> ...If the M8 is truly higher resolution than the R-D1 (duh), then your images are reversed...
The histogram screen shots were at different magnifications: the M8 one happened to be zoomed in closer. Sorry if this lead to confusion. It has no significance. If you go to the original link you can see that the M8 really does blow the RD1 away in resolution.
What the highlight (sky) peaks in the histograms do tell us though, is that (a) the RD1 sky is less saturated since the 3 colour peaks are closer to each other, and that (b) the M8 may have a tiny bit more noise in the blue channel since its blue peak width is slightly wider. But there may be many factors, including the changing colour of an evening sky.
The main hump is more interesting as it shows the near absence of greens in the M8 shot.
The very sharp blue peak near zero on the M8 doesn't seem right in this image. It suggests that black is not being rendered as black in the blue channel. The quest for dynamic range can go too far. It is possible that this is connected with the sludgy/noisy shadows. Suitable choice of black point in the blue channel should fix it.
I know nothing about these stats/maths operations to show 3D look, etc.
But I do trust my eyes and the M8 image if anything is much more threedimensional than the R-D1's.
When I opened the full images on my monitor it was very apparent also to my girlfriend.
Note that the R-D1 image was upressed (PS bicubic) to M8 resolution, hence actual pixels are slightly softer.
This is to give a comparable file size for printing,
Anyway M8 colors were off, maybe more pleasing for the back-lit leaves on the foreground, but not very accurate otherwise.
It was one of the first M8, firmware 1.06, without IR filter.
The R-D1 though didn't manage to differentiate the greens in the shadows, I see better tonal differentiation in the M8 (even if the colors were more brownish).
I now have a new M8 (firmware 1.09) but I still like R-D1 colors better (we need a good M8 profile), otherwise there's no comparison between the two cameras.
I initially thought to use the R-D1 as a backup, but now I think I'm gonna sell it... I rarely if never use it.
brainiac wrote:
Jack - I am surprised that someone who has "written resizing code that is used in the industry by large format printing professionals" wouldn't spot that those peaks are the sky, and that as flat colour at an undefined but very large distance, they are unlikely to convey much 3D info ;-)
No, but they should give any true brainiac a clue as to what was going on in the subtle midrange (very subtle in this set of images) that is driving the relative highlights in the lower tones, which in turn drives the human perception of dimension... The factor you did not take into consideration is the ramp on the curve in the midtones -- the longer it is, the easier it is to visibly distinguish luminosity values, again important in determining dimensionality. That stat does not get factored directly into your std dev between colors and you need to look at luminosity to get that -- and while i did not bother to load the files and look, I suspect the M8 file will do better there...
Finally, you can try this: Take each image and clip the lows off at around 30 where the human eye begins to get really good at distinguishing minute changes in luminance, and then clip the highs at around 225 just to moderate them so it doesn't throw your means off too much for the std dev calc and then take a look at your relative histo stats...
Frankly, we are talking about human visual perceptions, and one should consider the entire imaging environment and not just a single absolute value derived from a single metric when drawing a conclusion -- otherwise that conclusion is prone to error...
But I'm sure you'll disagree, so whatever...
PS: Marco: this is why you and I see more dimensionality in the M8 file
brainiac wrote:
The very sharp blue peak near zero on the M8 doesn't seem right in this image. It suggests that black is not being rendered as black in the blue channel. The quest for dynamic range can go too far. It is possible that this is connected with the sludgy/noisy shadows. Suitable choice of black point in the blue channel should fix it.
I think you read into far, regarding DR. The peak may just be IR. IR is not stopped by any R,G,B primary color filter array elelments. It go into all color bins. And show up as a specific color.
> ...The remaining issue seems to me to be the colour-accuracy, apart from your resolution-fixation.
Reg: But apart from the sanitation, the medicine, education, wine, public order, irrigation, roads, the fresh-water system, and public health, what have the Romans ever done for us?
Xerxes: Brought peace?
Reg: Oh, peace? Shut up!
