IM2GR

An exploration of algorithms for constructing graphs derived from image-like data in parallel. See our report (my senior thesis).

Examples

using IM2GR

data = <data source>
diff_fn = <image diff func>
d = <search distance>
image = im2gr(data, d, <construction mode>, diff_fn, track=true)
@show image.ei, image.ej, image.evd, image.evi

The result vectors are stored in image.

Testing

Use ] to enter Pkg REPL, and do:

pkg> add https://github.com/sinha-abhi/im2gr
pkg> test

or, do:

julia> import Pkg; Pkg.add("https://github.com/sinha-abhi/IM2GR.jl");
julia> Pkg.test("IM2GR)

TODO

• Option to construct graphs lazily
• Distributed mode for construction (multi-process)
• Better memory management for CUDA (actually, better memory management in general)
• Use multiple GPUs if available
• Tune CUDA kernel launch
• Batching for larger images