Here is a split view result of the unaltered shot on the left with the USm on the right.
Unsharp nodes.blend (694 KB)
Interesting. How’s this any different from the sharpen filter node at low values (say .10 or .20)?
The mask seems to be a lightweight version of the laplace filter (perhaps inverted) or some other difference filter.
Finally, the image does look blurry so it’s hard to say from the screenshot how much it has been changed.
Yeah it’s not especially clear from that render. I found a range of variation from various types of math node to derive the mask. Probably not better but seems ok as a test.
If it turns out that it only detects edges, then it might be cool to overlay it on the top of the original so that it can be used to enhance (i.e. sharpen) only edges and not every pixel much like the sharpen filter does. Thus, it could be of some use it is generically proven that it behaves like this.
Check the mask image to see what areas gets enhanced.
The image it generates actually looks alot like the high pass filter in Photoshop, which is used alot for manual sharpening.
Thanks for that find hype, I noticed that the mask works on outside edges too. Often edge enhancment makes a halo on the inside (of a dark object) as well.
Thank you for pointing me in the right direction, 3point. I just created a Sharpen mask using your template and it seems to work as expected. But of course, there’s always room for improvement. 
SharpenMask.blend (577 KB)
