多判别器协同框架:高品质图像的谱归一生成对抗网络

通过生成对抗网络的对抗学习生成仿真图像,已成为人工智能领域的一个研究热点.为了进一步提高生成图像的质量,本文提出了多判别器协同合作的网络框架——采用多个判别器为唯一生成器提供联合损失量,并通过不同的学习率保持各个判别器的差异性.同时,为了满足判别器的Lipschitz连续条件,本文所有的判别器网络一律进行谱归一化操作.实验表明,本文提出的基于多判别器合作框架的生成对抗网络表现较优.

Multi-discriminator Co-operation Framework:Spectral Normalized Generative Adversarial Networks for High Quality Generated Images

Generative Adversarial Networks(GANs)can generate lots of realistic images.It has been a hot issue in recent years in the research area of deep learning.In order to further improve the final quality of generated images,we propose a new framework of GANs based on co-operation of several discriminators.In this paper three discriminators are used with different learning rate to keep them distinguishing.In the mean time,spectral normalization is used on each of the discriminators to ensure that Lipschitz continuity is satisfied.Experiments show that our multi-discriminators networks framework has better performance on image compositing.