On Improving Robustness Of Deep Neural Networks For Computer Vision