caching speeds up batch processing

[swell-d]

Hi,
Thanks for the updates on the library.
Could you please check if caching would help speed up the batch processing of a large number of images?
I haven't delved into your code and don't understand what it does, but maybe it could help.

[BrianPugh]

I'll slightly tweak this PR and then merge it in. We need to use lru_cache instead of cache with a reasonably short cache size (lets say... 8)? Otherwise this could use an unexpected amount of memory. Additionally, since we are returning a numpy array (mutable) we should attempt to make it immutable by setting the write flag to false.

slice_chunk is a super fast function and barely benefits from caching. On my computer, a typical uncached call is about 5uS slower while a typical cached call is about 5uS faster. So let's not cache this one.

All the others are pretty fast (typically just a few mS), but would be faster if cached. This would be beneficial if processing a lot of images of the same resolution. For our test/demo image, this would make subsequent processing of same-resolution images ~25% faster.

[swell-d]

Super!

[BrianPugh]

if you're processing a large dataset (or any use in general, I suppose), make sure you have opencv installed! If it's installed, py360 will internally use its faster warping functions.

Thanks for the PR!

[swell-d]

Hi Brian,
Of course I'm using opencv. Here's my website: https://360-for-you.com/
In all the time users have already processed more than 100,000+ panoramas with your library.
Thank you a lot!

[BrianPugh]

awesome! I also just added sampler caching; so now all possible intermediate calculations are now cached for same input/output resolution processing.