A New Perspective on Stabilizing GANs Training: Direct Adversarial Training

Li, Ziqiang; Xia, Pengfei; Tao, Rentuo; Niu, Hongjing; Li, Bin

doi:10.1109/tetci.2022.3193373

Cited by 21 publications

(7 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…LeakyReLU activation functions are utilized in all layers except for the last deconvolutional layer, where the activation function is Tanh. Notably, spectral normalization is applied to the weight matrix of each convolutional layer in D. In doing this, the model becomes more robust against input perturbations, enhancing the overall model robustness [29].…”

Section: Sngan-based Pct Approachmentioning

confidence: 99%

Generative Deep Learning-Based Thermographic Inspection of Artwork

Liu

Wang

Jiang

et al. 2023

Sensors

View full text Add to dashboard Cite

Infrared thermography is a widely utilized nondestructive testing technique in the field of artwork inspection. However, raw thermograms often suffer from problems, such as limited quantity and high background noise, due to limitations inherent in the acquisition equipment and experimental environment. To overcome these challenges, there is a growing interest in developing thermographic data enhancement methods. In this study, a defect inspection method for artwork based on principal component analysis is proposed, incorporating two distinct deep learning approaches for thermographic data enhancement: spectral normalized generative adversarial network (SNGAN) and convolutional autoencoder (CAE). The SNGAN strategy focuses on augmenting the thermal images, while the CAE strategy emphasizes enhancing their quality. Subsequently, principal component thermography (PCT) is employed to analyze the processed data and improve the detectability of defects. Comparing the results to using PCT alone, the integration of the SNGAN strategy led to a 1.08% enhancement in the signal-to-noise ratio, while the utilization of the CAE strategy resulted in an 8.73% improvement.

show abstract

Section: Sngan-based Pct Approachmentioning

confidence: 99%

Generative Deep Learning-Based Thermographic Inspection of Artwork

Liu

Wang

Jiang

et al. 2023

Sensors

View full text Add to dashboard Cite

show abstract

“…Unfortunately, owing to the unstable training of GANs [ 40 , 41 , 42 , 43 ], unsuitable samples are inevitably generated. It is worth noting that the distributions of unsuitable samples do not properly match the distribution of the real data.…”

Section: Preliminariesmentioning

confidence: 99%

“…By generating virtual data that resemble actual data, the GAN enlarges the sample capacity to enhance the prediction performance. Although various improvements have been made in GANs, including alternative loss functions and training strategies, the training process remains unstable [ 40 , 41 , 42 , 43 ]. Therefore, the quality of the generated samples remains uncertain.…”

Section: Introductionmentioning

confidence: 99%

Enhanced Soft Sensor with Qualified Augmented Samples for Quality Prediction of the Polyethylene Process

Dai

Liu

Meng³

et al. 2022

Polymers

View full text Add to dashboard Cite

Data-driven soft sensors have increasingly been applied for the quality measurement of industrial polymerization processes in recent years. However, owing to the costly assay process, the limited labeled data available still pose significant obstacles to the construction of accurate models. In this study, a novel soft sensor named the selective Wasserstein generative adversarial network, with gradient penalty-based support vector regression (SWGAN-SVR), is proposed to enhance quality prediction with limited training samples. Specifically, the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) is employed to capture the distribution of the available limited labeled data and to generate virtual candidates. Subsequently, an effective data-selection strategy is developed to alleviate the problem of varied-quality samples caused by the unstable training of the WGAN-GP. The selection strategy includes two parts: the centroid metric criterion and the statistical characteristic criterion. An SVR model is constructed based on the qualified augmented training data to evaluate the prediction performance. The superiority of SWGAN-SVR is demonstrated, using a numerical example and an industrial polyethylene process.

show abstract

“…To reduce this problem, previous researches propose many methods which can be divided into two categories. The first kind of methods focus on the optimization process and divergence metrics [21, 25, 31–35] to stabilize the training process and mitigate the mode collapse problem. Salimans et al.…”

Section: Related Workmentioning

confidence: 99%

“…To reduce this problem, previous researches propose many methods which can be divided into two categories. The first kind of methods focus on the optimization process and divergence metrics [21,25,[31][32][33][34][35] [39] to capture more modes of the distribution. ModeGAN [40] and VEEGAN [41] utilize additional encoder networks to enforce the bijection mapping between the input noise vectors and generated images.…”

Section: Reducing Mode Collapsementioning

confidence: 99%

GSGAN: Learning controllable geospatial images generation

Xingzhe

Lin

Zheng

et al. 2022

IET Image Processing

View full text Add to dashboard Cite

Compared with natural images, geospatial images cover larger area and have more complex image contents. There are few algorithms for generating controllable geospatial images, and their results are of low quality. In response to this problem, this paper proposes Geospatial Style Generative Adversarial Network to generate controllable and high‐quality geospatial images. Current conditional generators suffer the mode collapse problem in geospatial field. The problem is addressed via a modified mode seeking regularization term with contrastive learning theory. Besides, the discriminator network architecture is modified to process global feature information and texture information of geospatial images. Feature loss in the generator is introduced to stabilize the training process and improve generated image quality. Comprehensive experiments are conducted on UC Merced Land Use Dataset, NWPU‐RESISC45 Dataset, and AID Dataset to evaluate all compared methods. Experiment results show our method outperforms state‐of‐the‐art models. Our method not only generates high‐quality and controllable geospatial images, but also enhances the discriminator to learn better representations.

show abstract

A New Perspective on Stabilizing GANs Training: Direct Adversarial Training

Cited by 21 publications

References 25 publications

Generative Deep Learning-Based Thermographic Inspection of Artwork

Generative Deep Learning-Based Thermographic Inspection of Artwork

Enhanced Soft Sensor with Qualified Augmented Samples for Quality Prediction of the Polyethylene Process

GSGAN: Learning controllable geospatial images generation

Contact Info

Product

Resources

About