Uncertainty Quantification for Deep Neural Networks: An Empirical Comparison and Usage Guidelines