Rationale

Why would you do this in Python instead of [insert your favorite compiled language here] ?

This is a classic “python as glue” app. MOSAICpy itself is more of a framework for chaining individual image processing steps together. Each processing step can actually be written in whatever language you want, you just need to write a simple wrapper (a subclass of ImgProcessor) to hand control over to your subroutines.

On the other hand, using python as “glue” allows for rapid development of some of the less computationally demanding parts of the image processing chain, such as detecting the format of a data directory and parsing metadata. It is also open source and accessible, allowing anyone to write a simple extension. If you would like to speed up your extension, there are plenty of ways to do so.

Tips for writing faster extensions

• Use a just-in-time compiler like numba to to accelerate frequently-called functions that may be computationally intensive. An example in this library is mosaicpy.camera.calc_correction()
• Use a python gpu-acceleration library like cupy, pytorch, or gputools. I find cupy to be particularly nice, as it can often be used as a drop-in replacement for numpy, leveraging CUDA libraries in the background… Letting you develop in numpy, or fall-back to numpy when a GPU is not available.
• Write a python extension for C++ code using something like cython, pybind11, ctypes, cffi, etc… In this library, the mosaicpy.libcudawrapper module is an example of using ctypes to access compiled shared libraries written in C++ from within python.
• Just call some external binary using the subprocess module from the standard python library. While it will be hard to truly “chain” together mosaicpy.ImgProcessors written using subprocesses, it might be the simplest way to simply trigger some external program at some stage in the processing pipeline.