Abstract
We introduce a distance metric between two distributions and propose a Generative Adversarial Network (GAN) model: the Simplified Fréchet distance (SFD) and the Simplified Fréchet GAN (SFGAN). Although the data generated through GANs are similar to real data, GAN often undergoes unstable training due to its adversarial structure. A possible solution to this problem is considering Fréchet distance (FD). However, FD is unfeasible to realize due to its covariance term. SFD overcomes the complexity so that it enables us to realize in networks. The structure of SFGAN is based on the Boundary Equilibrium GAN (BEGAN) while using SFD in loss functions. Experiments are conducted with several datasets, including CelebA and CIFAR-10. The losses and generated samples of SFGAN and BEGAN are compared with several distance metrics. The evidence of mode collapse and/or mode drop does not occur until 3000k steps for SFGAN, while it occurs between 457k and 968k steps for BEGAN. Experimental results show that SFD makes GANs more stable than other distance metrics used in GANs, and SFD compensates for the weakness of models based on BEGAN-based network structure. Based on the experimental results, we can conclude that SFD is more suitable for GAN than other metrics.
Highlights
Generative Adversarial Net (GAN) is one of the models drawing attention in the field of machine learning (ML) and computer vision [1]
Domain Adversarial Neural Network (DANN) and Adversarial Discriminative Domain Adaption (ADDA) considered image-to-image translation [2,3] and GAN with text manifold interpolation and image-text matching discriminator considered text-to-image synthesis [4]
We introduce a Simplified Fréchet distance (SFD) and propose a GAN model in which
Summary
Generative Adversarial Net (GAN) is one of the models drawing attention in the field of machine learning (ML) and computer vision [1]. Mode collapse is generating the similar or even the same outputs for different random input vectors, while mode drop is concerned with modes being dropped from the output distribution [1] These phenomena may occur when the distribution of real data cannot be represented correctly because of using inadequate optimizations or insufficient network resources that cause an inability of node counting [1,13,14]. A portion of the characteristics of FD could be explored when SFD is used in the training process of GANs. The structure of the proposed GAN model, the Simplified Fréchet Generative Adversarial. SFD reduces the complexity of FD, representing some characteristics of FD and enables stable training by compensating the weakness that the models belonging to BEGAN-based network structure have.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
