Overview
This personal research project explores neuromorphic computing applied to bird classification. Instead of using conventional frame-based video, the pipeline converts standard RGB video into event-based data — mimicking the output of a Dynamic Vision Sensor (DVS) — and feeds it into a Spiking Neural Network (SNN) classifier.
Pipeline
- Video Input — standard RGB bird footage
- Event Conversion — RGB frames are processed through the IEBCS DVS simulator to generate asynchronous event streams
- SNN Training — event data is fed into an SNN built with Norse (a PyTorch-based SNN library) for species classification
Why Neuromorphic?
DVS cameras and SNNs are inherently temporal and energy-efficient. They encode motion and change rather than full frames, making them well-suited for fast-moving biological subjects like birds. This project tests whether event-based representations can achieve competitive accuracy on a naturalistic classification task.
Tech Stack
- Python — core implementation
- PyTorch — deep learning backbone
- Norse — spiking neural network layers and dynamics
- OpenCV — video preprocessing
- NumPy — numerical operations
- IEBCS — event-based camera simulator