Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
GH2 ISO noise bug at 320, 640, and 1250
  • 62 Replies sorted by
  • I read about the bug issue awhile back and i taught it was only on high iso's coming from a canon its strange to hear a iso problem on low isos

  • so theres no issues with iso 160 i had seen a guy that said he did an interview at 160 and his had alot of noise. I also hear that if you go over fstop 8 that it introduceses noise

  • ETC mode is subject to a kind of electrical impulse noise when recording AVCHD video. It appears as intermittent horizontal streaks in certain brightness ranges, usually the darker ones. Here's a good example: vimeo.com/59801685 (sky at 00:08).

    The strength of the noise and which brightness ranges the noise appears in depend on which ISO setting you have chosen. Testing has shown which ISO settings are best and worst with respect to this noise in ETC mode. A good setting is one with weaker noise that appears across a narrower range of brightness levels. When I say "320-from-160", I mean setting the ISO to 160 and then to 320.

    • 160: bad. 200, 250, and 320-from-160: increasingly worse.

    • 320-from-400: good. 400, 500, and 640-from-320: increasingly worse.

    • 640-from-800: ok. 800, 1000, and 1250-from-640: increasingly worse.

    • 1250-from-1600: ok. Higher ISOs are increasingly worse.

    So the bottom line is that 320-from-400 is the best setting. ISO 400 is the next best: use 400 if don't trust yourself to avoid the ISO noise bug each time. Avoid 160, 200, and 250.

    It would appear that 160, 320-from-400, 640-from-800, and 1250-from-1600 are base ISO settings that put the sensor in a particular configuration. Higher ISOs use the same sensor configuration as the base ISO below it, and just increase the gain. Higher ISOs show the same impulse noise as their bases, but at a proportionally higher output brightness levels.

    The impulse noise is closest to black in the base ISOs, and can be somewhat reduced by pushing down levels in post.

    The noise does not happen in MJPEG mode or when the camera is not recording. It seems that the noise is being generated internally.

  • Increasingly I find myself sticking to ISOs 200, 400, 800, 1600 for everything. The divisibles seem to work best for me.

  • @driftwood I've also found this to apply to me as well (especially on the GH3)

  • Wasn't there a bunch of people talking about how great 160, 320, and 640 are for getting best image quality on GH2 (assuming iso bug properly avoided)? Do you guys prefer the 200, 400, 800, 1600 for image quality or is it a choice to avoid the iso bug?

  • Is it possible to turn off 1/3 ISO increments?

    @matt_gh2 I think it's subjective matter that multiples of 160 gives the best image quality, but we do know for sure that there is the noise bug. I'd stick to multiples of 200. Keep it simple.

  • @milada Thanks for the test. Is anyone else suddenly wanting to go with 640? On my screen that looks way cleaner than most of the lower ones.

  • That test is not representative of real-world images. It only shows one input brightness level.

  • oh, good point. What test do you like for best analysis? This seemed like a good option to me in that I always notice noise in the blacks and darker colors

  • http://www.maik.de/wp-content/uploads/2009/06/graubalken.jpg

    http://www.iris-digital.org/wordpress/wp-content/uploads/macbeth.jpg

    More info and custom color charts for Neat Video: http://www.neatvideo.com/profiling.html

    I use these to create noise profiles for Neat Video. Must admit that once done proper I can not distinguish 160 from 320 from 640. Anything higher will result in more visible noise you have to let trough, I usually use 1250 with good feeling in low light and up to 2500 with a little strange feeling in my stomach. (for anthing that must look clean at the end.)

    In the end if you are not heading for the big screens at a certain point pixel peeping for noise does not pay of.