In various threads around CR, I have noticed quite a few opinions that are not complimentary to the folks at DxO. The various individuals seem to take issue with DxO's methods and conclusions and generally disagree with pretty much everything they offer. Why? Is there some inherent fault with their methodology that would make their conclusions erroneous? (I am neither pro or con on this issue, but would just like some enlightenment.) Do you have any factual basis for disagreement? Comments?
It should be noted that, in addition to all the valid reasoning offered by everyone else, that DXO's scoring system is not just invalid for Canon. It is invalid, period. Because they utilize only ISO 100 measures in several scores, and because they do not FULLY publish all of their algorithms and methodologies, it invalidates the process as a whole. Using DXO to find the best Nikon camera is no better, and the results are just as misleading for them as for any other brand.
For example, Print DR is an EXTRAPOLATION based on DERIVED data based on Screen DR, which itself is determined via an unknown formula...it is not actually a "measure", despite the fact that they claim as much. Print DR is also based on the assumption that images are downsampled to a standard 3200x2400 pixel size. When it comes time to edit RAW images, ALL RAW images are ALWAYS edited at 100% size. It's the nature of RAW. So when DXO says some Nikon DSLRs are capable of achieving more than 14 stops of DR, they are being extremely misleading to people who don't understand what those numbers really mean. There is NO camera that uses 14-bit ADC that can achieve more than 14 stops of DR. Technically speaking, because of required overhead, even if that overhead is minimized, one could never really achieve 14 stops of DR period, you would always attain something just a little less.
Screen DR is the only valid measure of DR that DXO offers. In the case of the D800 and D600, the actual DR is ~13.2 stops, as much as 1.2 stops LESS than the 14.4 one might believe they have according to the Print DR extrapolation. The fact that Print DR IS an extrapolation means that even if someone downsamples their image by a considerable amount to 3200x2400 pixels, then tries to recover that extra 1.2 stops of information, it may not actually exist...Print DR is not a measure. It is derived, and therefor no one can really know for certain if a downsampled image from a D800 or D600 could actually achieve that much dynamic range. From an editing latitude standpoint...how much real-world freedom you have to push exposure around in post, you can only use DXO's Screen DR measure. The problem is, that measure is not shown by default, you have to find it yourself.
DXO claims a scientific approach to measuring cameras. They really fail to provide that on multiple fronts. For one, they don't measure cameras at all...they measure sensors, and only sensors. That fails to factor in ANY other camera features, such a AF system, metering system, body ergonomics, or other camera capabilities that may be as critical, if not more critical, to final IQ than the sensor. Because they do not 100% fully disclose all details of every methodology and algorithm they use, they can't really claim to be scientific (or at least, not openly scientific.) The fact that they clearly seem to have changed some of their algorithms over the years, and sometimes explicitly to cover up errors they made previously, only debases their claim to using a scientific method even more. At the very least, if an error is found in their methodology, it should be clearly explained and not slyly covered up.
Finally, as others have said...trying to reduce such a complex device as a camera to a single score that could theoretically be used for simple comparison of one camera to another is doomed to failure from the get-go....too many variables to consider, especially considering that DXO ignores the vast majority of them (i.e., all the non-sensor factors.) And don't even get me started on their lens tests...thats a debacle that puts their sensor testing to shame!