Originally Posted by skoobey
Exactly what I thought it's based on multiple exposures and tracking, something video editors have been doing for years.
That strategy or at least parts of it came from fluid dynamics, so it goes back further than that. It's not a new concept. It's an improvement, and I haven't seen anything that work that well even in demo form. The thing would require a huge amount of work to be of use to a casual user. Right now it uses a large amount of memory and requires several minutes to process an image of reasonable resolution. The second author posted the paper. It's currently the first link under news
, and it's quite readable compared to a lot of research papers. Some of the notation might be unfamiliar, but it's at least standard notation.
This stuff always involves iterative improvements, but if it doesn't break due to point source lights, flare, or motion in the scene, it would be a pretty enormous improvement. It's hard to get that stuff to work with little to now user interaction.