Method for migrating generative adversarial network with adversarial learning and discriminative learning

A discriminative and adversarial technology, applied in the field of deep learning neural networks, which can solve problems such as difficult image datasets, training, etc.

Active Publication Date: 2021-09-07
CHANGCHUN UNIV OF TECH
View PDF6 Cites 6 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to solve the problem that the generative confrontation network is difficult to train effectively and quickly in the target doma...

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method for migrating generative adversarial network with adversarial learning and discriminative learning
  • Method for migrating generative adversarial network with adversarial learning and discriminative learning
  • Method for migrating generative adversarial network with adversarial learning and discriminative learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0064] An adversarial neural network model ADT-GAN and a training method based on the combination of adversarial learning and discriminative row learning, which includes:

[0065] S1. Prepare image dataset

[0066] Prepare a source domain image dataset with a larger amount of data and a target domain image dataset with a smaller amount of data, and perform the following processing on the source domain dataset and the target domain dataset respectively:

[0067] 1) Split the image data set into source domain data set and target domain data set;

[0068] 2) Standardize the pictures in the dataset to the same resolution;

[0069] MNIST is a dataset of handwritten digits, consisting of 60,000 training data and 10,000 test data of images, and the present invention only uses training data. For the MNIST handwritten data set, each handwritten digit image is normalized to a grayscale image of 28×28 pixels and placed in the center of the image. In order to test the effect of ADT-GAN...

Embodiment 2

[0118] Example 2 Evaluation of the model after training

[0119] The evaluation of GAN (including ADT-GAN) can measure the similarity between the generated image and the real image by the Fréchet initial distance (FID) of the intermediate features of the network obtained from the generated image and the real image in the Inception v3 image classification model. The generated image and the real image are obtained in the Inception v3 image classification model. The intermediate features of the network can be modeled as a Gaussian distribution, and the mean values ​​are m r and m g , the covariance matrix is ​​Σ r and Σ g . The FID describing the statistical similarity of two intermediate features is defined as

[0120]

[0121] The smaller the FID, the more similar the two groups of images are; the larger the FID, the greater the difference between the two groups of images.

[0122] The FID values ​​of DCGAN, initialized DCGAN and ADTGAN in the same iteration are used ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a method for migrating a generative adversarial network by using adversarial learning and discriminative learning. The method comprises the following steps: S1, preparing a picture data set; s2, constructing a pre-trained GAN model; s3, constructing an ADT-GAN model through parameter migration; and S4, training the ADT-GANc. The ADT-GAN model initializes a generator and a discriminator through parameter transfer on the basis of a pre-trained GAN model trained by a source domain image data set by using transfer learning. A domain discriminator is added, a total objective function composed of an adversarial objective function and a domain discrimination objective function is optimized to drive a generator to generate image data of a target domain, and negative migration is avoided. Therefore, the training performance on a small target domain data set is improved, the number of iterations is reduced, and the image generation quality is improved.

Description

technical field [0001] The invention belongs to a deep learning neural network, in particular to a method for migrating a generative confrontation network with confrontational learning and discriminative learning. Background technique [0002] Generative Adversarial Networks (GANs, a type of deep model, have attracted a lot of attention, and the need to use GANs in many fields is also growing. Like other deep neural networks, GANs have high computational demands and need to be used in Training on large datasets, while quickly and efficiently training GANs on small training datasets and generating valid samples, becomes a particularly important and challenging research problem. [0003] Transfer learning which aims to improve the performance of a target learner in a target domain by transferring knowledge contained in a different but related source domain. Transfer learning has been used with GANs, mainly focusing on image-to-image translation and domain adaptation. Image-t...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/241G06F18/214
Inventor 李阳王宇阳文敦伟常佳乐
Owner CHANGCHUN UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap