p.16 #1 · M8/Leica People I'm Thinking About Switching
The M8 has a "clarity" (lightroom lingo :P ) that is quite high far before I start cranking the bars in LR, I guess that is the lack of AA-filter.
If I find an used one for 1600-1800 the next few weeks, I'm getting it. If I dislike it, I'll just selll it
Jun 20, 2009 at 06:59 PM
brainiac Offline [X]
p.16 #2 · M8/Leica People I'm Thinking About Switching
ulrikft wrote:
The M8 has a "clarity" (lightroom lingo :P ) that is quite high far before I start cranking the bars in LR, I guess that is the lack of AA-filter.
The downside is that the lack of AA filter introduces strong moire effects all over the place, because according to wave theory everything is part of some kind of pattern. Leica was therefore forced, by its own admission, to build in some kind of electronic anti-moire filtration (probably a Fourier frequency filter, akin to Photoshop's dust & scratches) which has the unfortunate effect of nuking detail around the 8-10 Mpixel frequency. IMO this is the reason that in comparisons of fine detail between 5D and M8, the M8 seems to contain none of the fine detail which the 5D captures. Of course it lags even further behind cameras like the 50D and 5D2 for detail.
What evidence do I have for this? First, Leica belatedly admitted that the camera had in-built anti-noise processing - you can look that up. Secondly, here are crops from a crude test that I made which show particular frequencies of data disappearing from the image despite the fact that they are well below the highest frequency of the sensor. The top crop is the M8. Although no noise reduction was used, note how the spokes of the umbrella have vanished, while broader objects are still well-defined: http://cyberphotographer.com/m8v5d/images/brollycrop.jpg
This illustrates that objects above a particular frequency are being smoothed over, while other edges remain sharp.
Note also that the Fourier detail-zapper doesn't fully solve the moire problem either, as illustrated by the funky colours all over the black text of the sign. The AA filtered version below lacks no detail by comparison, but successfully avoids the false colours that are the result of a Bayer matrix with no AA filter.
Ultimately the muddying effect of Canon's and Nikon's AA filters can be quite well undone by various USM techniques, when necessary, and their cameras manage to capture more detail, right up to the sensor's limit, as seen by the naked eye in tests. The M8 in-camera processing algorithm, on the other hand, removes detail in such a way that it can never be regained by a sharpening algorithm. The practice illustrates the theory.
For a long time people have dismissed this comparison for all sorts of convenient reasons, but the only person who ever bothered to try to repeat it was Jack Flesher. His comparison was much better than mine, but showed precisely the same thing: ironed out small-scale detail and ubiquitous red fringes in the M8 capture, against more detail and no moire/fringeing in the 5D file. He skewed the results a little by failing to display at equal magnification, but it is still clear that the M8 in his test showed less fine detail and no greater clarity.
One reason why people may be immediately impressed by the clarity of M8 files is that when they view them at 100% on screen they are viewing at a lower magnification than they have been used to with 12+ Mpixel cameras. Viewing a D30 at 100% illustrates the point well: the clarity and sharp pixels are beautiful. Only compare at equivalent magnifications - i.e. uprez to a common number of pixels.
"As you can see from the figures above the M8's built-in software anti-alias (moire-reduction) algorithm begins blurring the image fairly early and limits horizontal resolution to around 1900 LPH (compared to between 2400 and 2600 LPH from RAW). This puts the M8's JPEG resolution in the same ball park as the best eight megapixel digital SLRs (which is a pity considering the amazing resolution, thanks to not having an anti-alias filter, available from RAW)."
"Removing the differences between the camera's internal processing engines and using Adobe Camera RAW we can see that the M8 does deliver an incredible amount of detail and 'per pixel sharpness', no anti-alias filter means that there's no blurring of the image captured by the sensor and hence none of that has to be undone by excessive sharpening. The disadvantage of course is an increase in the potential visibility of moire in high contrast, high detail areas (which can be seen in some of the crops above) although frankly it's not distracting and is a trade-off I personally don't mind making. There are actually one or two of the EOS 5D crops above which look 'more detailed', in print this difference would be indistinguishable."
