5D Mkiii versus 5D Mkii high ISO test revier

Discussion in 'Canon EOS' started by davebell, Mar 22, 2012.

  1. Hi all,
    I've received my 5D Mkiii body today and decided to run a test of high ISO performance versus my Mkii. I've written a fairly detailed review including a bunch of images so please feel free to have a look here:
    http://www.bellissimaphoto.co.uk/photographers/canon-5d-mkiii-review-vs-5d-mkii-high-iso-test.html
    Any comments are more than welcome!
     
  2. Shot with one today. Didn't do any objective tests, but I was very impressed. ISO 16000 in a dark bathroom looked quite clean (for ridiculously high ISO).
     
  3. I'll buy any camera that'll clean my bathroom.
    :-D
     
  4. ...Seriously though, thanks for the test. I'm currently shooting a 5D mk I; in low light I usually use ISO 1250, which I find to be quite good when exposed and processed well. Looks like I'd easily be able to move up to 6400 with a 5D mk III; or perhaps up to 12800.
    Likewise, the shooting for detail and good color that I generally do at 400, could probably be done at 1600. That's a big deal because I often shoot group photos in marginal light at f/5.6. Shots I used to take at 1/20 could be shot at 1/80. Nice.
     
  5. Good information David and it makes me in less of a hurry to upgrade. I don't shoot that much above ISO 1600 with my
    5DII very often and find the AF is good enough for the subjects I shoot with the 5D. Since I shoot sports with a1dIIn or a
    7D the extra fps, AF and high ISO performance of the 5DII. I suspect I will end up with a 5DIII but I can wait a few months
    now
     
  6. Thanks, David! Exactly the images I've been waiting to see! I'm definitely impressed with the 5DIII, despite everyone else's griping and groaning. :)
     
  7. Up too 6400 they look pretty well the same - couldn't pick from these shots which was which in a blind test, but above 6400 then the MkIII is ahead. Doesn't make me want to upgrade to the MkIII. I have only shot at 6400 on two occasions that I can remember.
     
  8. david_henderson

    david_henderson www.photography001.com

    For me, most of the time I have to use a high ISO is with a dark subject in relatively low light, where I don't have the option of shooting at ISO 400 if I wish. This test doesn't replicate the conditions in which I use higher ISO, and therefore is of limited use to me since on the subject you choose I could shoot at ISO 100.
    On the other hand on a recent trip to India I was trying to make sense of some interiors for which, with no tripods allowed and the need for some dof, required me to shoot at ISO 3200 with my 5dmk ii, or ISO 6400 if I didn't want to alter the exposure or brightness afterwards. Unsurprisingly (to me anyway) the results weren't great noise-wise, and various noise-reduction routines have been tried without huge success. For me, if the Mk iii is decisively better in these circumstances, it would provide a rationale to upgrade.
     
  9. In your review: "Note: for the same aperture, shutter speed and ISO combination, the 5D Mkiii underexposed, versus the Mkii, by around 1/4 of a stop. This was compensated for by slowing down the shutter speed for each of the 5D Mkiii images, instead of adjusting in post processing."
    Unfortunately, you are giving the 5D mk III the advantage of more available light. You should have adjusted the exposure digitally. This is what the camera should have done with analog gain in the first place, to get the same image brightness at the same exposure. Effectively this means the ISO value of the mk III is lower than what it really says.
    If you download CR2 files from Imaging Resource of the 5D, 5D II and 5D III, you will notice that:
    • 5D III: ISO 3200, 1/400 f/8
    • 5D II: ISO 3200, 1/400 f/8
    • 5D: ISO 3200, 1/500 f/8 and image is even 0.33 stop brighter than the other two after equalizing white balance and contrast.
    I note several things in DPP: 5D III files do not sharpen by the same amount for the same slider settings (noise and sharpening). 5D III also has poorer detail in the red compared to both 5D and 5D II with all noise reduction and sharpening turned off. So I turned everything to zero, saved as TIFF and sharpened in another program. Strangely I can't get the 5D III image as sharp as the 5D II and it is not a focus/DOF issue. It looks like they crippled DPP because even Irfanview can get more detail out of it, in fact a little more than the 5D II. While I like the new HDR in DPP, it's disappointing to see it can't pull 5D III details the way it can with 5D and 5D II!
     
