CNNs on the GTSRB

A short exploration of CNN architectures/papers for German Traffic Sign Recognition Benchmark (GTSRB)

Baseline Architectures

• LeNet5: LeCun, Yann, et al. "Gradient-based learning applied to document recognition." Proceedings of the IEEE 86.11 (1998): 2278-2324.
• VGG16: Simonyan, Karen, and Andrew Zisserman. "Very deep convolutional networks for large-scale image recognition." arXiv preprint arXiv:1409.1556 (2014).

Papers

• Sermanet, Pierre, and Yann LeCun. "Traffic sign recognition with multi-scale convolutional networks." Neural Networks (IJCNN), The 2011 International Joint Conference on. IEEE, 2011.
• Ciresan, Dan, et al. "Multi-column deep neural network for traffic sign classification." Neural Networks 32 (2012): 333-338.

Implementation Details

Each pipeline was implemented using OpenCV 3.x and Tensorflow on python. This implementation can be extended for the evaluation of other models/papers just by following the presented modular design (model/pipeline/preprocess)and accordingly adjusting review.py to dynamically load the appropiate python module.

In this initial version, each paper/architecture evaluated include:

1. Pre-processing techniques like adaptive histogram equalization, intensity re-scaling, contrast normalization, histogram equalization, etc., based on open-cv and sk-image default implementations.
2. Model creation using Tensorflow primitives.
3. Training using hyper-parameters as described on each paper (for ImageNet or GTSRB).

For all models, a live plot is presented with the validation accuracy after each training epoch.