Image generation method based on discrete Fourier transform attention mechanism

An image generation and attention technology, applied in the field of computer vision, can solve the problems of high computational complexity and low computational efficiency, and achieve the effect of high computational complexity

Pending Publication Date: 2022-02-11
YANGTZE DELTA REGION INST (QUZHOU) UNIV OF ELECTRONIC SCI & TECH OF CHINA +1
View PDF0 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The invention is an image generation method of discrete Fourier transform attention mechanism, which solves the problems of high computational complexity and low computational efficiency existing in the existing generative confrontation network method based on self-attention mechanism

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
  • Image generation method based on discrete Fourier transform attention mechanism
  • Image generation method based on discrete Fourier transform attention mechanism
  • Image generation method based on discrete Fourier transform attention mechanism

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0063] Step 1: Preprocess the dataset;

[0064] Get the cifar10 dataset. The cifar10 dataset is composed of 10 categories of 32×32 natural color images and their corresponding category labels. It contains a total of 60,000 images and their corresponding labels. First, images can be classified into 10 categories according to the category labels of this dataset. Then, class labels are encoded using one-hot vectors. Finally, the image pixel values ​​are normalized to the range [-1,1], and the data is saved as a tensor for use by the generated adversarial network.

[0065] Step 2: Build a convolutional neural network;

[0066] This step builds a convolutional neural network that includes two subnetworks, one for the generator and the other for the discriminator; the input of the generator is Gaussian noise and picture category, its output is an image, and the input of the discriminator is an image and a picture category , output as a scalar. The first layer of the generator ne...

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 an image generation method based on a Fourier transform attention mechanism, and belongs to the field of computer vision. The method comprises the following steps: firstly, selecting a generative adversarial network as a basic framework, normalizing training pictures, and sampling normal distribution to obtain noise; improving the original attention mechanism, using the characteristic that discrete Fourier transform can be combined with all position information to carry out calculation for replacing an original calculation pixel point correlation feature map part with high complexity; and obtaining a feature map combining all position information. During training, noise and picture categories are input into the network at the same time, and a generative adversarial network algorithm is utilized to train the model. After the network is trained, the image generation task can be completed by inputting noise and picture categories in the generative adversarial. According to the invention, the calculation complexity and the time complexity of a self-attention mechanism can be remarkably reduced, and the quality of images generated by an existing method and the diversity of the images are improved.

Description

technical field [0001] The invention belongs to the field of computer vision, and mainly relates to the problem of image generation; it is mainly applied to the film and television entertainment industry, graphic design, machine vision understanding and the like. Background technique [0002] Image synthesis refers to the technology of using computer vision technology to understand image content and generate specified images according to requirements. It can generally be divided into two types: unsupervised image generation and supervised image generation. Unsupervised image generation refers to learning the mapping function from noise distribution to image distribution, and generating images through the mapping function. Supervised image generation refers to learning the conditional distribution of image data, and then generating images under given conditions. Image generation is a hot issue in the field of computer vision. It can not only solve the problem of missing vis...

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
Patent Type & Authority Applications(China)
IPC IPC(8): G06T11/00G06T5/00G06T5/10G06N3/04G06N3/08
CPCG06T11/00G06T5/002G06T5/10G06N3/084G06T2207/20056G06T2207/20081G06T2207/20084G06N3/045
Inventor 赵江伟唐佩军
Owner YANGTZE DELTA REGION INST (QUZHOU) UNIV OF ELECTRONIC SCI & TECH OF CHINA
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