BTGAN: Training GAN with Balanced Triplet Loss and Two-Branch Architecture
BTGAN: Training GAN with Balanced Triplet Loss and Two-Branch Architecture
复制标题
DOI：
10.1109/ijcnn52387.2021.9533969
复制
发表时间：
2021-07
期刊：
影响因子：
--
通讯作者：
Simin Yu;Kuntian Zhang;Chuan Xiao;Xianyu Bao;J. Huang;Mark Junjie Li
中科院分区：
文献类型：
--
作者：
Simin Yu;Kuntian Zhang;Chuan Xiao;Xianyu Bao;J. Huang;Mark Junjie Li
TripletGAN是生成对抗网络（GAN）的一个变体，它用三重损失代替了GAN的分类损失。尽管由于对抗性三重丢失的特性可以最大化生成样本之间的嵌入距离，TripletGAN比vanilla GAN提供了更好的模式覆盖，但其对抗性训练方法存在一些缺陷，即一些生成的图像往往会偏离真实的样本分布，并且随着我们增加训练迭代次数，会产生带噪图像。在本文中，我们提出了一个对抗平衡的三重损失与四个动态系数，以实现质量和生成的样本的多样性之间的权衡。我们还设计了一种新的网络架构，为GAN提供自动编码能力。大量的实验证明了我们提出的方法在缓解TripletGAN中的问题方面的有效性，以及在重建方面优于直接训练生成器和编码器（如O-GAN）的一些方法。
TripletGAN is a variant of Generative Adversarial Network (GAN) by replacing the classification loss of discriminator with a triplet loss. Although TripletGAN delivers better mode coverage than vanilla GAN thanks to the characteristics of adversarial triplet loss that maximizes the embedding distance between generated samples, its adversarial training method suffers from the drawback that some generated images tend to deviate from the real sample distribution and noisy images are produced as we increase the number of iterations of training. In this paper, we propose an adversarially balanced triplet loss with four dynamic coefficients to achieve a trade-off between the quality and the diversity of generated samples. We also design a novel network architecture to provide GANs with an auto-encoding ability. Extensive experiments demonstrate the effectiveness of our proposed methods in terms of alleviating the problem in TripletGAN and the superiority in terms of reconstruction over some methods that directly train generator and encoder such as O-GAN.