A research-grade masterclass rebuilding the entire Computer Vision stack from first principles. No OpenCV, no PyTorch—just pure math and NumPy. Implements custom convolutions, Canny edge detection, eigenvalue-based Harris corners, Lucas–Kanade optical flow, and a CNN layer with full backward passes from scratch.
computer-vision image-processing cnn neural-networks optical-flow mathematical feature-matching cnn-from-scratch canny-edge-detector classical-cv convolution-from-scratch
-
Updated
May 27, 2026 - Jupyter Notebook