【无标题】论文阅读 [TPAMI-2022] A Geometrical Perspective on Image Style Transfer With Adversarial Learning

论文阅读 [TPAMI-2022] A Geometrical Perspective on Image Style Transfer With Adversarial Learning

论文搜索(studyai.com)

搜索论文: A Geometrical Perspective on Image Style Transfer With Adversarial Learning

搜索论文: http://www.studyai.com/search/whole-site/?q=A+Geometrical+Perspective+on+Image+Style+Transfer+With+Adversarial+Learning

关键字(Keywords)

Gallium nitride; Manifolds; Task analysis; Generators; Face; Geometry; Analytical models; Generative adversarial learning; unsupervised learning theory; generalization theory; machine learning

机器学习; 机器视觉

监督学习; 无监督学习; 生成对抗; 风格迁移

摘要(Abstract)

Recent years witness the booming trend of applying generative adversarial nets (GAN) and its variants to image style transfer.

近年来，生成性对抗网络及其变体在图像风格转换中的应用呈现出蓬勃发展的趋势。.

Although many reported results strongly demonstrate the power of GAN on this task, there is still little known about neither the interpretations of several fundamental phenomenons of image style transfer by generative adversarial learning, nor its underlying mechanism.

尽管许多报道的结果有力地证明了GAN在这项任务中的作用，但对于生成性对抗学习对图像风格转换的几个基本现象的解释，以及其潜在机制，仍然知之甚少。.

To bridge this gap, this paper presents a general framework for analyzing style transfer with adversarial learning through the lens of differential geometry.

为了弥补这一差距，本文从微分几何的角度提出了一个用对抗性学习分析风格转换的一般框架。.

To demonstrate the utility of our proposed framework, we provide an in-depth analysis of Isola et al.’s pioneering style transfer model pix2pix [1] and reach a comprehensive interpretation on their major experimental phenomena.

为了证明我们提出的框架的实用性，我们对Isola等人开创性的转移模型pix2pix[1]进行了深入分析，并对其主要实验现象进行了全面解释。.

Furthermore, we extend the notion of generalization to conditional GAN and derive a condition to control the generalization capability of the pix2pix model.

此外，我们将泛化的概念扩展到条件GAN，并导出了控制pix2pix模型泛化能力的条件。.

From a higher viewpoint, we further prove a learning-free condition to guarantee the existence of infinitely many perfect style transfer mappings.

从更高的角度，我们进一步证明了一个无学习条件，以保证无穷多个完美样式转移映射的存在。.

Besides, we also provide a number of practical suggestions on model design and dataset construction based on these derived theoretical results to facilitate further researches…

此外，我们还根据这些理论结果对模型设计和数据集构建提出了一些实用建议，以便于进一步的研究。。.

作者(Authors)

[‘Xudong Pan’, ‘Mi Zhang’, ‘Daizong Ding’, ‘Min Yang’]