Seventeen megapixels anyone?
"Just to see how well both camera's images interpolate to larger sizes we processed both RAW's through Adobe Camera RAW using the seventeen megapixel output option (no sharpening and then 80% unsharp mask applied, as above). A single crop of the results can be found below and you can if you wish download these images and examine them in more detail. You will see that the M8's images interpolate very well (again thanks to a sharper starting point than a camera with an anti-alias filter) although some moire is visible on black & white high contrast detail."
just forget the brand B amateur analysis, take it from people who do this for a living, and whom while not faultlessly objective, can at least hold a note better than a crying child...
p.16 #4 · M8/Leica People I'm Thinking About Switching
The DMR also has no AA filter and has built-in software to reduce moire, which I don't use (because it's used only with jpg output). I use only RAW output with the DMR so I see the effects of aliasing occasionally, mostly in the fine detail of bird feathers. Some species are particularly prone to moire, such as the California Quail in the back and breast feathers:
Interestingly, the same species also shows moire when photographed with the Nikon D200 which has an AA filter. In the photo above I've used post-processing software to reduce the color moire in the affected areas. The inevitable softening introduced by the moire-reducer affects only the area where the moire was a visible problem, instead of the hardware AA filter's softening even where it's not needed.
I'm sure that Richard ('brainiac') will need the have the last word and with extreme pixel-peeping will demonstrate that even after the selective moire reducer there's some aliasing, i.e., the lack of AA filter introduces false data. Likewise the capture softening of the AA filter and the 'sharpening' (which is no such thing, it's local contrast enhancement) to supposedly bring the sharpness back is an introduction of false data.
The user has a choice: an AA filter + additional local contrast enhancement or no AA filter and selective moire repair, either way you'll find false data if you're a sufficiently persistent pixel-peeper.
Here are a few more photos where I've used selective moire reduction. Can you find the false data?
Can you find the false data? If you can, does it detract from the photo?
Here are a few more that have the full benefit of no AA filter, they have the full sparkle and brilliance the DMR can produce, and I have not seen any aliasing in any of them:
My experience has been that moire and aliasing are very minor issues and in the relatively few cases where it's a problem it's overshadowed by the brilliance and richness of the vast majority of the camera's output. I'm sure Richard will need to have the last word and demonstrate that the above photos are indeed fatally flawed by the lack of AA filter. All I know is that I'm very pleased with Leica's design choice of not using an AA filter on the DMR and that those who have purchased my photos have remarked about the brilliance of the photos and that nobody has complained about moire or aliasing.
The graph you have linked to is the "screen" comparison, which in DXOmark parlance means comparison at 100%, i.e. comparison of equal numbers of pixels. It is an entirely misleading metric, since it ignores the differences in the numbers of pixels in cameras, and therefore is a way of comparing crops from cameras at differeing magnifications. With this methodology it is possible to show a Minox outperforming a Hasselblad. I don't mind bullshit, as long as it's factually correct bullshit. If you don't understand why the "screen" side of the DXOmark site is thoroughly misleading, there are plenty of threads here on that subject. To get a better idea of relative camera performance, you need to click the "print" button which is a slightly less misleading method of comparison. Even DXOmark rates the picture quality of the 500D above that of the M8 in their DXOmark score, for the little that such an academic analysis is worth.
p.16 #7 · M8/Leica People I'm Thinking About Switching
What's that line..."you can't see the forest from the trees!".
Jun 21, 2009 at 12:08 PM
brainiac Offline [X]
p.16 #8 · M8/Leica People I'm Thinking About Switching
I've mentioned before DPReview's method of comparing 100% crops between cameras which have different numbers of pixels. It's the same error of comparison as the "screen" option on DXOmark. The files are enlarged by different factors, and then noise and fine detail are compared. It's completely unfair, and the DPReview examples shown above are no exception. You can see that the DPR crops above differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35, which is the ratio between the number of pixels of the respective cameras. Here are the DPR crops both downrezzed and uprezzed to equal magnifications. Obviously downrezzing the 5D samples throws away some detail.
