Trained model weights, training and evaluation code from the paper "A simple way to make neural networks robust against diverse image corruptions" - View it on GitHub