CNN Optimization For Raspberry PI 4

Optimizing a trained CNN for Inference on Raspberry PI 4. CNN used for classifying American Sign Language.

Data Sources

• ASL MNIST: https://www.kaggle.com/datasets/datamunge/sign-language-mnist

Methodology

• Data Processing and Training Baseline CNN: Standard preprocessing techniques for MNIST (normalizing)
• Model Optimization:
  • Unstructured Global Pruning
  • 8-bit Quantization
• Model Conversion to TFLite format

Technologies and Tools

• Keras
• Tensorflow Optimization Library
• Google Colab For Training
• Raspberry Pi 4 Model B 2GB for Inference

Team Members

• John Olusetire
• Jake Fitzgerald