[dahart]

Love PIL/Pillow, for just about the easiest image I/O there is.

I've been using it recently and needed a bit more industrial strength and speed than the library provides. PIL supports converting an image to/from a NUMPY array! (And the execution speed difference is definitely worth the effort...)

Now my template for a new PIL script looks like this (all error checking removed for simplicity):

from PIL import Image

import numpy as np

im = Image.open(args.input)

rgb8 = np.array(im)[...,:3]

rgb = rgb8.astype('float') / 256.

# ... do stuff to 'rgb' using np

rgb8 = np.clip(rgb * 256., 0, 255).astype('uint8')

im = Image.fromarray(rgb8)

im.save(args.output)