Random Erasing Data Augmentation

Zhong, Zhun; Zheng, Liang; Kang, Guoliang; Li, Shaozi; Yang, Yi

doi:10.1609/aaai.v34i07.7000

Cited by 2,385 publications

(1,305 citation statements)

References 15 publications

Supporting

Mentioning

1,181

Contrasting

Unclassified

Order By: Relevance

“…Krizhevsky et al [9] adjusted RGB channels, transformed the image, and made horizontal reflections to reduce overfitting. Zhong et al [19] performed data augmentation using a method of erasing part of an image and filling the erased part with random values. Similarly, DeVries and Taylor [25] removed certain parts, but proposed a method to fill the erased parts with zeros instead of random values.…”

Section: Data Augmentationmentioning

confidence: 99%

See 1 more Smart Citation

Self-Supervised Contextual Data Augmentation for Natural Language Processing

Park

Ahn

2019

Symmetry

View full text Add to dashboard Cite

In this paper, we propose a novel data augmentation method with respect to the target context of the data via self-supervised learning. Instead of looking for the exact synonyms of masked words, the proposed method finds words that can replace the original words considering the context. For self-supervised learning, we can employ the masked language model (MLM), which masks a specific word within a sentence and obtains the original word. The MLM learns the context of a sentence through asymmetrical inputs and outputs. However, without using the existing MLM, we propose a label-masked language model (LMLM) that can include label information for the mask tokens used in the MLM to effectively use the MLM in data with label information. The augmentation method performs self-supervised learning using LMLM and then implements data augmentation through the trained model. We demonstrate that our proposed method improves the classification accuracy of recurrent neural networks and convolutional neural network-based classifiers through several experiments for text classification benchmark datasets, including the Stanford Sentiment Treebank-5 (SST5), the Stanford Sentiment Treebank-2 (SST2), the subjectivity (Subj), the Multi-Perspective Question Answering (MPQA), the Movie Reviews (MR), and the Text Retrieval Conference (TREC) datasets. In addition, since the proposed method does not use external data, it can eliminate the time spent collecting external data, or pre-training using external data.

show abstract

Section: Data Augmentationmentioning

confidence: 99%

“…In order to alleviate the aforementioned problem, many studies take advantage of augmentation of the existing data [9,[19][20][21][22][23][24]. Data augmentation is used for small amounts of data, but also unbalanced data.…”

Section: Introductionmentioning

confidence: 99%

Self-Supervised Contextual Data Augmentation for Natural Language Processing

Park

Ahn

2019

Symmetry

View full text Add to dashboard Cite

show abstract

“…An over-fitting problem might also happen, which refers to a model that is too closely fit to the training data and leads to the poor generalization of test data. Data augmentation and regularization methods that generate pseudo samples to increase the amount of training data and enrich the diversity of the training data were proposed to reduce the risk of over-fitting [23,26,30,31]. Niall et al [30] introduced data augmentation by changing the background to generate samples.…”

Section: Related Workmentioning

confidence: 99%

“…Three kinds of deep models are used in person ReID: Siamese networks [13,15,20], classification networks [21][22][23], and triplet networks [24,25]. Additionally, many works have focused on coping with specific issues, such as occlusion [26], misalignment [27][28][29] and over-fitting [19,23,30,31].…”

Section: Introductionmentioning

confidence: 99%

Person Search via Deep Integrated Networks

Chen

et al. 2019

Applied Sciences

View full text Add to dashboard Cite

This study proposes an integrated deep network consisting of a detection and identification module for person search. Person search is a very challenging problem because of the large appearance variation caused by occlusion, background clutter, pose variations, etc., and it is still an active research issue in the academic and industrial fields. Although various studies have been proposed, following the protocols of the person re-identification (ReID) benchmarks, most existing works take cropped pedestrian images either from manual labelling or a perfect detection assumption. However, for person search, manual processing is unavailable in practical applications, thereby causing a gap between the ReID problem setting and practical applications. One fact is also ignored: an imperfect auto-detected bounding box or misalignment is inevitable. We design herein a framework for the practical surveillance scenarios in which the scene images are captured. For person search, detection is a necessary step before ReID, and previous studies have shown that the precision of detection results has an influence on person ReID. The detection module based on the Faster R-CNN is used to detect persons in a scene image. For identifying and extracting discriminative features, a multi-class CNN network is trained with the auto-detected bounding boxes from the detection module, instead of the manually cropped data. The distance metric is then learned from the discriminative features output by the identification module. According to the experimental results of the test performed in the scene images, the multi-class CNN network for the identification module can provide a 62.7% accuracy rate, which is higher than that for the two-class CNN network.

show abstract

“…Generally speaking, we used the UNIT [11] (Unsupervised Image-to-image Translation) framework as the basis of our overall translation works and then treated a full image as a combination of instance and background domains. First, identity matrices were used to record location information of instance domains that needed to be translated individually, and we successfully segregated the instances from the given image and assigned the original location (instance areas) with the mean pixel value of the original image [13]. Next, we translated the segregated background and instance domains, respectively.…”

Section: Introductionmentioning

confidence: 99%

DAGAN: A Domain-Aware Method for Image-to-Image Translations

Yin

Shin

2020

Complexity

View full text Add to dashboard Cite

The image-to-image translation method aims to learn inter-domain mappings from paired/unpaired data. Although this technique has been widely used for visual predication tasks—such as classification and image segmentation—and achieved great results, we still failed to perform flexible translations when attempting to learn different mappings, especially for images containing multiple instances. To tackle this problem, we propose a generative framework DAGAN (Domain-aware Generative Adversarial etwork) that enables domains to learn diverse mapping relationships. We assumed that an image is composed with background and instance domain and then fed them into different translation networks. Lastly, we integrated the translated domains into a complete image with smoothed labels to maintain realism. We examined the instance-aware framework on datasets generated by YOLO and confirmed that this is capable of generating images of equal or better diversity compared to current translation models.

show abstract

Random Erasing Data Augmentation

Cited by 2,385 publications

References 15 publications

Self-Supervised Contextual Data Augmentation for Natural Language Processing

Self-Supervised Contextual Data Augmentation for Natural Language Processing

Person Search via Deep Integrated Networks

DAGAN: A Domain-Aware Method for Image-to-Image Translations

Contact Info

Product

Resources

About