[论文解读] Generative Adversarial Networks: A Survey and Taxonomy.

Zhengwei Wang, Qi She+1 more|arXiv (Cornell University)|Jun 4, 2019

Generative Adversarial Networks and Image Synthesis参考文献 49被引用 67

一句话总结

本文全面综述并构建了生成对抗网络（GAN）的分类体系，聚焦于其在计算机视觉领域应对三大核心挑战的进展：高质量图像生成、生成多样性以及训练稳定性。文章回顾了主流的 GAN 架构与损失函数，基于关键应用领域中的实证性能，提供批判性分析与未来研究方向。

ABSTRACT

Generative adversarial networks (GANs) have been extensively studied in the past few years. Arguably their most significant impact has been in the area of computer vision where great advances have been made in challenges such as plausible image generation, image-to-image translation, facial attribute manipulation and similar domains. Despite the significant successes achieved to date, applying GANs to real-world problems still poses significant challenges, three of which we focus on here. These are: (1) the generation of high quality images, (2) diversity of image generation, and (3) stable training. Focusing on the degree to which popular GAN technologies have made progress against these challenges, we provide a detailed review of the state of the art in GAN-related research in the published scientific literature. We further structure this review through a convenient taxonomy we have adopted based on variations in GAN architectures and loss functions. While several reviews for GANs have been presented to date, none have considered the status of this field based on their progress towards addressing practical challenges relevant to computer vision. Accordingly, we review and critically discuss the most popular architecture-variant, and loss-variant GANs, for tackling these challenges. Our objective is to provide an overview as well as a critical analysis of the status of GAN research in terms of relevant progress towards important computer vision application requirements. As we do this we also discuss the most compelling applications in computer vision in which GANs have demonstrated considerable success along with some suggestions for future research directions. Code related to GAN-variants studied in this work is summarized on this https URL.

研究动机与目标

• 解决在实际计算机视觉应用中生成高质量、多样化且稳定的 GAN 输出的长期挑战。
• 通过聚焦实际挑战而非仅理论新颖性，批判性评估 GAN 研究的最新进展。
• 基于架构与损失函数的变体，构建结构化分类体系，以组织和比较现有 GAN 变体。
• 识别在图像生成、图像到图像翻译以及面部属性操控等关键计算机视觉任务中表现最成功的 GAN 方法。
• 基于当前 GAN 实际部署进程中的差距，提出未来研究方向。

提出的方法

• 本文系统性回顾了关于 GAN 的已发表科学文献，重点聚焦于针对特定挑战的架构与损失函数。
• 基于架构设计（例如，条件 GAN、StyleGAN、BigGAN）与损失函数改进（例如，对抗性损失、感知损失、循环一致性）对 GAN 变体进行分类。
• 通过定性与定量基准，评估每种 GAN 变体在三大核心挑战（图像质量、多样性、训练稳定性）上的表现。
• 对不同 GAN 设计之间的权衡进行批判性讨论，例如模式崩溃的缓解与保真度的提升。
• 评估架构创新（如跳跃连接、归一化层、渐进式增长）在改善训练动态与输出质量方面的影响。
• 整理并总结所研究 GAN 变体的代码仓库，以支持可复现性与进一步研究。

实验结果

研究问题

• RQ1 近期 GAN 架构在计算机视觉任务中，对提升生成图像质量的改善程度如何？
• RQ2 不同损失函数在增强生成输出的多样性并避免模式崩溃方面有何贡献？
• RQ3 哪些架构与训练策略实现了更稳定的 GAN 训练，且在不同基准上如何比较？
• RQ4 哪些 GAN 变体在图像到图像翻译与面部属性操控等应用中表现出最显著的成功？
• RQ5 在将 GAN 部署用于实际计算机视觉应用时，其关键局限与开放挑战是什么？

主要发现

• 架构创新（如渐进式增长与基于风格的归一化）显著提升了训练稳定性和图像质量，如 StyleGAN 与 BigGAN 所示。
• 损失函数改进——特别是感知损失与循环一致性损失的整合——有效增强了生成样本的多样性并减少了模式崩溃。
• 条件 GAN 及其变体在受控图像生成任务（如面部属性编辑与图像到图像翻译）中表现出色。
• 尽管已有进展，训练不稳定与模式崩溃仍是持续挑战，尤其在高分辨率图像生成中。
• 本文提出的分类体系通过架构与损失函数对 GAN 变体进行有效分类，有助于更清晰地比较与识别有前景的研究方向。
• 所综述的 GAN 变体的代码仓库已整理并公开可用，支持可复现性与未来基准测试。