24 II and/or 300 2.8 IS II at numerous ISOs and light levels.

That's great, it really needs a controlled setup to be able to say anything beyond guessowork - and btw the dr curve might be different between different camera models like 5d3 & 6d (unfortunately there is little dxo data on the latter).

Alex (the main Magic Lantern dev, great guy) also just did a quick test, do read about it here, no need to duplicate tests: http://www.magiclantern.fm/forum/index.php?topic=9867.msg95051#msg95051

So, ISO 160 multiples seem to have 0.1 EV more DR and ISO 250 multiples seems to have 1/3 EV less DR than ISO 200 multiples.

White levels on ISO 160 & co was always below the white level of 200 and 250, but white(200) was equal to white(250). This shows there's something getting saturated at 200 and 250, so it confirms my previous theory that the real ISO is somewhere between 160 and 200 (most likely around 185, 375, 745 and so on).

 

[mackguyver]

Re: ISO 100 vs. 160

Hmm, Alex's comments are interesting, but for me, I'd like to try testing with real images (even if they are test charts ) to see the results with my own eyes. I specifically want to see what happens when I run the RAW files through ACR and DxO PRIME as well. I have no plans to test JPEG or video output.

 

[Marsu42]

Re: ISO 100 vs. 160

I specifically want to see what happens when I run the RAW files through ACR and DxO PRIME as well.

Good luck with that, at least with ACR I doubt you'll get good results as highlight recovery is a bit on the artistic side ... alex uses dcraw, I suppose this dumps the whole dynamic range in on go.

I'm really looking forward to see you results, if iso160 is a good choice I'll add a SmartISO(tm) function to the auto_iso Magic Lantern module I wrote so it automatically selects -1/3ev pulled iso for the low isos, and full iso steps above.

 

[mackguyver]

Re: ISO 100 vs. 160

I specifically want to see what happens when I run the RAW files through ACR and DxO PRIME as well.

Good luck with that, at least with ACR I doubt you'll get good results as highlight recovery is a bit on the artistic side ... alex uses dcraw, I suppose this dumps the whole dynamic range in on go.

I'm really looking forward to see you results, if iso160 is a good choice I'll add a SmartISO(tm) function to the auto_iso Magic Lantern module I wrote so it automatically selects -1/3ev pulled iso for the low isos, and full iso steps above.

Definitely true about ACR, so I'll probably start with DxO and try the "No corrections" preset initially. It's going to be a learning process I'm sure and I'm curious to see what happens at higher ISOs

 

[Sporgon]

Re: ISO 100 vs. 160

I've never really tested it seriously, but on the 5D mkii my feelings are that if I'm going to have to seriously lift shadows then ISO 160 appears to a have a very slight advantage in terms of lower noise. ISO 100 appears to give fractionally better saturation overall. I can't see any difference in noise between 100 and 160 in data like skies. However the difference is so low I would always shoot at 160 when handheld to gain faster shutter speed.

ISOs that are lifted from native ones, on the mkii that's the likes of 125, 250 etc are fractionally more noisy.

Going back to the original 5D, ISO 50 ( or overexposing 100 by a full stop in raw, then bring down in post ) definitely gave the best IQ in terms of saturation and low noise.

I'll be interested to see what your conclusions are with the mkiii.

 

[Marsu42]

Re: ISO 100 vs. 160

Btw the Title should be "ISO 200 vs. 160" or exactly "100x iso multiples vs. 1/3ev pull" - I don't doubt it's "better" to take shots @iso100 if you can, but it might be clever to use 160/320 and maybe 640 instead of the according base iso.

 

[mackguyver]

Re: ISO 100 vs. 160

Btw the Title should be "ISO 200 vs. 160" or exactly "100x iso multiples vs. 1/3ev pull" - I don't doubt it's "better" to take shots @iso100 if you can, but it might be clever to use 160/320 and maybe 640 instead of the according base iso.

