I shot a few image sets this afternoon to compare how the camera would behave in DR auto versus HR priority mode. I wanted to test two things ... first, how well are highlights protected? Second, how much resolution do you lose in 5mp mode?
The first image is a house across the street from me with the sun shining on it *from behind me* ... it has white trim and grey siding. Here are the crops:
I upsized the 5mp image to 10mp to make them comparable. Since this is doubling the size, this puts DR mode at a sever disadvantage. But it is much more interesting than downsizing 10mp mode, since it handles the enlargement case and it makes real differences much easier to see.
What astonished me was that, as I was working on the images, I thought they were reversed. I.e. I thought the one on the left was the upsized image. But it's the one on the right that was shot in 5mp ... hmmm.
Note that there isn't too much difference in highlight quality, so I'd call it a wash. But that's just because DR mode chose DR100 ... not sure why, but probably because it could.
Not so, though with tone curves on the second image, facing west into the sun.
This image shows a tremendous difference in highlight retention and shadow tone. The DR version is much better on both counts.
And again the DR version has better detail definition, even after the upsize operation.
1. The DR version has slight halos from the upsize operation. But these are unlikely to affect the final image.
2. Watch for the fact that I reversed the HR/DR image sequence here ... I shot the Dr, the HR, then turned around and shot the HR and then the DR. This minimized settings changes, but added to my confusion while doing the crops. I referred to the camera itself several times to be sure, since ACR does not display these kinky makernote settings.
Bottom line ... no real reason to shoot HR mode. Does not hold highlights as well, and upsizing properly retains even more detail.
Why does it do that, you might ask? Well, the answer should almost certainly lie in the fact that the binned modes increase the detail size by double *before the NR algorithms see the image* ... which probably takes these fine details out of the range of "noise". I've speculated on that before, and now I am pretty sure about it.