Various?

[alan2here]

I don't know whats important and whats not, and these questions could depend highly on each other:

• The existing nearest neighbour scaling function would probably be better as a bicubic one, in examples so far I've been scaling manually as opposed to using the automatic approach I coded.

• It might be important that the input image is high quality, losslessly compressed etc...

• Should the scaling function use the same sort of scaling for (where used, downscaling the ground truth) and (upscaling the 'orig' image each iteration)?

• Is it always better to use a single iteration, or when this really works well will multiple iterations become more effective?

Ticked items are resolved.