Deformable convolution kernel method based on WGAN (Wasserstein-Generative Adversarial Network) model

Abstract

The invention discloses a deformable convolution kernel method based on a WGAN (Wasserstein-Generative Adversarial Network) model, and belongs to the field of the deep learning neural network. The method comprises the following steps that: S1: constructing an original generative adversarial network model; S2: constructing a Wasserstein distance as the judgment index of the adversarial network model; S3: initializing random noise, and inputting the random noise into a generator; S4: in the WGAN model, utilizing a deformable convolution kernel to carry out convolution on an image; and S5: inputting a loss function obtained by a deformable convolution operation into the generator for subsequent training. By use of the deformable convolution kernel method, which is constructed by the invention, based on the WGAN model, a convolution way generated after a discriminator and the generator receive a picture is changed, the discriminator and the generator can automatically change the size of the convolution kernel according to a training situation, so that the characteristics of a dataset image can be adaptively learnt, and the robustness of whole network training is improved.

Description

Technical field [0001] The invention relates to the field of deep learning neural networks, in particular to a deformable convolution kernel method based on a WGAN model. Background technique [0002] The Generative Adversarial Network (GAN) is a deep learning framework proposed by Goodfellow in 2014. It is based on the idea of ​​"game theory" and constructs two models: generator and discriminator, the former The image is generated by inputting uniform noise of (0,1) or Gaussian random noise, and the latter discriminates the input image and determines whether it is an image from a data set or an image generated by a generator. [0003] In the traditional adversarial network model, there is no unified criterion for the quality of the image generated by the generator. Therefore, it is urgent to propose a method that uses the Wasserstein distance as the evaluation index to generate the adversarial network, so that the entire model can be trained correctly. In addition, the method of ...

AI Technical Summary

Problems solved by technology

[0003] In the traditional adversarial network model, there is no unified evaluation standard for the image quality generated by the generator. Therefore, it is urgent to propose a method that uses the Wasserstein distance as the evaluation index of the generative adversarial network, so that the training of the entire model can go to the correct direction. In addition, the method of using deformable convolution to learn image features improves the training efficiency of the entire network.

Images