Part of this is expectations and semantics I suppose. We all agree there is an improvement in underlying image quality of conversion.
When I first started working with the early betas I was not convinced I saw any improvement. Part of this was due to some very poor presets for noise reduction and sharpening in the early betas for the camera I was testing (mostly Phase One and Canon). It would not surprise me if some of the lower-market-share cameras like the Sony still need some tweaks to the presets (perhaps even some of the underlying math).
Also, the improvements are not huge on every image. They will be greater:
- when the raw is already sharp. Out of focus images benefit much less from the improved debayer (though they still benefit from the better noise reduction). So images shot at unsharp low apertures (like the Canon 1.2 lenses wide open) higher apertures that are diffracted (e.g. f/16 on a D800) or with mediocre quality lenses. Likewise my initial evaluation leads me to believe the AA-free sensors benefit more than the AA'd sensors - though I don't have any direct comparisons to make a more rigorous evaluation.
- when the subject matter has very fine detail, especially if the detail is of a repeating or geometric nature
- when the subject matter has significant single-pixel wide lines
Also, the increased sharpness/detail is only half the story to look at. The other part is the much lower occurrence of artifacts. Less moire (even without post-debayer moire reduction), less single pixel noise (and better post-debayer single pixel noise reduction when needed), and less, and better angled lines. This is a nice improvement when the image is presented at defaults, but is even more impressive when making enlargements (e.g. uprezzing for a really big print) or adding sharpening.
Lastly the improvements are made to the guesswork of filling in the missing information from the bayer pattern. Which inherently means that there are a small number of instances when v6 actually produced a more correct/pleasing result for a specific part of the image. In this shot
for instance the very tight-knit clothing looks MUCH better in v7 in about 70% of the fabric, moderately better in about 20% of the fabric and moderately worse in 10% of the areas of fabric. This is also the case in other test images I've spent extensive time with. So from that point of view it's possible to over analyze and look for any area of weakness rather than making a holistic evaluation.
Whether it's overhyped, a major improvement, or subtle change, is semantics and relative to your expectations. I honestly did not think there was room to improve debayering math at this point (considering the math has been around in some form since the mid 70s). So for me my expectations are low when it comes to new software providing significantly better rendering of detail for existing raw files, and my expectations were greatly exceeded.
The improvements in highlight/shadow recovery, high ISO noise reduction (not addressed in this post) are quite large in my opinion.