benchmark pyvips vs cv2 for image read

bench-read.py

#!/usr/bin/python3

# setup with:
# 
#   mkdir sample
#   cd sample
#   for i in {1..300}; do cp ~/pics/k2.png $i.png; done
# 
# on this modest 2015 laptop running Ubuntu 20.04, I see:
#
#   $ ../bench-read.py *.png
#   cv2: read 300 files with BGR->RGB conversion ...
#       took 39.46 sec
#   pyvips: read 300 files ...
#       took 37.10 sec

import sys
import pyvips 
import cv2
import time
import numpy as np

files = sys.argv[1:]

start_time = time.time()
d = []
print(f'cv2: read {len(files)} files with BGR->RGB conversion ...')
for f in files:
    b = cv2.imread(f)
    b = cv2.cvtColor(b, cv2.COLOR_BGR2RGB)
    d.append(b)
print(f'    took {time.time() - start_time:.2f} sec')

start_time = time.time()
d = []
print(f'pyvips: read {len(files)} files ...')
for f in files:
    c = pyvips.Image.new_from_file(f, access='sequential')
    c = np.ndarray(buffer=c.write_to_memory(),
                   dtype=np.uint8, shape=[c.height, c.width, c.bands])
    d.append(c)
print(f'    took {time.time() - start_time:.2f} sec')