With proper exposure, and watching your highlights you can get great results. "BUT" I still get contour banding on out of focus areas, sky, and graduations in light falloff. As far as i can tell, only a higher bitrate will allow the codec to not through out all that graduation. So if it is 150 mbs I take it, to get smooth images. We will see if 100mbs+ will even be possible on a GH-2?
rozor is right, however higher bit-rate will help somewhat by allowing codec of the camera not only to concentrate on the areas of high frequency detail in Lightness and color channels (more attention is usually paid to lightness information channel) but also on the areas of low frequency like the sky. Sky is the area of the continuous tone but its appearance does not only comes from lightness info: Blue/Yellow channel is also very important. You won't necessarily get film like appearance of smooth tonality change with a higher bit rate but you should certainly get gradation steps that are more smooth themselves and look more like a film to digital transfer, rather than digital acquisition to begin with due to the reduction of the macroblock patterns. That being said if sky gradation is really important to you try so called expose to the right (using a histogram) method and then color grade to bring down you mid-tones and shadows if they appear to be too bright. By exposing to the right you will be able to squeeze more of those 256 values per channel available where it is important to you. Exposing to the right is a good practice for any kind of subject matter anyway, but you will have to color grade in post.
I've seen this gradation step issue, although normally it's almost unnoticeable. In one case though the steps were actually moving across the picture (in one direction, diagonally) where there was a flat area of white wall in the image. This was with a manual lens and fixed exposure, so the exposure wasn't changing. Therefore I'm not sure what was happening there and why the steps were constantly moving across the image. That's the only time I've really noticed it, and it was in a shot with a single bright area (a window) elsewhere in the frame, so maybe that was affecting the exposure enough to make the gradation steps visible. Still don't know about the movement, though.
Dkitsov's idea about changing exposure is an interesting one - not one I've tried. But then, I've not seen the gradations on the LCD, only afterwards, probably because the LCD seems to go to a slightly lower quality during recording, so I wouldn't have been aware of it at the time I shot the footage. But something now to watch out for in particular scenes.
@dkitsov, you mention that this is essentially a LF effect - which I suppose it is - so I wonder if there's any post-treatment (chroma blur?) which can improve it?
Also, do you think that by changing the in-camera contrast settings, it could help to use more of the values available where it matters?
Finally, Vitaliy, any chance of the hack making it possible to record 10 bits per channel? One can always hope!!!
Edit: have looked again at the shot with the moving banding. I was very slightly changing exposure during it (just playing around with a new lens). But in a more conventional shooting situation, I could see how you might get that effect of changing exposure with lighting / shadows that vary during a shot.
I think its like dkitsov explained it. Dark saturated gradients like a blue sky at dusk will alway have banding issues on a GH2/AVCHD/h264. There are h264 profiles that support 4:2:2 chroma and more than 8bit. I don't think that they are implemented into the AVCHD encoder inside the GH2. A higher bit rate could help to avoid compression artifacts for high detail fast motion shots but it wont be that helpful for color banding issues.
Color is a real issue on the GH2. To me, this greenish tint and the banding are a way more often experienced problem then rolling shutter and aliasing on the Canons I use. And its very hard to fix the green GH2 look in post. You have to plan your GH2 shoot even more carefully than you have to with the Canons.
//Moved from original location// Exposing to the right saves the trouble of using LCD for anything but framing (EVF is better for focusing). the point of exposing to the right using the histogram is precisely that ( I would also turn on the highlights blinking) you will get a good exposure without picture confirmation. Good exposure in that sense that you record maximum tonal information, the image might look washed out unless you grade it in post. Low contrast Nostalgic or Cinema are always good. Though honestly I rarely see gradient problem with what I get out of GH2. For me the real reason to expect the hack is so that codec would give more attention to the areas that are below the mid-tones so when I try to grade them up if needed there would be actual detail and not just areas of flat fill with squared edges. There is no fix for sky banding short of 8bit>16bit>color key for the sky color>blur keyed region>apply noise to the keyed region only and it must match unaffected areas or will look fake. And it is very tedious and slow, and painstaking. So no - there is no fix. To answer for Vitaly no probably there is no easy way that 10bit color can be enabled (as he only tweaks existing firmware and does not write new one, it also may be hardwired so no firmware may change it)
Does the Canon 5D2 handles banding better than the GH2? I've sold my 5D some time ago and I can't remember whether there were much banding.
My GH2 banding (posterisation) is especially pronounced when watch on a 50 inch Panasonic VT25 plasma panel. On a 24" Apple Cinema Display, the banding was less of a problem because I reckoned it was heavily dithered.
Hey - my 2 cents: when the 5d2 first came out I scoured videos online to see if it was worth buying, and it was then that I first noticed banding in DSLRs as it was obvious in quite a few 5d2 videos I saw. I've seen banding in my GH2 in one shot. So I guess they're similar.
