From previous discussions on this site I always found that people tend to consider less noise and more dynamic range as more valuable than higher resolution. If you pursue hyper-realism, and that is a big "if" thank-you-very-much, IF you pursue hyper-realism, does resolution become more important than either dynamic range and noise?
I am deliberately contrasting the usual factors so as to get a discussion on what people think contributes most to apparent detail. Is it even "apparent" detail that should be pursued in hyper-realism? Or should there be true detail? Does the equation change significantly for downsampling?
Back in the days we used to have a rule of thumb that 4x oversampling is sufficient for perfect anti-aliasing, which basically means that if one starts with 4x the resolution, downsampling should be equal to a full color scan (non-bayer scan) of the scene. Interestingly enough most popular downsampling algoritms use 4 samples from the original to produce 1 destination sample (4x4 pixels for 1 destination pixel). If you sample down an image with 8x the resolution, changes are that the algorithm is skipping half your pixels anyway. A 2 step process would probably be more accurate. Or even a 3 step process using simple bilinear interpolation.
I am mentioning this somewhat technical issue because whenever I read "downsampling", I am not sure whether people realize this. You can have a zillion megapixels, but if you downsample, especially more than 25%, then you may well be throwing out a lot of data unknowingly...
