赞
踩
什么是生成对抗网络(GANs)?简而言之,它是由两部分组成,一个是生成器(Generator),另一个是判别器(Discriminator)。生成器负责生成新的样本,而判别器则负责判断输入是否为真实数据。两个模型之间通过对抗的方式训练,希望使得生成的数据看起来像真实数据。训练过程可以被认为是一种博弈游戏,生成器想要欺骗判别器,而判别器则想通过自身学习去区分真假样本。
今天要分享的是基于PyTorch实现的一个具体的GAN模型——DCGAN,并结合MNIST手写数字数据库来进行实例化。主要面向零基础读者,以最简单易懂的文字和代码为主。文章重点将会在于介绍GAN的基本原理、核心算法和PyTorch库中相关模块的应用。
Generative Adversarial Networks (GANs) are one of the most interesting ideas in deep learning today. They allow us to automatically learn to generate new data with high fidelity from a given set of training data without being explicitly programmed for it. GANs have many applications in creativity and media generation, image-to-image translation, and reinforcement learning. The basic idea behind GANs is that we train two models simultaneously:
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。