Survey: Image Mixing and Deleting for Data Augmentation

Abstract

Data augmentation has been widely used to improve deep nerual networks performance. Numerous approaches are suggested, for example, dropout, regularization and image augmentation, to avoid over-ftting and enhancing generalization of neural networks. One of the sub-area within data augmentation is image mixing and deleting. This specific type of augmentation either mixes two images or delete image regions to hide or make certain characteristics of images confusing for the network to force it to emphasize on overall structure of object in image. The model trained with this approach has shown to perform and generalize well as compared to one trained without imgage mixing or deleting. Additional benefit achieved with this method of training is robustness against image corruptions. Due to its low compute cost and success in recent past, many techniques of image mixing and deleting are proposed. This paper provides detailed review on these devised approaches, dividing augmentation strategies in three main categories cut and delete, cut and mix and mixup. The second part of paper emprically evaluates these approaches for image classification, finegrained image recognition and object detection where it is shown that this category of data augmentation improves the overall performance for deep neural networks.

ChessMix: Spatial Context Data Augmentation for Remote Sensing Semantic Segmentation

“…ChessMix is a data augmentation method created especially for the scenario of semantic segmentation with remote sensing data. It can be classified as part of the over-sampling category [7], more specifically under the "cut and mix" [29] subcategory. ChessMix creates synthetic images by collecting image mini-patches from the different training samples available for the problem and placing them with a transformation in the new image with a chessboard-like grid pattern.…”

“…As for mixing images methods, cut and mix approaches are the closest to the proposed method, since they replace selected regions with some regions of other images [29]. CutMix [30] is an augmentation method for image classification that samples images coordinates and replaces the selected patch with a patch from another random image from the mini-batch during training.…”

A Comprehensive Survey of Image Augmentation Techniques for Deep Learning

“…First, we do not confine ourselves to a specific type of image, such as face images [8]. Second, many types of image augmentation algorithms are included, instead of a specific type, such as only generative adversarial networks [9] and image mixing [10]. Third, we do not focus on one specific application, such as object detection [5].…”

Survey on Videos Data Augmentation for Deep Learning Models

“…Reviews on image data augmentation for DL models have already been published [19][20][21][22][23]. In their paper, Shorten et al [19] realised a complete survey on image data augmentation for DL, covering both basic image manipulation and DL approaches.…”

“…On the other hand, DL methods are evolving rapidly and will gain more popularity given their ability to tackle complex data synthesis problems such as cross-domain image synthesis. Image mixing and deleting data augmentation strategies are reviewed by Humza Naveed in his survey [23]. The reviewed papers are split into three categories: erasing image patches; cut image regions and replacing them with patches from other images; and mixing multiple images.…”