Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Face image generation method based on GANs

A face image and image technology, applied in the computer field, can solve the problems of many parameter update iterations, complex optimization targets, easy loss of small patch areas or detailed textures, etc., to achieve reasonable clarity, improve quality, and avoid gradient disappearance. effect of the phenomenon

Active Publication Date: 2020-01-17
SOUTHWEST JIAOTONG UNIV
View PDF5 Cites 9 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

BEGANs improves the discriminator with the help of encoding and decoding ideas, and then optimizes the W distance, but the optimization goal is more complicated, and small patch areas or detailed textures are prone to be lost in the generated images.
In addition, in order to achieve a better training effect, they need more parameter update iterations.

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
  • Face image generation method based on GANs
  • Face image generation method based on GANs
  • Face image generation method based on GANs

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0074] Please refer to figure 1 , figure 2 and Figure 4 , the embodiment of the present invention provides a kind of face image generation method based on GANs, comprises the following steps:

[0075] S1. Obtain a training set X, which consists of several face images.

[0076] In this embodiment, two methods of obtaining the training set X are given: the first is to cut the center of the CELEBA face data set into fixed-size face images such as: 64×64 96×64, etc.; the second One is to use crawler technology to crawl pictures of people on the public network, and then use face recognition technology to crop out face images, and finally scale the image size to a fixed size such as: 64×64 96×64, etc.

[0077] S2. Extract hidden features of all face images in the training set X to obtain a hidden feature set C of human face images.

[0078] S3, face image decoding training

[0079] S31. Sampling batchsize face images x sequentially without repeated sampling from the training ...

Embodiment 2

[0109] Please refer to image 3 , for step S2 in Embodiment 1, it specifically includes:

[0110] S21. Sampling is not repeated sequentially from the training set X (in the same epoch cycle, when the sequence of training set samples is determined, sequential sampling is non-repetitive sampling) batchsize face images x 1 ,x 2 ,...,x k (k=batchsize), and transform the pixel value scale to [-1,1] according to formula (5), the transformed image is still recorded as x 1 ,x 2 ,...,x k (k=batchsize).

[0111]

[0112] Among them, i is the label of an image in batchsize images, i∈[1, batchsize].

[0113] S22, using the batchsize face images after the pixel value scale transformation in step S21 to train the feature learning network, as follows:

[0114] Construct an initial feature learning network, and convert the batchsize pixel value scale-transformed image x in step S21 1 ,x 2 ,...,x k (k=batchsize) is fed into the feature learning network, and the mean square error l...

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 face image generation method based on GANs, and relates to the technical field of computers, and the face image generated by a generator not only can be associated with a random vector, but also can be associated with a feature vector, thereby indicating that the generated image is directly affected by the features of a training image, and improving the interpretability.Gradient disappearance can be effectively avoided, decoding training can be possibly carried out before binary adversarial training is carried out, the gradient disappearance phenomenon caused by JS divergence optimization can be avoided, and then the quality of a generated image is improved. A decoder can learn good image structure features, so that a generator learns better structure features, face distorted images are reduced, and meanwhile, the definition of the images can be learned more reasonably. Due to the fact that feature decoding constraint is carried out, the gradient descent direction is also constrained to a certain extent when the target function is optimized, and fewer epoch numbers can be used in the training process.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a method for generating face images based on GANs. Background technique [0002] The image generation method based on GANs is one of the hotspots in current artificial intelligence research. In theory, the GANs image generation method can effectively simulate many image contents, such as: human faces, buildings, indoor scenes, flowers, animal images, etc. The generation of these images also has practical significance. For example, the effective generation of real faces or cartoon faces can save the virtual generation of some general characters in film and television or animation works, thereby saving costs; the generation of indoor scenes can effectively protect certain The indoor background information that some photographers want to protect; the number of images of a certain category is small, and more images of this category can be obtained to achieve the purpose of ...

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): G06T11/00G06K9/00G06K9/62G06N3/08G06N3/04
CPCG06T11/00G06N3/08G06V40/168G06N3/045G06F18/214
Inventor 和红杰陈泓佑陈帆
Owner SOUTHWEST JIAOTONG UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com