Traffic Light Recognition Network

This model was created for the nexar challenge 1. It reached 93.6% accuracy on the challenge test data.

This work is based on the squeezeNet model and it's kera's implementation.

Using

in order to use the model to classify image to 3 classes:

• no traffic light
• red light
• green light

just edit the variable TEST_FOLDER in pred.py file. then:

python pred.py

it will create csv file: results.csv

Auther

Guy Hadash