German Traffic Sign Recognition Benchmark (GTSRB)

This repository aims to implement classifiers networks models from the GTSRB challenge.

Getting started

Clone this repository:

git clone https://github.com/raymas/German-Traffic-Sign-Recognition-Benchmark.git

Prior training networks, there are prerequisites.

Docker installation (Work in progress)

Simply build using the provided Dockerfile:

docker build -t gtsrb-nn .
docker run -it gtsrb-nn [args]

See How to use for correct list of arguments. Please note this docker image is using the GPU version of tensorflow.

Conda environnement

Create a new virtual environnement using the lastest tensorflow packages (GPU or not) from anaconda.

conda create -c anaconda tensorflow[-gpu] pip

PS: please remove the '[]' if you want to have to gpu acceleration or simply delete '[-gpu]' for cpu only tensorflow.

Install the required softwares:

pip install -r requirements.txt

How to use

Train and test by launching one of the provided model:

python main.py --model DKS --train

For help:

python main.py -h

With docker :

docker run -it gtsrb-nn --model DKS --train

Results

DeepKnowledgeSeville

After 30 epochs :

Accuracy : ~ 95%
Loss : 0.07

Loss	Accuracy

TODO

Spatial extractor.

Contributing

Fork, publish a new branch, add your name to CONTRIBUTORS.md

Sources

GTSRB J. Stallkamp, M. Schlipsing, J. Salmen, and C. Igel. The German Traffic Sign Recognition Benchmark: A multi-class classification competition. In Proceedings of the IEEE International Joint Conference on Neural Networks, pages 1453–1460. 2011.

DeepKnowledge Seville CNN with 3 Spatial Transformers, DeepKnowledge Seville, Álvaro Arcos-García and Juan A. Álvarez-García and Luis M. Soria-Morillo, Neural Networks

Sermanet Multi-Scale CNNs, sermanet , Traffic sign recognition with multi-scale Convolutional Networks, Traffic sign recognition with multi-scale Convolutional Networks, P. Sermanet, Y. LeCun, August 2011, International Joint Conference on Neural Networks (IJCNN) 2011