marcoc2 / pixel-art-remaster-gpu Goto Github PK

So, I've been playing around with a few implementations of the http://johanneskopf.de/publications/pixelart/paper/pixel.pdf paper when I stumbled on your work. I've been able to get it running, but my results are not what I would expect.

In all of the examples the similarity graph naturally creates a smoothing affects along curves creating more of an approximation of the original image. What I see when I run the code from this repo is a perfectly scalled 1-1 representation. The graph visualization might show some subdivsion but all of the individual squares seem to be present in it. I was wondering if maybe CUDA could be silently failing some routine on my machine or if I need to uncomment some line of code in the source tree.

I'm hoping to modify the source to write out its results as either very high res PNGs or SVG graphics, but I'm currently using it as is to study how various input images interact.