A Full Stage Data Augmentation Method in Deep Convolutional Neural Network for Natural Image Classification

Abstract

Nowadays, deep learning has achieved remarkable results in many computer vision related tasks, among which the support of big data is essential. In this paper, we propose a full stage data augmentation framework to improve the accuracy of deep convolutional neural networks, which can also play the role of implicit model ensemble without introducing additional model training costs. Simultaneous data augmentation during training and testing stages can ensure network optimization and enhance its generalization ability. Augmentation in two stages needs to be consistent to ensure the accurate transfer of specific domain information. Furthermore, this framework is universal for any network architecture and data augmentation strategy and therefore can be applied to a variety of deep learning based tasks. Finally, experimental results about image classification on the coarse-grained dataset CIFAR-10 (93.41%) and fine-grained dataset CIFAR-100 (70.22%) demonstrate the effectiveness of the framework by comparing with state-of-the-art results.

Fighting fire with fire: A spatial–frequency ensemble relation network with generative adversarial learning for adversarial image classification

“…Antoniou et al 36 proposed to train a conditional GAN (DAGAN) to perform data augmentation. It is also worth mentioning that regularized deep learning [37][38][39][40][41] is an efficient and vital way to improve the generalization ability: the full stage data augmentation 37 plays the role of the implicit model ensemble without introducing additional model training costs; PReLU 38 is a new activation function to improve the classification performance with a fast convergence rate; LLb-SGD 40 is a simple gradient-based optimization method with computational efficiency; two-stage training method 39 could regularize the feature boundaries of deep networks from the point of view of data punishment so as to improve the generalization ability of the networks; drop-path 41 could reduce model parameters of deep networks and accelerate the network inference. On the other hand, MetaGAN 42 is a simplistic and versatile framework for improving the performance of few-shot learning models, based on the idea that generators generate fake samples that help the classifier to understand more explicit decision boundaries between different categories from several samples.…”

An intelligent nonintrusive load monitoring scheme based on 2D phase encoding of power signals

“…Moreover, aiming at accelerating the learning process, a pruning scheme is introduced in Reference [38], in which the model parameters of 2D deep CNN are reduced. By Zheng et al, 39 the performance of CNN models is improved using full‐stage information augmentation strategy. This results in an implicit model ensemble that does not require extra model training costs.…”

Damage Analysis in Dual-Phase Steel Using Deep Learning: Transfer from Uniaxial to Biaxial Straining Conditions by Image Data Augmentation

“…Data augmentation is a general technique to improve both the generalizability and convergence of neural networks, 10,17 enhancing the training and convergence of the neural networks, 17 since it tackles the qualitative and quantitative limitations of the employed data. 10,25,26 Data augmentation techniques can be loosely grouped into two categories: The more traditional approach 17,27 is based on methods such as image rotation, shearing, cropping, translation, color transformation, etc., whereas generative adversarial networks (GANs) 10,25,28 show great promise to generate ''fake'' images.…”