Traffic Light Classifier

This project is to train a traffic light classifier end-to-end for testing on Lincoln MKZ. The CNN trained is a SSD model for traffic light state classification using learning transfer from pre-trained MobileNet models using Google TensorFlow Object Detection API.

Since it is an end-to-end CNN training, the following key stages are involved:

• Dataset collection and purify
• Labeling and record file generation
• Model preparsion and training
• Model graph export
• Inference and validation

Detailed information could be found here.