M8 uprezzed: http://cyberphotographer.com/m8v5d/images/m8rezzedv5d.jpg
I can only say what I see, which is this: the M8 file has red and blue moire around fine detailed edges like the black on white text, whether it be English or Japanese. Unfortunately there are few items in the image which constitute fine texture, but the oak tree on the bottle label qualifies as such. The 5D captures more texture in the oak tree. I see none of the extra clarity that the M8 is supposed to show, so I can only put those reports down to the effect of lesser magnification.
What it boils down to is that the M8 is practically as good as the legendary 5D if you shoot raw, except that it has a very slight shortfall in detail. The lack of AA filter does not provide it with some mystical ability to capture extra detail and clarity. It does suffer from coloured fringeing around high contrast small-scale texture, and it also can't keep up with the 5D or 500D at 2500 iso and above due to increasing noise. Overall image quality is rated as inferior to the 500D on DXOmark.
It's a nice camera if you like rangefinders, don't mind moire, and can spare the coin.
p.16 #9 · M8/Leica People I'm Thinking About Switching
brainiac wrote:
I've mentioned before DPReview's method of comparing 100% crops between cameras which have different numbers of pixels. It's the same error of comparison as the "screen" option on DXOmark. The files are enlarged by different factors, and then noise and fine detail are compared. It's completely unfair, and the DPReview examples shown above are no exception. You can see that the DPR crops above differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35, which is the ratio between the number of pixels of the respective cameras. Here are the DPR crops both downrezzed and uprezzed to equal magnifications. Obviously downrezzing the 5D samples throws away some detail.
I can only say what I see, which is this: the M8 file has red and blue moire around fine detailed edges like the black on white text, whether it be English or Japanese. Unfortunately there are few items in the image which constitute fine texture, but the oak tree on the bottle label qualifies as such. The 5D captures more texture in the oak tree. I see none of the extra clarity that the M8 is supposed to show, so I can only put those reports down to the effect of lesser magnification....Show more →
which everybody dealt with including Askey
I wrote THIS in for a reason
"Seventeen megapixels anyone?
"Just to see how well both camera's images interpolate to larger sizes we processed both RAW's through Adobe Camera RAW using the seventeen megapixel output option (no sharpening and then 80% unsharp mask applied, as above). A single crop of the results can be found below and you can if you wish download these images and examine them in more detail. You will see that the M8's images interpolate very well (again thanks to a sharper starting point than a camera with an anti-alias filter) although some moire is visible on black & white high contrast detail."
just forget the brand B amateur analysis, take it from people who do this for a living, and whom while not faultlessly objective, can at least hold a note better than a crying child..."
as to the difference between 10.3Mp without an AA filter, and 10.8Mp with an AA filter
its less than the root of (12.8/10.3) = 11.4%,
it is NOT "differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35," which is the wrong math in a noise/resolution pretext
having no AA filter will release more detail, ipso facto detail is not an issue with this camera, and its IQ compares well with 5D.........live with it...
as the OP wrote
"I'm thinking about getting a used M8 and 28-35-50 or 16-18-21 as my everyday travel camera and an M3 for film. It just seems like the perfect city camera, small and fewer eyes looking at be while I make the shot."
M8 is fine for this if you can handle the eccentricities of a RF, that was after all the subject, not some B grade backyard analysis made years ago with now ancient firmware in a few hours of inexperienced hacking about which is obvious from your examples
and Richard wrote
"One can be subjected to a lot of abuse for telling the truth about the M8"
and one can be accused of being a troll, plain and simple...
the latter is your form
p.16 #10 · M8/Leica People I'm Thinking About Switching
@Rob Riley: the problem is that DPReview's image analysis is 'B Grade backyard analysis' based on procedures originally invented by somebody who had no clue what they were doing at the time (Phil Askey, before he went full-time with DPReview). Richard's point about how 100% crop comparisons of different resolution cameras are very misleading is thoroughly accurate, whether applied to the M8, or comparing high ISO between say a 1DsmIII and a D3.