Edit: 2 shots. Both of them were areas of plain, light (not white) wall.
Exposing to the right is not a euphemism for anything. It is a name of a technique that allows one to employ the maximum dynamic range of a linear sensor. While the technique came from the digital photography world it is especially important for digital SLR video as we have only 256 step of gray to work with as opposed to possible 4096 steps possible with some still cameras. There is a good write up on the exposing to the right on the luminous landscape. http://www.luminous-landscape.com/tutorials/expose-right.shtml
Thanks for that link. Really useful. While I've intuitively tried to do this anyway when shooting, the explanation makes huge sense to me. Relating it to audio, it's like recording at a high enough level that you hit max without distorting anything (and you can get away with the odd bit of clipping on transients). So you use all the bits you can. It's probably an exact equivalent.
A big plus is that the GH2 clips in a pretty good way, so there's no excuse not to keep that histogram going all the way up to the right hand side.
Since we can't increase the sample size, the only other way the banding could be improved would be to have more dithering. Basically you need 1 bit's worth of noise prior to being encoded into 8 bits per channel, which then effectively removes the banding but at the expense of image noise. We do the equivalent with digital audio, but in audio we have at least 16 bits to play with so we're not sacrificing significant S/N ratio.
With audio, some cheaper systems don't add dither because the self-noise in the system is enough to act as dither prior to A-D conversion, so if there's a video equivalent, maybe the fact we're noticing banding in these cameras is down to both the low (8-bit) sample sizes and also the low sensor noise / lack of dither. I'm only speculating - audio I know about, but not video! Anyone know?
Just did a bit of experimenting to test how the sensor noise might act as "dither" to affect banding. Here are two shots and I've deliberately not used all the dynamic range available (the histogram was going up to about halfway, giving levels of approximately 0-155). I wanted to see if I could deliberately generate banding and see if high ISO (hence more noise) would smooth it out.
The first photo is a still from a sequence shot 50i at 160 ISO. The second photo is a still from a sequence shot 50i at 2500 ISO, with the lens stopped down to bring the histogram back to within a similar range.
To see it, you will need to click on the images to enlarge (you have to be logged in to do that). It's not a totally scientific test, but I can see banding in the 160 ISO image on the wall to the right of the harp, but at 2500, the system noise seems to smooth out the steps in the image.
By the way I also tested at 24p and had similar results. It seems to confirm that manufacturers rely on at least some system noise to add dither, and if the system noise is low (for example, low-noise sensor) then you see the "stepping" associated with transitions from one level to another. It's especially visible if you don't use all of the available dynamic range when shooting - in other words, you don't get the camera histogram as wide as possible. Hence, to avoid it, just try to use the full range available when shooting.
I'll stop being boring now. Later on I'm hoping to post some steadicam footage of my kittens climbing the stairs in slow motion, set to music. Just kidding!
Good test. I would however caution people, who do not realize that this discussion is somewhat academic, against shooting at higher than needed for good exposure ISO rating. As we are dealing with a fixed sensor that has only one real ISO rating and whatever ISO numbers you set in camera change the gain of the sensor - the dynamic range of captured image is being reduced.
Is that situation true of film as well? As you go higher ISO does the exposure lattitude narrow and noise increases? I guess it doesn't really matter since no one uses film anymore. Just curious. I never played with varying ISO films to understand the tradeoffs.
How are you going to get the kittens to walk in slow motion. :) don
donf, This situation is not true with film. It is actually the opposite but for completely different reasons. For all intents and purposes, with film it is impossible to change the dynamic range of the emulsion - you have to use a different sensitivity film to do so. (changing the development time of the film (push/pull) doesn't count). The lower film sensitivity the lower it's dynamic range, the higher the film sensitivity the higher it's dynamic range. Although for certain reasons (ie photographic emulsion non-linear response to light) the effect of ISO on the dynamic range is not very pronounced.
I am using film, i love it.there is no digital camera can shoot 4x5 inch.that is why i still find diffecult to like digital imaging. It is also amazing new generation never going to use analog system. They will born with digital and grow up with it.future visual esthetic will shape around it.
What do you think about making topic for 4x5 and other medium and large format film cameras? It is interesting to see real difference and talk about it.
Anytime one of the young kids on any production I am involved with ask me about what can they do to become a better filmmaker I say "Go buy a cheap 35mm camera with a 50mm lens and take an introduction B&W photography at the community college"
>Go buy a cheap 35mm camera with a 50mm lens and take an introduction B&W photography at the community college.
I do not agree with you. As I see no point of advising novices 35mm film cameras. Learning something is always result of trial and error. In digital it simply happens faster.