Static - All four settings are very close to each other, however I feel 66M AQ3 holds slightly finer detail than the others.
Motion - Once again, very close, but there are differences. 132M AQ2 holds fine detail the best, followed by a tie between 66M AQ2 and 66M AQ3. 44M AQ4 brings up the rear by a whisker.
COMMENT: When it comes to holding fine detail, the static picture is more important one. The eye has time to linger and pick it up. The 66M AQ3 holds the edge here and personally I would like to use it as my standard setting. But there's a problem - for the first time since using the hack, I had an image break up. It lasted about five seconds and then corrected itself. Could a stable version of 66M AQ3 be made?
Lastly, cosimo_bullo requested that I shoot still JPEGS during the tests, which I did. I'll post them after the clips.
I think you misinterpreted my intent. I think what Ralph is doing is absolutely tops - some of the best testing so far. If anything I was trying to confirm his findings analytically. Come to think of it, his tests largely confirm what the codec should be doing theoretically! I also understand the value of rendering noise better.
@Ralph - a small request (one in a series...!). While you're doing your HDMI testing, can you also include a 1080 jpeg photo taken with the same settings. I'd like to see how different the jpeg looks from the HDMI signal and see if that could be used as a low-tech baseline for future tests for those without HDMI capture.
@cbrandin - These are interesting points, but they're very theoretical, and, thanks to Ralph, we no longer need to operate in a theoretical mode. Everything should be tested, even if it seems the numbers don't add up.
"It looks like after ISO 1600 the codec will outperform the sensor. Actually, considering Nyquist limit, etc... it may outperform the sensor at 800."
The codec - other than the one night time 132M test - is clearly not outperforming the sensor in any mode. There's a lot of valuable info/noise that is being dumped, info/noise many want to keep if at all possible. Remember, this is art as much as science.
And, to be clear, I SO value what you are doing, Chris, just want to be clear that Ralph's tests have changed my thinking dramatically as to what's possible.
I should point out that I am not saying that there is no reason to shoot at an ISO below 800. Rather, I am saying that above 800-1600 the sensor limits quality - not the codec, and the ability to improve anything is pretty limited.
Nice chart - thanks. It sure looks like ISO 800-1600 is the breaking point - as the SNR there is pretty close to 0.4% (accounting for 250 steps, or so).
The higher ISO, the lower DR, the lower SNR. True.
But if high ISO can't be avoided, I'd prefer finer noises. It would give more information to denoising plug-in during post-processing. But I wouldn't go higher than ISO 1600. Some reported noise differentiation starting from ISO 640.
Noise is band limited. Remember that noise is mainly a function, which you can see as a signal (if you plot the function you get a curve which is your signal). In signal processing it's possible to take a signal and convert it from spatial domain to frequency domain. This operation gives a result from which it is possible to see the different frequencies that a signal is made of. Don't worry too much if you don't understand yet what it means to go from spatial to frequency space (it is off topic here but you can check the lesson on Fourrier Transform). All you need to remember is that the noise function is potentially made of multiple frequencies (low frequencies account for large scale changes, high frequencies account for small changes). But one of these frequencies dominates all the others. And this one frequency defines both the visual and frequency (if you look at your signal in frequency space) appearance/characteristic of your noise function. Why should we care about the frequency of a noise function? When the noise is small in frame (imaging an object textured with noise far away from the camera) it becomes white noise again which is a cause of what we call in our jargon, aliasing. This is illustrated in the following figure (xx).
If it is possible to completely disable the noise reduction, we might see more finer noise which might become white noise or aliasing. It will look better only if it goes through post-processing to remove the noise. Of course the consumer camera manufacturers wouldn't want that as most consumers would prefer seeing more coarse grained noise. I meant for video. The white noise aliasing would freak out most people who are not into videography.
It occurs to me that with high ISO settings you lose a proportional amount of dynamic range. Thus all you will get with higher quality settings is better rendering of noise - the camera is limited in it's ability to handle gradations because of the higher ISO value and the effect that has on dynamic range before the codec even processes anything.
Not that it is a bad thing - just something to consider.
I often use stills to identify what the camera's limits are vs. the codec's.
@stonebat, I think so, yes. Finer, more detailed grain has a more filmic feel - and if you want to clean it up, you're starting with a much more accurate source than if NR has already been applied or compression has smoothed it over.
"Also, be aware if you can see more finer noise, this means you are able to see more fine detail which results in a better detailed image."
+1
And let's not forget, the GH2's ability to resolve fine detail is one of the things that sets it apart from every other DSLR on the market. This is it's strength. So it makes sense to maximize it. cosimo_bullo is using it in conjunction with Red 4K. That speaks volumes.
Yep, cool, I get that, you're right. I must admit I was playing a lot with the 132M AQ2 GOP 3 yesterday and was completely gobsmacked by it's stability more than anything. I actually don't like the look of motion with GOP 3, but as said I would expect a lower GOP to dramatically improve noise rendering (and therefore NR in post). I don't think I'll ever use it.. but I know its there, if I ever have to do that one shot where I know I'll have to do a lot of work in a deep shadow area. Say, if I wanted to do a VFX shot where I'll have a CG element travelling from deep shadow to light or somesuch. But it's not something I'll reach for everyday.
132 definitely has finer noise reproduction. No question about that. Some people will find it useful. But shorter recording time and possible in-camera playback issue. I'm sure some people wouldn't mind that.
We are learning more pros and cons. I like the details. Hopefully the same testing can be done when we get new PTool as VK and Chris are working on improving the codec.