This tool upscales images using Google's RAISR algorithm.
TODO: Finish the training part of the algorithm. Currently, this tool uses a filterbank that is exported from movehand's Python implementation.
USAGE:
rusty_raisr [FLAGS] [OPTIONS] <input> <output>
FLAGS:
-b Benchmark GPU performance
-g Whether or not to use the GPU
-h, --help Prints help information
-V, --version Prints version information
OPTIONS:
-d <FILE> Name of filter image to write
-f <FILE> Filterbank to use [default: filters/filterbank]
ARGS:
<input> Input image
<output> Output image
The -g option uses an OpenGL compute shader to perform both the hashing and upsampling steps. You will need OpenGL 4.3 or higher. This backend is still a WIP -- there are a variety of small optimizations that I have not tried yet.
The "filter image" is a useful tool for visualizing which filter will be used per pixel. This feature does not work with the GPU backend.
(from left to right: nearest neighbor, bicubic, and RAISR)