  10. Edit: found the reason: "High Quality" (moiré reduction) versus "High Speed" in DPP preferences. Being used to the 5D, there never was a large difference and I had it at "High Quality". Now it is HUGE. If you want detail in the 5D III using DPP, set it to "High Speed". After doing this and setting the same sharpening, at ISO 3200 I see effectively very little differences between these three cameras in these test images: maybe 1/3rd stop improvement of the 5D III compared to 5D II and at the pixel level it looks the same as the 5D, which is 1/4+1/3 stop faster at the same ISO though, so effectively all benefit comes just from the increase in number of pixels. Still would be good to verify with more tests.

    Since the original poster did his test using DPP, was it set to "High Quality" or "High Speed"? I strongly recommend the latter.
     
  11. Thanks for the illuminating test reports, David and Oscar.
    Since I never shoot above ISO 1600, and usually am in the 100-400 range, I'll definitely be holding onto my 5DII.
     
  12. I agree with Oscar about the moire reduction. I wrote a report with images here: http://www.dgrin.com/showthread.php?t=217561
     
  13. Giving the Mk III more exposure really does through off the results. I know that under exposing even a little can cause
    noise to become unacceptable with the mk II. But,even so, the Mk III looks to be better. But, not enough for me to
    spend that kind of cash. If it really was two stops it might be worth it.
     
  14. Oscar and Jack... interesting find there, good to know.
     
  15. Fascinating. Unfortunately there are several critical flaws in the methodology which undermine the validity of any conclusions one could draw from the results, especially since there were no real 'conclusive' results.
    Exposure approach: Manual. Aperture set to a constant F5.6 to ensure plenty of DOF. ISO from 400 to 25600 with corresponding increase in shutter speed for each ISO increment. Note: for the same aperture, shutter speed and ISO combination, the 5D Mkiii underexposed, versus the Mkii, by around 1/4 of a stop. This was compensated for by slowing down the shutter speed for each of the 5D Mkiii images, instead of adjusting in post processing.​
    Since the mk3 underexposed (by default, or firmware, or metering flaw), necesetating a bump in exposure time, every single image recieved a greater 'signal' than the 'similar' mk2 image. 1/4 of a stop drastically affects the amount of noise in a file (as we all know well). Since the mk3 got a 'bump' throughout the entire test, one cannot conclude that the 'apparent' performance is natively better.
    The proper methodology would have been to 'balance' the exposure difference (between the mk2 and mk3) and set the exact same exposure to each camera (using manual) for each subsequent shot.
    5D Mkiii files: as above, set the noise reduction to zero. Here comes the first interesting observation – the Mkiii files are a bit softer than those of the Mkii so I set the sharpening level to “2″. If anything, this will degrade the observed noise but I decided to do this anyway. Also note that the softer images of the Mkiii sharpen up very nicely so I don;t see this as an issue at all.​
    Canon's algorithm for converting the RAW data from the mk3 to a JPEG format is different than that which converts mk2 RAW data to JPEG format. Assuming that a specific instruction to an algorithm (whose mechanism is not completely understood) will result in a particular behavior is scientific fallacy. Canon (themselves) have stated that the in-camera conversion (RAW->JPEG) produces a 'two-stop' improvement in noise, implying that they have 'tweaked' it in some undisclosed manner to make for more effective NR during the RAW to JPEG conversion process. Altering the setting from the mk2 to mk3 only further contaminates the data.
    Proper methodology would be to NOT convert from RAW to JPEG for the purposes of high ISO noise comparison (or at least not w/ Canon's proprietary software). Instead, one could view @ 100% ->200% (for clarity) side by side, and take a screen shot of both side by side for comparison.
    All images were ten converted to JPEGs as maximum quality level.​
    As above, since the conversion process is done by different instruction sets, how effective that conversion is is what you are now evaluating, NOT how well the camera natively performed at high ISO noise.
    So, in a nutshell, if the exposures were the same, the conversion was the same, the software (algorithms) used to convert were the same, and the settings used for conversion were the same, and conclude that the mk3 has between 1/2 -> 1 stop better native RAW high iso noise performance. However, since the exposure was NOT the same, the convertion process was NOT the same, and the 'settings' were NOT the same, it's hard to make any specific conclusion at all.
     