Okay, changed the title to something more appropriate

 

[mackguyver]

Still thinking this through, but I think I'll do it as such:

Set Up:
Target: Lasersoft A4 sized IT8 target (I used to beta test for them)
Lighting: 2 x Einstein 640 lights at 45 degrees to target, 5000K modeling lights, using the strobes seems too annoying
Gear: Canon 5DIII + 300mm f/2.8 IS II, on solid tripod, RRS head, weighted down with 10lbs of sandbags
Tethered shooting with EOS Utility from 20' USB cable-connected laptop
Settings: RAW files, manual focus, manual WB, aperture priority @f/5.6, disable: highlight tone priority, ALO, noise reduction, etc.

Shooting Sequence:
Use shortest shutter speed possible to minimize sensor heat influence (not sure I can actually light it this way, but will try):
1. Adjust Lighting power to achieve correct exposure of 1/8000s at f/5.6 at ISO 12,800
2. Shoot a frame at each 1/3 ISO stop from 100-12,800 (ISO 50 is known to be a reduced DR mode)
3. Repeat sequence at half light (1/4000s, f/5.6, ISO 12,800), quarter light, and possibly 1/10 light
4. Repeat full set using Expose to the Right (ETTR) exposure.

Batch Processing (initial test):
1. DxO 9 (No Correction Preset), normal NR - 16-bit TIFF output
2. DxO 9 (No Correction Preset), PRIME NR - 16-bit TIFF output

Results (still thinking about this):
Graph RAW & final file sizes by ISO, processing (decent indicator of noise/resolution at least in terms of standard devations)
Prepare side-by-side comparisons of the gray patches, color patches by ISO, setting
TBD...

Your thoughts and opinions are welcome - as I said, I'm still thinking about how I want to do this...

 

[ejenner]

I've messed with this to my satisfaction with the 5DII with real images with much greater DR than the camera can handle. With brights where you can see where clipping occurs and deep shadows that show noise, even at low ISO.

One thing that gets me is that it is always between ISO160 and 100 multipliers. If the 160 is 'true' ISO, then surely 125 would be less noisy because it would be 160 pulled.

The fact that I've had people on the net swear blind that ISO320 is better than ISO100 is enough for me to always do my own tests (or even that ISO160 is better than 100).

Agree 100% with ACR though. It's good, but definitely does 'stuff' behind the scenes (sharpening for one). even with everything 'off' or zero.

 

[Marsu42]

Your thoughts and opinions are welcome - as I said, I'm still thinking about how I want to do this...

When in doubt, also use ufraw - it's oss and that's why the Magic Lantern devs use them, but then again it might dump less "cooked" results than commercial software.

The fact that I've had people on the net swear blind that ISO320 is better than ISO100 is enough for me to always do my own tests (or even that ISO160 is better than 100).

The net is full of urban myths, but I think it's worth really figuring this out - for example when I first got my 60d I often shot @iso1000 which in hindsight is an extremely bad idea, I should have decided either for 800 and accept a lower keeper rate or go straight to the next higher analog iso and expose properly rather than underexpose.

One thing that gets me is that it is always between ISO160 and 100 multipliers. If the 160 is 'true' ISO, then surely 125 would be less noisy because it would be 160 pulled.

Nope, Canon doesn't work this way ... the 1/3ev spacing setting means that ~-1/3ev from 100 mutiples is pulled, ~+1/3ev is pushed. The internal Canon firmware values are different altogether, iso 100 is "72", +1/3ev is +3, +1/2ev is +4, +2/3ev is +5 ... so there is some rounding error even at this stage. Here's a part of the Magic Lantern source I recently added to lens.h:

#define ISO_100 72
#define ISO_125 75
#define ISO_160 77
#define ISO_200 80
#define ISO_250 83
#define ISO_320 85
#define ISO_400 88
#define ISO_500 91
#define ISO_640 93
#define ISO_800 96
...

The notable fact is that the iso setting on the camera isn't the "real" iso set in any case, for example iso 100 setting actually is measured iso 80 ... so there might be some effects at work which make -1/3ev gain a tiny bit of more dynamic range. The noise/dr should thus also be correlated to the actual iso, not only to the nominal iso set on camera, the data can be gotten from dxo or sensorgen.

