WGAN model method based on depth convolution nerve network

Abstract

The invention discloses a WGAN model method based on a depth convolution nerve network, and belongs to the deep learning nerve network field; the method comprises the following steps: S1, forming a Wasserstein generative confrontation network WGAN model; S2, forming a discriminator with the depth convolution nerve network structure; S3, forming a generator with a transposition convolution nerve network structure; S4, employing a Wasserstein distance loss function as a discriminator loss function; S5, preparing a data set so as to train the formed network. The method creatively proposes the depth convolution nerve network and WGAN combined forming mode according to the generative confrontation network model features, forms the generator and the discriminator as the depth convolution nerve network mode, and uses the WGAN loss function mode; the method can learn image features in the training process, and can use the loss functions to reflect the generative image quality.

Description

technical field [0001] The invention relates to the technical field of deep learning neural networks, in particular to a WGAN model method based on a deep convolutional neural network. Background technique [0002] Generative Adversarial Network (GAN for short) is a framework proposed by Goodfellow in 2014. It is based on the idea of ​​"game theory" and constructs two models of generator (generator) and discriminator (discriminator). Uniform noise of (0, 1) or Gaussian random noise generates images, and the latter discriminates the input image to determine whether it is an image from the dataset or an image produced by the generator. Every time the discriminator completes a judgment, it returns the result error to the generator. [0003] However, in the traditional GAN ​​model, the loss function of the discriminator cannot indicate the direction of network training, that is, none of the indicators can reflect the quality of the generated image. The result of this situation...

AI Technical Summary

Problems solved by technology

[0003] However, in the traditional GAN ​​model, the loss function of the discriminator cannot indicate the direction of network training, that is, none of the indicators can reflect the quality of the generated image

The result of this situation is that the network will be trained endlessly, that is to say, there is no clear termination condition for the training of the generative confrontation network

[0004] In the standard WGAN model, the generator and the discriminator are traditional fully-connected layer structures. During the training process, the generator can only correct the image generated by itself through the back propagation of the loss function error of the discriminator.

In this case, the speed of network training is relatively slow, and the generator has no way to learn the characteristics of the data set.

Images