Personal View site logo
GH2 settings for low light shooting
  • 64 Replies sorted by
  • @LPowell Therefore we should remain in "the finest quantizing state" even though the picture is darker in order to preserve details in shadows. Is it possible? I suppose that Panasonic and some other producers have chosen such path (to increase macroblocks in darker areas) in order to get better results on noise tests. However, is increasing macroblocks in darker areas a part of AVCHD specifications or does it depend on a vendor?

  • @LPowell: The problem is here "finest perceptible degree". We all do grading later on the footage which changes the perceptible threshold dramatically. The consumer won't do grading. Perception is logarithmic and an increase from 5% to 10% in levels is the same as 50% to 100% in levels. So, if in grading we change the blacks just a tiny bit all the artefacts appear because the eye is very sensitive in the blacks. @crunch: the MPEG-4 spec won't specify which DCT coefficients have to be thrown away (and I read it!). It's entirely up to the manufacturer to make this decision. That's the reason why the blacks look so different between Canon, Panasonic and Sony. EOSHD has shown this recently with the Sony NEX: http://www.eoshd.com/content/7862/sony-nex-fs100-as-a-dslr-alternative-first-impressions Personally I would say black encoding from my experience is worse with the GH2 and best with Sony cameras. Canon is somewhere in the middle.

  • @crunchy @berndporr

    Here's the technical shortfall inherent in the GH2's 8-bit sampling of dark images, as explained by Charles Poynton, in "A Technical Introduction to Digital Video":

    "In an image coding system, it is sufficient, for perceptual purposes, to maintain a ratio of luminance values between adjacent codes of about a 1 percent... For luminance values below 100, as the code value decreases toward black, the difference of luminance values between adjacent codes becomes increasingly visible: At code 25, the ratio between adjacent codes is 4 percent, which is objectionable to most observers. These errors are especially noticeable in pictures having large areas of smoothly varying shades, where they are known as contouring or banding."

    http://www.poynton.com/PDFs/TIDV/Gamma.pdf

    While PTool gives us the ability to use finer quantization factors to encode each macroblock, nothing can be done to improve the coarse 8-bit sampling of the darkest areas that are being quantized. Even if the quantization was flawless, we'd still be stuck with coarse 4% steps between the darkest shades of gray.

    When you raise the illumination of shadow details in post-production grading, the coarse sampling becomes much more noticeable. In addition, the eye's insensitivity to pure blue and purple hues makes those shades appear darker than other hues encoded at the same level. The crude sampling also degrades the saturation scale, making brightened shadow details look blotchy and over-saturated.

    The reality is that when enhancing shadow detail, you're attempting to salvage low-resolution data, and it often requires filtering as well as color correction. One approach is to add dithering noise to obscure the edgy digital artifacts. A technique I favor is to desaturate the RGB color channels, linearly rather than perceptually. That restores a more natural-looking gray tone to the shadows and eliminates the spurious contrast in blue and purple hues.

    Ultimately, however, the best way to enhance shadow details is to lift them out of the dark digital noise floor in the first place. Since the GH2 lacks an in-camera pedestal adjustment, a practical way to do this is to use fill lights to pull the shadows out of black into gray. Outdoors, it's best to select shaded, low-contrast backgrounds, rather than areas in direct sunlight. When grading an image, pushing grays down into black will always produce higher quality results than lifting the darkest shades into gray.

  • <

    p>PETER "KID

  • Sorry to dredge up old thread but its 7 months later and wondering if things have moved on and if @driftwood would still recommend cluster v2f for high ISO (1600+) low light shooting?

    OR anything else I should try (I've got 45mb/s cards but I'll buy a 95mb/s one if it'll make a big difference). Anything to limit the macro-blocking and chunky shadow noise. I dont use long takes.

  • @gethinc, in my experience, Lowrider at iso 800 to 1600 performs well, as does flowmotion 2.02 mjpeg. shooting above 1600, regardless of hack, leaves a lot to be desired

  • Thanks @hay, I just loaded the sanity 5.1 patch, correctly moved to 2500 (ie from 3200 not from 5000) shooting at 720 50p and I thought it looked OK. Compared it to nikon d800 at 2500 iso and I'd say it was better.

