Affine Transformation: C++17, OpenMP, Python

This project explores how C++17 and OpenMP can be combined to write a surprisingly compact implementation of n-dimensional parallel affine transformations which are linked into Python via the affine_transform module.

While this project is still under development, the following features are supported:

• Linear and cubic (without prefiltering) interpolation
• Constant boundaries
• Compiling code for arbitrarily dimensional data
• Parallelism via OpenMP
• Arbitrary shaped output arrays, allowing e.g. to only extract a transformed slice

Short example usage

import numpy as np

from affine_transform import transform
from mgen import rotation_from_angle


# Create a simple white square in an image
original = np.zeros((601, 401))
original[100:300, 100:300] = 1

# Rotate by 22.5° (around the centre of the square (200,200))
# and shift +200 in x and +100 in y
transformed = transform(
    original, rotation_from_angle(np.pi / 8), np.array([200, 100]), origin=(200, 200)
)