Neural style transfer based on deep feature synthesis

Li, Da‐Jin; Gao, Wenran

doi:10.1007/s00371-022-02664-2

Cited by 10 publications

(17 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The results of PSGAN-GD are poor, which is essentially due to the instability of the inversion procedure in the latent space. [8] 135 ± 468 178 ± 190 0.56 ± 0.09 0.085 WCT [12] 29.8 ± 132.9 10 ± 18 0.44 ± 0.07 3.5 PSGAN-GD [2] 208 ± 611 381 ± 607 0.64 ± 0.11 24…”

Section: Visual and Quantitative Resultsmentioning

confidence: 99%

“…-TextureCNN [4], as one of our goals is to shortcut this process with a feedforward auto-encoder; -PSGAN [2], as they introduced the idea of augmenting a generator with periodic content; -Neural texture [8], since they also use a texture encoder; -Whitening Coloring Transform (WCT) [12], because this methods performs universal texture synthesis with no learning step.…”

Section: Assessed Methods Datasets and Evaluation Metricsmentioning

confidence: 99%

“…through latent spaces. The authors of [12] were the first to propose such an approach, leveraging the concept of adaptive instance normalization [9] to provide a flexible and universal style transfer and texture synthesis method, where each texture is represented by several thousands normalizing parameters related to VGG19 features. To allow for user control [1] proposes a modification of PSGAN [2] by introducing a convolutional encoder.…”

Section: Universality and Latent Representations Of Texturesmentioning

confidence: 99%

“…Therefore, methods have been proposed to perform synthesis using generative neural networks, such as networks learned for each new examplar [17] or GANs [10]. More recently, such approaches have been extended to perform the synthesis from a whole set of textures, by using adaptive normalizations of generative networks [12] [13].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A geometrically aware auto-encoder for multi-texture synthesis

Chatillon¹,

Gousseau²,

Lefèbvre³

2023

Preprint

View full text Add to dashboard Cite

We propose an auto-encoder architecture for multi-texture synthesis. The approach relies on both a compact encoder accounting for second order neural statistics and a generator incorporating adaptive periodic content. Images are embedded in a compact and geometrically consistent latent space, where the texture representation and its spatial organisation are disentangled. Texture synthesis and interpolation tasks can be performed directly from these latent codes. Our experiments demonstrate that our model outperforms state-of-the-art feed-forward methods in terms of visual quality and various texture related metrics.

show abstract

Section: Visual and Quantitative Resultsmentioning

confidence: 99%

Section: Assessed Methods Datasets and Evaluation Metricsmentioning

confidence: 99%

Section: Universality and Latent Representations Of Texturesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A geometrically aware auto-encoder for multi-texture synthesis

Chatillon¹,

Gousseau²,

Lefèbvre³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…1 Since then, the image transfer task has stepped into the wave of deep learning. [2][3][4][5][6][7] The advent of generative adversarial networks 8 (GANs) has improved the quality of images generated by feed-forward neural networks. 9 Based on the above advantages, a recurrent adversarial network is proposed to solve the problem of dataset mismatch.…”

Section: Introductionmentioning

confidence: 99%

AttCST: attention improves style transfer via contrastive learning

Zhou

Zou

2023

J. Electron. Imag.

View full text Add to dashboard Cite

.Arbitrary style transfer has flexible applicability and aims to learn the painting styles of different artists through one model training and re-render to everyday photos. Existing domain-to-domain style transfer methods based on generative adversarial networks achieve good results in image generation quality but need more flexibility in various style transfer tasks. In the image-to-image style transfer method, focusing on the fusion of content image and style image can learn local detail texture very well. Still, there are some defects in the preservation of the content structure. To alleviate the above two problems, we leverage both advantages, using attention improves style transfer via contrastive learning. Specifically, the attention mechanism is used to fuse the multi-scale features of the content and style images. The previous shallow features are fully utilized to supplement the texture information lost by the deep features. Then the currently learned style image is distinguished from other style images by contrastive learning modules, thereby enhancing the ability of the network to learn style images. We also derive a local loss based on the transferred blocks to improve the visual quality of the generated images. We conduct extensive qualitative and quantitative analyses, the results of which demonstrate the effectiveness of our approach.

show abstract