  16. Thanks for all of your comments! Oscar you have raised a really good point about the DPP settings. Firstly the images shown in the test had DPP set to "high quality". I have never really used DPP so I wasn't even aware of this setting. At least the setting was common between the Mk ii and Mk iii images for the review.
    I did a little more investigation and found the following, working on a single Mk iii 6400 ISO file:
    For zero noise reduction in all cases except where indicated.....:
    1) DPP set to "high quality" => noise as shown in review, image a bit soft.
    2) DPP set to "high speed" => noise distinctly worse than (1), image noticeably sharper.
    3) ACR => image looks very very similar to (2)
    4) ACR with Lum/Col noise reduction set to 20 (other NR sliders at their defaults) => lower noise and sharper than (1) but not as sharp as (2) or (3).
    So it almost seems that DPP may be applying some kind of softening when set to "high quality" even though the noise reduction was set to zero.

    I will update the review with the images tomorrow morning.....
    Regarding the adjustement I made to the Mk iii exposures to match those of the Mk ii, I considered both options, i.e. slow down the MK iii shutter or crank up the exposure in the RAW adjustment tool. I think both are not ideal but I thought the former would be better. Maybe, maybe not.
     
  17. Marcus, thanks for your post.
    1) "The proper methodology would have been to 'balance' the exposure difference (between the mk2 and mk3) and set the exact same exposure to each camera (using manual) for each subsequent shot."
    I actually ran the test three times, and I started off doing exactly this. This is easy to say from a theoretical standpoint, but the downfall of this method is that when it comes to side by side comparisons, the exposure difference is too distracting to make a meaningful comparison. It was particularly problematic in the shadow areas, which are of course very important. So this was not an option.
    Another thought I had at the time was to reduce the Mk ii exposures in DPP to match those of the Mk iii. This should improve the apparent noise in the Mk ii images which would weight the results in favour of the Mk ii, meaning that any possible overall Mk iii "improvements" would be on the safe / conservative side.
    2) "to make for more effective NR during the RAW to JPEG conversion process. Altering the setting from the mk2 to mk3 only further contaminates the data."
    Not really, the effect on noise is miniscule and negligible for the purpose of this test. I pixel peeped for ages to confirm this so I don't see this as an issue.
    3) " Instead, one could view @ 100% ->200% (for clarity) side by side, and take a screen shot of both side by side for comparison."
    That sounds like a fantastic idea and if I had thought of it I would probably have done it this way.
    Marcus, one can nit pick until the cows come home and while the test is not perfect, and having spent many hours mulling over the results whilst conducting the test, I believe the results are a good indication of the difference in performance. I agree that the issues raised (and thank you for doing so) will have aslight impact on the results, but my estimate is that what we see is a good indication of what we can expect.
    Lastly, I think that adding the RAW-JPG conversion algorithm into the equation is maybe a stretch too far. Unless you have insider knowledge, we have no idea how it is done in DPP and since it is a manufacturer supplied tool valid for both cameras, we need to assume that the conversion process is "equal" as we don;t have much other choice. But to negate this, you good idea of screen captures would get around this.
    Many thanks again for your detailed comments. I may re-run the test with a tweak or two and see how it goes.
     
  18. I have observed this issue as well. It seems that disabling "moire' reduction" somehow tricks the software not to apply what appears to be a heavy luminance NR to the 5D3 CR2 files to make them appear cleaner.
    However even with disabling the moire reduction, slightest change even in the chroma NR slider will push the software to default back to the heavy NR which smears the details.
    I have notified my contact at Canon about this. At the mean time please take some time to report this issue so they can fix it-thanks.
     
  19. Arash thanks for sharing that bit of information. I'll look into it.
     
  20. Hi everyone! I just got my Canon 5D III yesterday and was so excited I just had to take some pictures to try it out and give a little review of what I have found so far! I have to say… compared to the Canon 5D II I am very impressed with the camera so far. The AF is unbelievable and far exceeds the MK II and MK I! The Canon 5D III is super accurate, fast and will focus in the dark – literally…. The AF is definitely on level and may even exceed the 1D series. I have some images that were shot at ISO 102,000 and 51,000, but I haven’t had a chance to process them yet. I wanted to put something up for others to see, so I am posting images from ISO 12,800 to begin with and ISO 6,400 to follow. I will be adding more images and expanding my hands on review in the coming days.


    Click on the following link to see RAW files and Jpeg Examples.

    http://johncarnessali.com/lens-tests/3201
    00aDlx-455061684.jpg
     

Share This Page