I think resolution and colour accuracy are pretty fundamental aspects of a digital camera. The rest is usability, versatility, cost, and lenses, right? It's clear that nice pictures can be taken with the M8, just as nice pictures can be taken with a Nikon Coolpix 700. The only rigorous comparison I have seen with a 5D is my 'crap' 'flawed' one for 'morons', which suggested to my unfocussed eye that the M8 would represent, for me, a slight step downwards in colour rendition and resolution. After 190 pages I'm still looking forward to less moronic comparisons than mine. This is not a 'demand' to anyone, by the way, it's just an expression of interest. I know you said you might do a comparison yourself Carsten. I will certainly be keen to look at your results and hear your opinions on them.
I think I forgot a smiley. If you recall, I am the one person, other than shirozina who beat me to it, the poofter, who sees why your wedding panoramas do justify your resolution fixation.
So here is the smiley for you:
Colour accuracy is certainly a major remaining point. Apparently Leica is already up to about firmware 1.14, in spite of customers having seen only 1.09, so something is cooking (fast) in Solms...
FWIW, I don't find your comparisons moronic, and they had value. There were some points which one could criticise, to be sure, but these points don't invalidate the tests to me.
I do expect that in the end, the conclusion is that the M8 and 5D are roughly equivalent, with each camera having its own strengths, so I am not too excited about the outcome, personally. My M8 is taking so long to return that I am wondering if I can hang on to my 5D long enough to properly compare. I am not infinitely rich, after all.
There is a bug in your one sentence, by the way. It should read like this: "It's clear that nice pictures can be taken with the M8, just as nice pictures can be taken with a Hasselblad H3D-39".
I do believe that you are correct that the M8 is currently a step down compared to the 5D, wrt. colour and resolution. I expect that when the M8 firmware revisions gain the maturity of the 5D firmware, the verdict on the colour issue will be reversed. The resolution issue is unlikely to change, though the *characteristics* of the images may swing that around, i.e. the M8 makes very nice *looking* images, even if their resolution is technically lower.
Pondria - imagine a sensor whose pixels are only sensitive to green. How is processing going to rebuild a full colour image? In a way all sensors are probably slightly like that, since they probably don't have perfectly linear response to different frequencies at different intensities. In other words, sensors have gamuts too, and this will limit possibilities even before raw processing enters the fray. Some sensors have better sensitivity to red or green or blue than others, by design. Processing might fix the balance but it can't do so without reducing the number of output colours (tone) in some part of the colour wheel.
Pondria wrote:
I think the color accuracy is NOT a fundamental aspect of a camera but of a RAW converter.
I bet some/all cameras adjust RGB pre-raw. I doubt the naked sensor is linear, and there may be S/N advantages to not making the output linear with the filter ( R, G or B) on individual sensors.
Thinking about it more, I guess not. A Raw converter always has a camera profile, right?
brainiac wrote:
Pondria - imagine a sensor whose pixels are only sensitive to green. How is processing going to rebuild a full colour image? ...
Your point is well taken. I agree. My statement was too extreme. Current unique situation with M8 may be the example that won't support my generalization.
dcmiller wrote:
I bet some/all cameras adjust RGB pre-raw. I doubt the naked sensor is linear, and there may be S/N advantages to not making the output linear with the filter ( R, G or B) on individual sensors.
Thinking about it more, I guess not. A Raw converter always has a camera profile, right?
The sensor that you're talking about is the array of raw photo diodes. They are pretty much linear photon-collectors. But What Richard is talking about is the sensor covered with the tinted color glass. The color filter is the part of the signal.
The photo-diodes need to be guided, which involve UV and IR blocking and Anti-Aliasing. If you go really naked, you would get pretty bare signal with great sensation. But there is downside as well as upside of the risk.
I personally believe that Leica took the risk too aggressively. Having to have the IR filter on every lens is not a viable option that a responsible company would offer.
I predict that Leica will recall all the M8s to put the IR filter on the sensor as soon as they are ready to do so.
I thought it was crystal clear, in fact. I was happy, even, for whatever reason.
