“…Deep neural networks have shown tremendous success in image classification [38,24], object detection [37,36], and semantic segmentation [3,57]. These networks are data hungry with millions of parameters that make them prone to overfitting [28,53,34]. In this regard, many approaches have been suggested to avoid overfitting, for example, regularization [27], dropout [43], and data augmentations [40] like image rotation, random cropping, jittering and etc.…”