Note I'm not arguing anything about the M8's IQ here. Just that DPreview's IQ methodology is broken (As is DXO's, but for other reasons. They simply don't seem to be able to acccurately repeat test results, which means that they will rate cameras with absolutely identical sensors and processing systems differently on IQ. It's usually not a large difference, but is also often enough to wonder just how rigorous their testing is when the D3 and D700 rate slightly differently and the D90 and D300 rate a full 5 slots apart. It seems the primary issue is in their low-light testing. The D90, with nearly identical processing, sensor and results should not have a rating 50% higher than the D300)
p.16 #11 · M8/Leica People I'm Thinking About Switching
mawz wrote:
@Rob Riley: the problem is that DPReview's image analysis is 'B Grade backyard analysis' based on procedures originally invented by somebody who had no clue what they were doing at the time (Phil Askey, before he went full-time with DPReview). Richard's point about how 100% crop comparisons of different resolution cameras are very misleading is thoroughly accurate, whether applied to the M8, or comparing high ISO between say a 1DsmIII and a D3.
which would in ordinary circumstances would share some truth
but he did upres both files to 17Mp, which pushes them both beyond their habitat
and i did identify the actual difference
like i said, if they were both AA less or AA equipped cameras the difference would be 11.4%, since they are not the results differ dependent on the form of examination, but it is not as great as Richard would have you believe
which in a fair context, i would have thought a most obvious statement to make
but this isnt about 'fair' is it ?
p.16 #12 · M8/Leica People I'm Thinking About Switching
Rob Riley wrote:
which would in ordinary circumstances would share some truth
but he did upres both files to 17Mp, which pushes them both beyond their habitat
and i did identify the actual difference
like i said, if they were both AA less or AA equipped cameras the difference would be 11.4%, since they are not the results differ dependent on the form of examination, but it is not as great as Richard would have you believe
which in a fair context, i would have thought a most obvious statement to make
but this isnt about 'fair' is it ?
I'm not arguing about Richard's position on the IQ of the M8, nor do I necessarily agree with it (He has valid points but is overstating the matter IMHO). I'm simply pointing out that his arguments on the problems with DPreview's methodology are valid. The 17MP uprez test is not part of DPReview's standard test methodology and was a one-off for the M8 test.
p.16 #13 · M8/Leica People I'm Thinking About Switching
mawz wrote:
I'm not arguing about Richard's position on the IQ of the M8, nor do I necessarily agree with it (He has valid points but is overstating the matter IMHO). I'm simply pointing out that his arguments on the problems with DPreview's methodology are valid. The 17MP uprez test is not part of DPReview's standard test methodology and was a one-off for the M8 test.
that is the test at hand
and it covers the issue between M8 and 5D file differences,
im not at this point as concerned with Askey's methodology and policy as you appear to be.
Richard not only 'overstates' the differences as i pointed out
"You can see that the DPR crops above differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35, which is the ratio between the number of pixels of the respective cameras. Here are the DPR crops both downrezzed and uprezzed to equal magnifications. Obviously downrezzing the 5D samples throws away some detail."
is quite simply...factually wrong
the difference is less than 11.4%, with the caveats i provided earlier
p.16 #14 · M8/Leica People I'm Thinking About Switching
Rob Riley wrote:
Richard not only 'overstates' the differences as i pointed out
"You can see that the DPR crops above differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35, which is the ratio between the number of pixels of the respective cameras. Here are the DPR crops both downrezzed and uprezzed to equal magnifications. Obviously downrezzing the 5D samples throws away some detail."
is quite simply...factually wrong
the difference is less than 11.4%, with the caveats i provided earlier
No, the linear magnification is less than 11.4% difference. But the difference in magnification is as Richard states as he's using 2D numbers (Area), not linear numbers. You're both factually correct.
