1. Reminder: Please user our affiliate links to get to your favorite stores for holiday shopping!

GH3 = fake ISO12800 and ISO25600

Discussion in 'Panasonic Cameras' started by deejjjaaaa, Dec 5, 2012.

  1. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
  2. danska

    danska Mu-43 Top Veteran

    945
    May 21, 2012
    Portland, OR
    Joe
    Gotta love when someone spends a bunch of time creating a graph and then doesn't label it completely. It seems to happen so often in this world of photography and I just don't get it.

    I see that this probably your creation, but the link below the graph has many links that a lot of people aren't going to spend the time to go through. It would make sense to include more of a description of what is being shown.
     
  3. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    that is only for people who are interested in low level details, not every Buonarroti needs to know how his chisel works.
     
  4. Amin Sabet

    Amin Sabet Administrator

    Apr 10, 2009
    Boston, MA (USA)
    The press release suggested native 12,800 and extended 25,600, but that never seemed likely, especially with all the evidence that it uses the same sensor as the E-M5.

    Your graph is interesting. I would expect an extended ISO 12,800 to have the same DR as an extended ISO 25,600 for a given exposure (since the sensor analog gain is the same for both), but most test methods give the high expansion ISO a lower exposure and result in a lower measured DR. I see that you are using Bill Claff's protocol, and virtually all his results show lower "photographic DR" with increasing ISO, even into the expansion ISOs.
     
  5. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
  6. Amin Sabet

    Amin Sabet Administrator

    Apr 10, 2009
    Boston, MA (USA)
    That's always the case for expansion ISOs, no? The difference is that most medium format and Sigma Foveon cameras have just one native ISO and the rest are "fake" whereas most DSLRs and mirrorless cameras have a range of native ISOs and a few "fake" values outside of that.

    For example, the original Canon 5D had native ISO up to 1600 and a "fake" (by your standards) ISO 3200. Claff's data shows a 1-stop lower photographic DR for the 5D at nominal ISO 3200 compared with ISO 1600, yet the camera is doing nothing different at those settings other than the tag. You could get an "expansion ISO 6400" on the 5D by setting it to 3200 with -1EV exposure compensation, and that RAW file would contain the same data and DR as an ISO 1600 file shot under the same circumstances with -2EV exposure comp.
     
    • Like Like x 1
  7. zapatista

    zapatista Mu-43 Top Veteran

    668
    Mar 19, 2012
    Denver, Colorado, USA
    Mike
    Amin,

    How dare you explain it it plain terms so it actually makes sense.
     
  8. arentol

    arentol Mu-43 Veteran

    269
    Jun 29, 2012
    The manual says that 25600 is only available when Extended ISO is set. In most cameras these days all ISO's available without turning on Extended ISO are usually hardware ISO while all extended ISO's are usually software ISO's. However, there is no rule that says this is how it has to be, and I don't know that Panasonic ever claimed that 12800 was hardware ISO.

    So this is good to know, but it isn't a big deal at this point. Also calling it "Fake" is adding an unnecessary and unwarranted negative connotation to a common industry practice.
     
  9. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    no, not always what manufacturer calls expanded/extended ISO is done just through tags (like MF) w/o actual raw data modifications... for example take Pentax K5 (ISO range = 100 to 12800 and extended ISOs = 80, 25600, 51200) and compare.
     
  10. Amin Sabet

    Amin Sabet Administrator

    Apr 10, 2009
    Boston, MA (USA)
    That makes sense. You can define an "expansion ISO" as an ISO done through tags (how I define it) or you can define it as an ISO which is beyond the normal use or auto ISO limits for the camera (some manufacturers using it this way).

    Either way, it's not surprising that ISO 6400 would correspond to the max analog gain for the GH3 sensor. What is surprising is that you would get results using Claff's methods that show the same photographic DR for those three nominal ISO values, because his graphs don't show that for other cameras which offer ISO values that are done through tags (like the Canon 5D at ISO 3200).
     
  11. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    and in GH3's case "claff's" (his instructions) data shows the same DR for ISO6400 through ISO25600, that's the difference... and I 'd assume you actually did not see the raw data from 5D raws that you are talking about, did you ?... did you actually see the difference between how it is done through tags in MF and not just through tags in 5D as illustrated by DxO - see ISO sensivity graph for 5D @ DxOMark - Canon EOS 5D and see ISO sensivity graph for example for Phase1 P45+ @ DxOMark - Phase One P45 Plus
     
  12. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    true... however there is a difference between doing that just through a tag (instruction for a raw converter) and applying gain digitally (raw data actually differs).


    they way it is done is not common... do an exercise - find how many cameras does that through tag only and how many through digital gain - adobe dng converter and checking for changes in baseline exposure tag (that is the instruction for a raw converter on your computer to apply the gain out of camera - basically to apply exposure correction hidden from the end user) will help you.
     
  13. arentol

    arentol Mu-43 Veteran

    269
    Jun 29, 2012
    My understanding, which apparently was wrong, has always been that it was always done through digital gain in camera. I didn't know there was another option. So you are saying that the GH3 method is "fake" (relative to other methods???) because it is not done with digital gain?
     
  14. Amin Sabet

    Amin Sabet Administrator

    Apr 10, 2009
    Boston, MA (USA)
    I think that's a peculiarity of DxOmark testing. I've done my own comparisons of 5D ISO 1600 files pushed one stop vs ISO 3200 files, and they were the same as far as I could tell. Plenty of other people analyzed the RAW files and found the same thing. Likewise, ISO 6400 and up are "fake" on the 5D II, yet DxOmark indicates that all of them are real.

    To an extent, it is immaterial whether these high ISO values are "real" (analog gain, ie sensitivity) or "fake" (digital gain, ie pushing). With most cameras the higher analog gain settings have lower read noise, so using fake ISOs will slightly increase noise/banding in deep shadows while adding to highlight headroom in the RAW files. With some of the newer sensors, read noise is similar across ISO values, so there is little to lose by going the "fake" route.

    There's an interesting older but still relevant discussion of how the 5D II ISO 6400 and up ISOs are fake (and why that doesn't matter much) here: http://forums.dpreview.com/forums/post/29828654
     
  15. goldenlight

    goldenlight Mu-43 All-Pro

    Jan 30, 2010
    Essex
    John
    I'm not interested in whether the high ISO values are "real" or "fake."
    It's how it actually performs in practice that matters.
     
    • Like Like x 1
  16. jloden

    jloden Mu-43 Hall of Famer

    May 15, 2012
    Hunterdon County, NJ
    Jay
    Well, that's it then. I guess I'll go return my GH3 ...to its case until the next time I plan to use it :biggrin:
     
  17. jnewell

    jnewell Mu-43 All-Pro

    Jun 23, 2011
    Boston, MA
    It only works badly if you are...under...a...bridge...:redface:
     
    • Like Like x 2
  18. meyerweb

    meyerweb Mu-43 Hall of Famer

    Sep 5, 2011
    We have a winner!
     
  19. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    and a nail



     
  20. deejjjaaaa

    deejjjaaaa Mu-43 Rookie

    21
    Aug 28, 2011
    first of all you do not understand that there are 2 ways to "fake" - through tag only (raw data itself is intact) and pushing is left to raw converter - that is the situation with GH3, MFDB (some of them - I can't say whether all MF are doing this way) and through digital gain applied before raw file is written (hence raw data actually different - whether the digital gain is applied by hardware or there is actually a calculation directly coded in firmware is irrelevant for now) and raw converter does not need to deal w/ that...