That paper is exactly the kind of thing I mean when I said "no existing purpose-built design fits the problem you have well enough to be practical" A custom bu

That paper is exactly the kind of thing I mean when I said "no existing purpose-built design fits the problem you have well enough to be practical" A custom built vision processor for the algoritm they're using would outperform all of the solutions in the paper by at least an order of magnitude. Just because FPGA solves a single, very specialized type of problem well doesn't mean that it solves every type of problem well.