May I know the actual process for benchmarking?

[rabbitism]

I think the CSharp script is not well optimized. I have rewrite the code according to C++ implementation in this repository
https://github.com/rabbitism/SkeletonTracingCSharp
From my rough testing it's about 40% faster than currently c# implementation. But in your README you said you excluded thinning process, so may I know what is the actual code you run for the 1000 replication?

[LingDong-]

Hi! My benchmark process was very basic. I run the thinning algorithm once to get the thinned image first (without recording time), then repeatedly trace the thinned image 1000 times in a loop (recording the time). My c# implementation was tested in Unity3D. I don't have that piece of code with me right now to paste here, sorry!

Congrats on the fast implementation! Seems like you're using linked lists for the polylines and threading too (I was lazy and just used builtin array), so it makes sense yours is faster :) I'll be happy to put a link to your implementation in the README, or to merge the code here as alternative to the current c# one, if you'd like. Thanks!