Unsupervised image style migration method based on dual learning

An unsupervised, style technology, applied in the field of computer vision, can solve the problems of local distortion, style transfer image noise, etc., to achieve the effect of eliminating image noise

Active Publication Date: 2019-11-15
BEIJING INSTITUTE OF TECHNOLOGYGY
View PDF3 Cites 20 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the method based solely on the confrontational network to obtain the style transfer image often has shortcomings such as more noise and local distortion.

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
  • Unsupervised image style migration method based on dual learning
  • Unsupervised image style migration method based on dual learning
  • Unsupervised image style migration method based on dual learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0067] This embodiment is the overall process and network structure of the unsupervised image style transfer model.

[0068] An unsupervised image style transfer method based on dual learning, such as figure 1 shown, including the following steps:

[0069] Step 1: Preprocess the training data. Obtain high-resolution images on the public data set as training data; the training data set contains multiple pictures of different sizes. In order to facilitate the design of the network structure and reduce the amount of calculation, first ignore the aspect ratio of the original image and uniformly scale it to 284× 284 size; in order to make up for the lack of training data, a 256×256 area is randomly cropped on the zoomed image to achieve data enhancement; this size is used to facilitate multiple internal downsampling operations during model operations (per Once downsampling, the image size will be halved, so only odd-sized images can be down-sampled; and 256x256 can ensure that th...

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 relates to an unsupervised image style migration method based on dual learning, and belongs to the field of computer vision. The method comprises the following steps: firstly, preprocessing training data, and then designing network structures of a generator and a discriminator; designing a loss function and training the generator and the discriminator by using the training data and the loss function to obtain an unsupervised image style migration network ST; introducing an aesthetic scoring model to maximize the aesthetic quality score of the generated image; taking basic pixel features and advanced semantic features of the images as dual consistency constraints of unsupervised training, and dynamically adjusting the weights of the two features; adaptively adjusting convergence rates of the model in different style migration directions by using a style balance technology; and finally, performing style migration on the input image by applying ST. Compared with an existingmethod, the method has the advantages that the target image with higher quality can be generated, good universality is achieved, meanwhile, the training process of the model is more stable, and selection and design of the network structure are more flexible.

Description

technical field [0001] The invention designs an unsupervised image style transfer method based on dual learning, and in particular relates to a method based on a so-called generative confrontation network and using multiple loss function training for unsupervised image style transfer, which belongs to the field of computer vision technology. Background technique [0002] With the deepening of the era of artificial intelligence, a large number of image applications have sprung up like mushrooms, one of which is a variety of image beautification apps that focus on filter functions, and the key technology of filter functions is image style transfer. [0003] Image style transfer refers to converting the original image into another style of image while keeping the main content of the image unchanged, such as the conversion of landscapes in different seasons, the conversion of different painting styles, etc. Unsupervised image style transfer based on neural networks means that th...

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): G06T3/00G06N3/04G06N3/08
CPCG06T3/0012G06N3/08G06N3/045
Inventor 宋丹丹李志凡
Owner BEIJING INSTITUTE OF TECHNOLOGYGY
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