Texture Synthesis Using Convolutional Neural Networks

By Leon A. Gatys, Alexander S. Ecker, Matthias Bethge. 2015

1,451 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Generative Models

Citations in the corpus

ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness (2018, background): “…such as style transfer (Gatys et al., 2016), single image super-resolution (Gondal et al., 2018) as well as static and dynamic texture synthesis (Gatys et al., 2015; Funke et al., 2017) all produce excellent results using standard CNNs, while CNN-based shape transfer seems to be very difﬁcult…”
Regionwise Generative Adversarial Image Inpainting for Large Missing Areas (2019, background): “the perceptual loss [8] or style loss [9].”
EdgeConnect: Generative Image Inpainting with Adversarial Edge Learning (2019, neutral): “This is similar to perceptual loss [24, 16, 15], where activation maps are compared with those from the pre-trained VGG network.”
Image-to-Image Translation with Conditional Adversarial Networks (2016, methodology): “rkov random ﬁeld, assuming independence between pixels separated by more than a patch diameter. This connection was previously explored in [38], and is also the common assumption in models of texture [17,21] and style [16,25,22,37]. Therefore, our PatchGAN can be understood as a form of texture/styl”
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (2016, positive): “Perceptually more convincing results were obtained for both super-resolution and artistic style-transfer [15, 16].”
Semantic Image Inpainting with Deep Generative Models (2016, methodology): “In earlier work, back-propagation to augment data has been used for texture synthesis and style transfer [8, 7, 20].”
A note on the evaluation of generative models (2015, background): “It is therefore unsurprising that other approaches than density estimation are much more effective for image synthesis (Portilla & Simoncelli, 2000; Dosovitskiy et al., 2015; Gatys et al., 2015).”
Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNs (2021, neutral): “Gatys et al. [10] ﬁrst mentioned absolute position in the latent representations of CNNs in a footnote which comments on how zero padding effects generative modelling.”
Perceptual Losses for Real-Time Style Transfer and Super-Resolution (2016, methodology): “[10,11] propose the following style reconstruction loss.”
A Learned Representation For Artistic Style (2016, methodology): “The second point is motivated by the observation that the artistic style of a painting may be interpreted as a visual texture (Gatys et al., 2015a).”
Amortised MAP Inference for Image Super-resolution (2016, methodology): “Using textures is also motivated by the Julesz conjecture (Julesz, 1962; Gatys et al., 2015), the assumption that the human visual system is sensitive only to a certain set of statistics when distinguishing between textures.”
Stacked Generative Adversarial Networks (2016, methodology): “[14, 13, 61, 26] use an additional “style loss” which minimizes the L2 distance between Gram matrices on some representation space.”
Attribute2Image: Conditional Image Generation from Visual Attributes (2015, methodology): “ion to the variable z, which we solve by the ADAM method [13]. Our algorithm actually shares a similar spirit with recently proposed neural network visualization [42] and texture synthesis algorithms [6]. The dierence is that we use generation models for recognition while their algorithms use recogn”
EnhanceNet: Single Image Super-Resolution Through Automated Texture Synthesis (2016, methodology): “CNNs have also been found useful for the task of texture synthesis [15] and style transfer [16, 24, 53], however these methods are constrained to the setting of a single network learning to produce only a single texture and have so far not been applied to SISR.”
ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (2018, background): “o helps to improve the perceptual quality. Perceptual-driven approaches have also been proposed to improve the visual quality of SR results. Based on the idea of being closer to perceptual similarity [29,14], perceptual loss [13] is proposed to enhance the visual quality by minimizing the error in a”
Image2StyleGAN: How to Embed Images Into the StyleGAN Latent Space? (2019, background): “[6, 7] observed that the learned filters of the VGG image classification model [20] are excellent general-purpose feature extractors and proposed to use the covariance statistics of the extracted features to measure the high-level similarity between images perceptually, which is then formalized as t”