Traffic sign classifier

Machine learning course project. We build a traffic sign classifier with multi-scale Convolutional Networks using Keras starting from this publication, obtaining the result of 99.33% in the GTSRB dataset version online-competition (the same used on the publication) improving the accuracy on test-set.

You can find all the details on this Jupiter notebook written in Italian only :'( .

Requirements

If you want to use Docker read the docker/README.md, otherwise you need on your system:

• python3
• jupyter
• numpy
• pandas
• scikit-learn
• matplotlib
• scikit-image
• glob2
• opencv
• keras

Usage

cd scripts
python preprocessing.py --help
python training.py --help
python test_on_testset.py --help
python test_on_new.py --help

python training.py --augmentation --blur --epochs 40 -- dataset online