Classify Traffic Signs from German Traffic Sign Data set. Each image is 32x32 pixels and belongs to one of 43 classes. Training and inference are achieved using Keras with either Tensorflow or Theano as backend. The data set (124 MB) is downloaded automatically and consists of three parts: train, valid, test.

You can start by running the inference script to make sure that prerequisites are correctly installed. Commands should be run in Terminal (macOS/Linux) or Command Prompt (Windows) unless otherwise specified.

Perform inference using existing pre-trained model.

python inference.py

Train new model from scratch.

python train.py

In project root directory, run

tensorboard --logdir=logdir

In browser, navigate to

http://localhost:6006

Version numbers below are of confirmed working releases for this project.

python 3.6.5
keras 2.2.2   
numpy 1.14.3
pandas 0.23.4
pillow 5.1.0
scikit-learn 0.19.1
scipy 1.1.0
tensorflow 1.10.1
tensorflow-tensorboard 1.10.0
tqdm 4.26.0

It is recommended to use a virtual environment so that python packages can be easily managed. Instructions for installation using Anaconda will make it easier to prepare your environment for this project.

1. Install Anaconda Python 3

2. Add Anaconda directories to PATH as necessary (e.g. for Windows: Anaconda3, Anaconda3\Scripts)

3. Create environment conda create -n traffic_signs

4. Activate environment

   • Windows: activate traffic_signs
   • Mac/Linux: source activate traffic_signs

5. Install packages

   pip install keras scikit-learn tensorflow pandas tqdm pillow

6. [Optional] Save model diagram (Ubuntu, macOS)

    macOS: brew install graphviz
    Ubuntu: apt install graphviz
    pip install graphviz pydot