Resizing images (Deep Learning)

speed vs quality vs implementation
coding
python
deep-learning
computer-vision
Author

Soumendra Dhanee

Published

February 18, 2020

Resizing images: speed vs quality vs implementation

Main filters for resizing: NEAREST, BILINEAR, BICUBIC, LANCZOS/ANTIALIAS

  1. Speed: NEAREST > BILINEAR > BICUBIC > LANCZOS/ANTIALIAS
  2. Output (downscaling) Quality: LANCZOS/ANTIALIAS > BICUBIC > BILINEAR > NEAREST
  3. Pillow vs cv2 resize: Pillow seems to come out better (but really old discussion). Also, the outputs (resulting image pixel values) are different!

Sources

Back to top