Hi all,
I needed to pack lots of rectangles into a larger rectangle for my current project, and found that there wasn’t yet a suitable Julia algorithm, so I wrote one up and I thought I’d share it. I took the algorithm from this blog post about lightmap/sprite packing and generalized it to N dimensions. I took the binary tree datastructure from the AbstractTrees.jl examples.
Here is the code on Github
Here are a few example runs in 2D:



This is is my first Julia package I’m posting public, so I’d love any feedback on the algorithm, the code correctness/efficiency, or the package structure. I haven’t actually added it to the package repository yet, but I imagine that isn’t too hard.
Any suggestions/PRs are welcome!
Cheers,
Oliver
5 Likes
Awesome 
Would be amazing, if we could add that to:
https://github.com/JuliaGeometry/Packing.jl
Where I already ported 2 algorithms!
I think there is quite a bit of value to conform to the same interface and being able to just switch out the algorithm 
3 Likes
Great! If I had found your repo first, I probably wouldn’t have written it myself, since it’s basically the same algorithm
, although I generalized to N dimensions. I’d be glad to somehow include my generalization into your existing binary tree algorithm in Packing.jl.
Two discussion points are occurring to me:
- If we have an N-dimensional binary tree algorithm and a 2-dimensional guillotine algorithm, it seems like we need to expand the interface to N-dimensions, and throw an error when trying to use the guillotine algorithm for N != 2. Does this seem appropriate? Any thoughts here?
- I went back and forth for a while about using GeometryBasics.jl. On one hand, I like using existing packages and a common interface. But on the other hand, it seems like a
GeometryBasics.Rect
must have a position, which is not relevant for the input to a packing problem. So I ended up not using it in an attempt not to require extraneous input data, especially since I wasn’t using any GeometryBasics
methods. Do you know if it’s possible to have a size-only Rect
? Does the extra 0, 0
not seem like a big deal to you? Do you think it’s advantageous to use the existing package in this situation?
Thanks for the reply - I’d love to hear your thoughts!
Oliver