Belgium Traffic Signs Classification

Develop deep learning architectures for classifying Belgium traffic signs. Goals of this project are as follows -

1. Create structured and automated workflow of classification for reproducible research.
2. Learn and play with convolutional neural networks in TensorFlow

Download data

1. Download training data - http://btsd.ethz.ch/shareddata/BelgiumTSC/BelgiumTSC_Training.zip.

2. Download testing data - http://btsd.ethz.ch/shareddata/BelgiumTSC/BelgiumTSC_Testing.zip

3. './data/download_data.py' will download data and create './data/Training' and './data/Testing'

4. Create empty .gitkeep files (Only if this repo is not being cloned) echo $null>>.\data\Training.gitkeep $null>>.\data\Testing.gitkeep

Git does not store empty directories. .gitkeep enforces directory persistence.

3. Create .gitignore echo $null >> .gitignore

References

1. Deep MNIST TensorFlow tutorial - https://github.com/tensorflow/tensorflow/blob/r1.2/tensorflow/examples/tutorials/mnist/mnist_deep.py
2. Blog post - https://beckernick.github.io/neural-network-scratch/
3. Tutorial for Belgium TS data set - https://www.datacamp.com/community/tutorials/tensorflow-tutorial#gs.C4=SPAQ
4. Belgium TS data set - http://btsd.ethz.ch/shareddata/
5. Data extraction and reading - https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/learn/python/learn/datasets/mnist.py
6. Code structure: an Object pattern for TensorFlow - https://github.com/wpm/tf_model_session
7. Radu Timofte*, Markus Mathias*, Rodrigo Benenson, and Luc Van Gool, Traffic Sign Recognition - How far are we from the solution?, International Joint Conference on Neural Networks (IJCNN 2013), August 2013, Dallas, USA. (* equal contributions)