Abstract Much of the recent progress made in image classifification research can be credited to training procedure refifinements, such as changes in data augmentations and optimization methods. In the literature, however, most refifinements are either brieflfly mentioned as implementation details or only visible in source code. In this paper, we will examine a collection of such refifinements and empirically evaluate their impact on the fifinal model accuracy through ablation study. We will show that, by combining these refifinements together, we are able to improve various CNN models signifificantly. For example, we raise ResNet-50’s top-1 validation accuracy from 75.3% to 79.29% on ImageNet. We will also demonstrate that improvement on image classifification accuracy leads to better transfer learning performance in other application domains such as object detection and semantic segmentation.