cpx

Note: Currently, cpx is a rough draft of an initial CPU benchmark for a set of base utilities that enable memory efficient and fast processing of calcium imaging data.

Most source extraction libraries (that I'm aware of) have trouble scaling past ~50 GBs of data, are not designed with memory efficiency in mind, and additionally leave a lot of optimizations and flexibility on the table. Early CPU benchmarks lead me to believe that a ~5-10x increase in performance (with better memory management) is possible on CPUs alone, while maintaining similar (if not exact) source extraction results. Additionally, GPUs might lead to another order of magnitude increase in performance.

The goal of cpx in the long run, would be to have a set of battle tested utilities (for filters, metrics, registration, extraction, etc.) that are designed from the bottom up with memory efficiency and performance in mind (on both CPUs and GPUs). This also extends to 2-photon stacks.

If your interested in this problem and would like to talk (either as an end-user or developer), send me an email at RyanIRL (at) icloud (dot) com.

Installation

pip install cpx

License

MIT, see LICENSE.txt.