cuconv

Simple image convolutions in PyCUDA. Mainly a learning exercise.

Requirements

• CUDA
• PyCUDA

Usage

• For a speed comparison between cuconv, Scipy.signal, and a basic CPU convolution, run 'main.py'.
• Results are stored in the 'results' folder.
• Basic usage:

cuconv.convolve(input_image_as_array, conv_kernel_as_array)

• The function expects a pair of 2D numpy-arrays, with the first corresponding to the input image, and the second being an odd-dimensioned convolution kernel.

Existing data

• See 'results' folder for image results of speed comparison in 'main.py', on horizontal and vertical Sobel edge detection, and 9x9 box blur kernels.