论文标题

InfoMax-GAN:通过信息最大化和对比度学习改善了对抗图像的生成

InfoMax-GAN: Improved Adversarial Image Generation via Information Maximization and Contrastive Learning

论文作者

Lee, Kwot Sin, Tran, Ngoc-Trung, Cheung, Ngai-Man

论文摘要

尽管生成的对抗网络(GAN)是许多生成建模应用的基础,但它们却遭受了许多问题。在这项工作中,我们提出了一个原则性的框架,以同时减轻gan中的两个基本问题:灾难性忘记歧视者和发电机的模式崩溃。我们通过采用gans进行对比学习和共同信息最大化方法来实现这一目标,并进行广泛的分析以了解改进的来源。我们的方法显着稳定了GAN训练,并在相同的培训和评估条件下针对最先进的作品进行了五个数据集的图像合成的GAN性能。特别是,与最先进的SSGAN相比,我们的方法不会在图像域(例如面部)上的性能较差,而是显着提高了性能。我们的方法易于实施和实用:它仅涉及一个辅助目标,计算成本较低,并且在各种培训设置和数据集中都在没有任何超参数调整的情况下执行稳健性。为了获得可重复性,我们的代码可在模仿中提供:https://github.com/kwotsin/mimicry。

While Generative Adversarial Networks (GANs) are fundamental to many generative modelling applications, they suffer from numerous issues. In this work, we propose a principled framework to simultaneously mitigate two fundamental issues in GANs: catastrophic forgetting of the discriminator and mode collapse of the generator. We achieve this by employing for GANs a contrastive learning and mutual information maximization approach, and perform extensive analyses to understand sources of improvements. Our approach significantly stabilizes GAN training and improves GAN performance for image synthesis across five datasets under the same training and evaluation conditions against state-of-the-art works. In particular, compared to the state-of-the-art SSGAN, our approach does not suffer from poorer performance on image domains such as faces, and instead improves performance significantly. Our approach is simple to implement and practical: it involves only one auxiliary objective, has a low computational cost, and performs robustly across a wide range of training settings and datasets without any hyperparameter tuning. For reproducibility, our code is available in Mimicry: https://github.com/kwotsin/mimicry.

扫码加入交流群

加入微信交流群

微信交流群二维码

扫码加入学术交流群,获取更多资源