Hi all, I hope no one minds me mailing this. I help maintain vips, a free image processing library, and we've just released a new version, vips8. This new version has a much better Python interface. It supports python2 and python3, and it works on any Unix, OS X and Windows. It comes with a large test-suite, written in Python, which it passes with no memory leaks.
The speed and memory use varies with the operations you are performing but typically it's 5x faster than PIL or Pillow and only needs 1/10th of the memory. There's a benchmark here: http://www.vips.ecs.soton.ac.uk/index.php?title=Speed_and_Memory_Use with some sample source code. On that test at least, vips is even quicker than OpenCV. vips is interesting because it's a demand-driven library. Most image processing systems load up the input image, perform a series of operations one after the other, then write the result back to disk. This is simple, but can use huge amounts of memory if the images are large. Instead, vips breaks the image into small chunks and streams it through your system. All operations are active at the same time, including the load and save operations, and no more of the image is kept in memory than is needed for computation. It runs a separate pipeline of operations for each core in your computer, so threading is efficient. This page introduces the Python interface: http://www.vips.ecs.soton.ac.uk/index.php?title=Python And this chapter in the manual goes into more detail: http://www.vips.ecs.soton.ac.uk/supported/current/doc/html/libvips/using-from-python.html John _______________________________________________ Image-SIG maillist - Image-SIG@python.org https://mail.python.org/mailman/listinfo/image-sig