    I'll check out lowrider, thanks :)

  • I know this thread is ancient but I didnt know where else to post...So, I've been shooting with the GH2 commercially now since early last fall and still haven't found ideal settings for lower light situations that get rid of digital noise/ weird banding and odd artifacts that showup at ISO higher than 320. The other day I decided to try out Sanity 5.1 (I've tried 5-8 differenct hacks most of which were all great in various ways in daylight or well lit scenes..none of which where acceptable in low light). I shot some night scenes last night using a rented Voightlander 17mm f.95. This is the second time I've rented this lens. After the stunning quality (lack of any noise what so ever) I think there is really no other option than shooting with this lens (or its 25 or 45mm equivalent) in situations that are NOT full daylight or well lit scenes. I'm also begining to think that my camera may have a funky sensor that makes it break up on ISO higher than 320..?? Curious as to other peoples experiences with low light issues. I've researched tons but not discovered any real world practical solutions....meaning solutions that you would use with clients on the job as opposed to personal shoots or for fun..Sorry for the run on sentences..

  • @LiftCM: You're right. I also experience visible noise above ISO320, no matter which hack is used.

  • The GH2 sensor can't do any better than this I don't think, regardless of the firmware. It's actually impressive as is, but then I've never had a F0.95 lens or shot at ISO320 at night. Short of switching to a 10bit 422 camera the most I can suggest is doing any image manipulation after conversion to a mezzanine codec like Cineform. The advantage of Cineform, being a wavelet codec, is that you can playback at 1/2 1/4 resolution and get more going in realtime on an older system.

    My workflow is like this: Install latest Cineform Studio, it has a free encoder.

    • trim out unwanted takes on multiple timelines (latest Premiere Pro CC 2014.1+)

    • Consolidate project and transcode to Cineform (there's a dialog that allows for selecting which codec to transcode to on consolidation).

    • open the newly created project and work from there.

    What this does is upconverts 8bit 420 to 10bit 422, and interpolates the chroma channels; essentially you get to an alternate representation of the data, without the original H.264 macroblocks from AVCHD. You don't have more info, but what you have is less prone to breaking up on manipulation. I don't have a handle on the wavelet transform to explain it in any detail, just see for yourself and confirm it with your own eyes.

    From this point I:

    1) Clean up with NeatVideo

    2) Color correct

    3) Add back noise :) Usually (but not always) I use FilmConvert to finish - what that does in one pass is not only bakes in a specific film LUT, but also adds quality grain (dithering) which you can dial in. I've saved quite a few night shots with banding this way.

    Have a look at this short, shot with a GH2 and a Canon 24-70/f2.8, ISO 800/1000/1250, with the post workflow above:

  • @radikalfilm...thats quite a workflow, but then again your footage looks really nice esp considering the low light conditions. This is exactly what inspired me in the first place to jump aboard the GH2 wagon despite most DPs in NYC thinking it was a joke. I've saved a lot of time and money having to rent Cameras the last 6-8 months in addition to shooting a lot more on my own for fun either of my family, stuff for our sales and marketing efforts or my own documentary.. I tend to think as one of the posters mentioned, there may be more value in finding the "ideal" presets and combos in the film modes and in relation to various lenses used than anything else. ie what settings work best with Voightlander vs Canon Cine vs consumer or cheap FD mounts... Also your comments about it still being an 8-bit sensor holds very true. I work with Final Cut pro. I havent used Cineform since my Sony Vegas days. So I tend to just import direct into FCPX and let it convert to prores. But I find it interesting upconverting to 422 not that you gain or add anything..(back to my addage..garbage in garbage out) but if it helps to hold up the image in color grading...that may be an interesting workflow/theory.. Curious what the science behind it is.... At the end of the day I think as long as fast lenses are used when there is inadequate light according to the world of the GH2 sensor...Its a camera that will hold up to my Red and Canon rentals..esp considering the costs incurred...my 2 cents

  • From my experience, you can shoot up to 640 ISO and transcode the footage to ProRes 4444. I talked about low-light shooting in one of my videos. I use Voigtlander lenses and Neat Video in post to kill off any small noise left behind.

  • @Gardner-I've been meaning to watch your video with the sound on..!! But this footage looks awesome..But then again thats all Voightlander all the way! I'm obsessed with those lenses. Have you tried the 45 yet? I am looking to purchase the 17.5 but need to rent the 25 to see how that focal length looks on interviews which I do a ton of..The 17 is just too wide for an interview but I love taking it out a night! My feeling was that even at 640 I was getting too much grain...But Im going to run Sanity 5.1 for a while as I'm liking it so far..and its pretty reliable in terms of footage showing up at download time.

  • @LiftCM I haven't tried the 45 yet, I've been using the Nikon 50mm 1.2. The 25mm is excellent for interviews though and have you thought about trying other hacks since you've been getting grain? I assume you've been getting unusable grain if 640iso doesn't work for you.