theotherphil

It turns out that most of the time in the canny benchmark is spent in image::imageops::blur. I've been meaning to implement efficient approximate Gaussian blur via iterated box filters for...

Looks like a very useful addition, thanks!

@tafia has been making a lot of large performance improvements recently. It might eventually be nice to write a short performance tutorial/discussion working roughly through the commit history of e.g....

I'm conflicted on this. My initial not-really-a-plan was just to wait until specialisation landed in the compiler and have faster versions of functions specialised to images that we know consist...

Ok, great. Anything that made functions faster on ImageBuffers would be directly useful for my hypothetical new world of custom traits, and getting the API for these traits right will...

I've just created a "performance" branch for this.

For future reference, here are all the current benchmarks according to my laptop (ignore the "baseline" ones - they're to let us make sure that whatever we add is faster...

It looks good. I've not had the time to work on this library much recently but I've just started reading up on halide a little and think we could nick...

https://github.com/libmir/dcv This D library is based heavily on the ndslice type, so might provide some useful ideas.

Yes, everything is on the CPU at the moment. It certainly would be interesting to add GPU support - do you have any ideas for how we might go about...