On-Device-Deep Learning

On-Device-Deep Learning is a recent project that developed a custom deep learning gesture recognition network, specifically tailored for Raspberry Pi. The project utilized the Jester dataset and Pytorch, focusing on efficiency and performance, enhancing the model's architecture with channel-based pruning and 4-bit K-means quantization, achieving a near-optimal balance with accuracy minimally reduced from 71.7% to 69.8%.