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