 

[mackguyver]

Unfortunately things didn't work out over this weekend to run my tests. I'll try to get things going soon.

 

[Gert Arijs]

I look forward to your tests, have you managed to do it yet?
Gert

 

[Marsu42]

I look forward to your tests, have you managed to do it yet?

Hopefully not, the latest Magic Lantern research adding 1/3-1/2 stop to full isos make the 160x values completely obsolete, even if they're a tiny bit better than Canon's defaults.

 

[pensive tomato]

I look forward to your tests, have you managed to do it yet?

Hopefully not, the latest Magic Lantern research adding 1/3-1/2 stop to full isos make the 160x values completely obsolete, even if they're a tiny bit better than Canon's defaults.

Marsu42, I'm not fully following what you're saying, do you mean that the ML folks determine that using full ISOs is the recommended course of action to decrease noise and optimize DR? Or do you mean that an ML hack already makes this a moot point? Thanks for any clarification!

 

[mackguyver]

I look forward to your tests, have you managed to do it yet?
Gert

Sorry guys, still healing up and trying to keep up with my day job & the teaching job I've had to take on to make up for my lost photography income . I have the night off, so I'll try to do it tonight.

 

[Marsu42]

do you mean that the ML folks determine that using full ISOs is the recommended course of action to decrease noise and optimize DR? Or do you mean that an ML hack already makes this a moot point?

The latter, though I wouldn't call it a hack - by fine-tuning the full iso stops the bleeding edge Magic Lantern is able to crank an additional 1/3-1/2 stop from full isos, much more than any potential improvement you'd get from Canon's 1/3 ev steps.

I have the night off, so I'll try to do it tonight.

Really, under these circumstances don't bother - the ML rand libraw devs are on this with a lot of more time (and probably expertise) than you could currently manage: http://www.magiclantern.fm/forum/index.php?topic=10111.msg98298;topicseen#msg98298

 

[pensive tomato]

do you mean that the ML folks determine that using full ISOs is the recommended course of action to decrease noise and optimize DR? Or do you mean that an ML hack already makes this a moot point?

The latter, though I wouldn't call it a hack - by fine-tuning the full iso stops the bleeding edge Magic Lantern is able to crank an additional 1/3-1/2 stop from full isos, much more than any potential improvement you'd get from Canon's 1/3 ev steps.

Thanks for the reply, that's informative. As you pointed out in another thread, that's one more reason to consider ML (although I'm a bit hesitant with my 6D still being under warranty).

By the way, I meant hack as a compliment, more along the lines of creative tinkering of existing resources

 

[LetTheRightLensIn]

More or less ISO160 and ISO200 and ISO320 and ISO400 etc. are the same. You can set ISO200 and EC +1/3 and get the same as ISO160 or ISO160 and EC -1/3 and get the same as ISO200. The RAW files are stored a bit differently but the end result is basically the same. Who knows maybe one actually has 0.1 stop more DR or something, but from a practical standpoint who can tell +0.1 DR from +0.0 DR???

Once you above ISO1600 for most of the older bodies everything is just pushed ever 1/3 more, more recent bodies had that shift point at ISO3200. There may be or two where it is ISO6400, I forget.

The only thing to avoid is the ISO125 and so on.

The only way practical at all to get more DR might be the new ML thing which apparently gives you 1/2 more DR at any ISO. Focusing on whether ISO160 has 0.1 stops more DR than ISO200 seems kind of useless to me.

 

[bholliman]

Sorry guys, still healing up and trying to keep up with my day job & the teaching job I've had to take on to make up for my lost photography income . I have the night off, so I'll try to do it tonight.

Sorry to hear things are not going well. Get healed up and if you have time down the road, do the test.

I do not plan to use ML on my 6D until it reaches a beta level and that might be another year or two, so myself and probably many others, would be interested in your results.