Convolution but fast

Little repo following my blog post : Convolution, but fast

This purely to ulistrate the implementation of some conv algorithms. You can use it for your own project ;D.

You can also play with it on some images to learn more about the convolution, to apply gaussian blur or extract edges for example.

Quick Usage

Install dependencies:

pip install numpy matplotlib pillow

Run the script:

python convolve.py --image_file 'path/to/image.ext' --filter 'edges_extract'

• --image_file: Path to the image file.
• --filter: Choose from:
  • edges_extract
  • gaussian
  • invert

Example: Apply a filter in code

from convolution import memstrided_matrix_convolution_2d
from convolve import get_image_greyscale, show_images

img = get_image_greyscale('path/to/image.jpg')
kernel = np.array([[1, 0, -1], [1, 0, -1], [1, 0, -1]])  # Edge detection
result = memstrided_matrix_convolution_2d(img, kernel)
show_images([img, result], cmap='gray')