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.
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