Building a Bayesian neural network

For this project, we will use the German Traffic Sign Dataset (http://benchmark.ini.rub.de/?section=gtsrb&subsection=dataset) to build a Bayesian neural network. The training dataset contains 26,640 images in 43 classes. Similarly, the testing dataset contains 12,630 images.

Note

Please read the README.md file in this book's repository before executing the code to install the appropriate dependencies and for instructions on how to run the code.

The following is an image that's present in this dataset: 

You can see that there are different kinds of traffic sign depicted by different classes in the dataset.

We begin by pre-processing our dataset and making it conform to the requirements of the learning algorithm. This is done by reshaping the images to a uniform size via histogram equalization, which is used to enhance contrast, and cropping them to only focus on the traffic signs in the image. Also, we convert the images to grayscale as traffic signs...