Jun 21, 2009 at 02:18 PM
Andi Dietrich Offline [X]
p.16 #15 · M8/Leica People I'm Thinking About Switching
Can somebody please post a photograph of a beer can, might bring something new to the discussion
Jun 21, 2009 at 03:28 PM
brainiac Offline [X]
p.16 #16 · M8/Leica People I'm Thinking About Switching
Rob Riley wrote:
"You can see that the DPR crops above differ in magnification by a ratio of about 1.3:1. In fact the ratio is 12.72:10.35, which is the ratio between the number of pixels of the respective cameras. Here are the DPR crops both downrezzed and uprezzed to equal magnifications. Obviously downrezzing the 5D samples throws away some detail."
is quite simply...factually wrong
Regardless of my arguable misuse of the word 'magnification', the resized versions that I presented here are resized correctly, and the DPR versions are incorrectly compared for per pixel noise and detail, instead of per square inch of equal-sized print.
the difference is less than 11.4%, with the caveats i provided earlier
Any testing method that skews results 10% in favour of one of the cameras is a bad method. In this case it's a mere 10% since the 5D and M8 are quite close in megapixellage. When it's 21 against 12 the skew is worse. The method is wrong, and 10% skew is too much. DXOmark has the cheek to make that broken method of comparison the default. I've heard it argued around here that DXOmark visitors are clever enough to know that the results are skewed, and clever enough to be able to mentally compensate when using the DXOmark comparer. I am not that clever.
At any rate, DPR's 17 Mpixel crops show quite clearly that the absense of an AA filter does not compensate for the M8's fewer pixels, and the 5D really does deliver slightly more detail in spite of its AA filter. DPR's other comparison shows that the 5D suffers less moire too.
p.16 #17 · M8/Leica People I'm Thinking About Switching
So, the conclusion is... the m8 is on par with the 5d when it comes to resolution. It can't beat the 5d on iso (but you can handhold it with slower shutterspeed, and you have 0.95, 1.1, 1.2 etc glass in ranges that your 500d can only dream about?). It _is_ on the other hand, far smaller, and less... "I'm in your face with a camera" than a 5d with a 35 1.4L is?
p.16 #18 · M8/Leica People I'm Thinking About Switching
ulrikft wrote:
So, the conclusion is... the m8 is on par with the 5d when it comes to resolution. It can't beat the 5d on iso (but you can handhold it with slower shutterspeed, and you have 0.95, 1.1, 1.2 etc glass in ranges that your 500d can only dream about?). It _is_ on the other hand, far smaller, and less... "I'm in your face with a camera" than a 5d with a 35 1.4L is?
or?
Are we level then? :P
Yes, that's pretty accurate for the most part, although you can have f1 glass on a 5D if you're willing to pay Leica-esque prices (Last I recall, the 50mm f1L ran around $3k for a used copy, not all that far off a Noctilux f1 price). The M8 of course gets all the uber-fast 3rd party glass that's only available for it and the G1/GH1 (Since the R-D1 is marginal for focusing ultra-fast glass due to a short rangefinder baseline).
p.16 #19 · M8/Leica People I'm Thinking About Switching
ulrikft wrote:
So, the conclusion is... the m8 is on par with the 5d when it comes to resolution. It can't beat the 5d on iso (but you can handhold it with slower shutterspeed, and you have 0.95, 1.1, 1.2 etc glass in ranges that your 500d can only dream about?). It _is_ on the other hand, far smaller, and less... "I'm in your face with a camera" than a 5d with a 35 1.4L is?
or?
Are we level then? :P
Well, then it will come down to the price.... . I am about to leave for Norway/Spitsbergen in less than 24 hours but I will have to leave my M8 at home .
p.16 #20 · M8/Leica People I'm Thinking About Switching
You are coming here, and you are not bringing your m8?! ... ach :P Staying in Oslo any amount of time? Let me know if so, I'll get you a cup of coffee and ask you lots of questions about the m8.
mawz: if I'm getting one (and it looks like I will), it will be m8 + 35 1.2 + 50 1.1 (both cv) or 35 1.4 + 50 1.4.
.