Skip to content
Open to SDE roles · full-time and internships +91 90504 57303
Work/B-side
2024 · Team project

Image processing library

CNN building blocks from scratch: FFT-based 2D convolution, pooling and activations.

Team projectC++FFTCNN basics

What it does

A recursive Cooley-Tukey FFT, extended to 2D, does convolution in the frequency domain. Max and average pooling and ReLU, tanh and softmax complete the core of a convolutional layer, all without an ML framework. Built as a team.

Size

516 lines of code.

Next project
Library lending system
Keep going
CV Contact