Unlock instant, AI-driven research and patent intelligence for your innovation.

An unsupervised image translation method and system

An unsupervised, image technology, applied in the field of image translation, can solve the problem of insufficient authenticity of translated images, and achieve the effect of improving model discrimination and generation ability and accurate capture.

Active Publication Date: 2021-02-09
XIAMEN UNIV
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

For example, even the relatively powerful CycleGAN model is still insufficient in terms of mapping accuracy between different ensemble domains, and the realism of translating images while capturing geometric structural features and global features.

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
  • An unsupervised image translation method and system
  • An unsupervised image translation method and system
  • An unsupervised image translation method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0078] figure 1 It is a flow chart of the unsupervised image translation method in Embodiment 1 of the present invention. Such as figure 1 As shown, an unsupervised image translation method, including:

[0079] Step 101: Divide the original image data into source domain data and target domain data.

[0080] Step 102: Design the generation confrontation network, initialize the weights and hyperparameters of the generation confrontation network; the generation confrontation network includes: generator G t2s , discriminator D s2t_1 , discriminator D s2t_2 , generator G s2t , discriminator D t2s_1 and the discriminator D t2s_2 .

[0081] Step 103: Perform the first conversion task from source domain data to target domain data, specifically including:

[0082] Carry out the task of converting source domain data to target domain data, and extract image data X from the source domain data according to the batch size s respectively input to the discriminator D s2t_1 and the ...

Embodiment 2

[0106] figure 2 It is a structural diagram of an unsupervised image translation system according to Embodiment 2 of the present invention. Such as figure 2 As shown, an unsupervised image translation system includes:

[0107] The original image division module 201 is configured to divide the original image data into source domain data and target domain data.

[0108] The generation confrontation network initialization module 202 is used to design the generation confrontation network, and initialize the weights and hyperparameters of the generation confrontation network; the generation confrontation network includes: generator G t2s , discriminator D s2t_1 , discriminator D s2t_2 , generator G s2t , discriminator D t2s_1 and the discriminator D t2s_2 .

[0109] The first conversion module 203 is configured to perform a first conversion task from source domain data to target domain data.

[0110] The first discrimination loss calculation module 204 is configured to ca...

Embodiment 3

[0124] image 3 It is a network frame diagram of the unsupervised image translation method in Embodiment 3 of the present invention. Embodiment 3 of the present invention The unsupervised image translation method includes:

[0125] S1: Divide the original image data into source domain data and target domain data to form training data set and test data set.

[0126] S2: All sub-network weights and hyperparameters in this framework are initialized, and the model is established.

[0127] S3: For the conversion task from the source domain (Source Domain) to the target domain (Target Domain), fetch the image data X from the source domain according to the batch size s respectively input to the discriminator D s2t_1 (convolutional network) and discriminator D s2t_2 (Improved Capsule Network) to distinguish between real and fake (Real or Fake).

[0128] S4: For the above discriminator D s2t_1 and D s2t_2 Make the true and false discrimination of the source domain image for disc...

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 unsupervised image translation method and system. Taking two different image collection domains of the same object as the research object, based on the dual-capsule competition network and multi-agent generation confrontation, an unsupervised image translation method and system are proposed, which improves the model's discrimination and generation capabilities, and is used to generate more Rich global and local feature images, and can more accurately capture the distribution of image domains and learn the mapping relationship between different domains.

Description

technical field [0001] The invention relates to the field of image translation, in particular to an unsupervised image translation method and system. Background technique [0002] With the emergence of information multimedia technology, the technology of image as the main communication medium has been developed rapidly, so the technology related to image processing is becoming more and more important. Thanks to breakthroughs in artificial intelligence technology, especially deep learning technology, computer vision technology has been widely used. Among many computer vision tasks, many problems require synthesizing images, such as texture synthesis, image analogy, image super-resolution, image segmentation, style transfer, season transfer, and image understanding. Image translation technology that fuses features from different domains is expected to solve the above problems as a unified framework. For example, this technology can be used to synthesize images of different s...

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 Patents(China)
IPC IPC(8): G06K9/62G06N3/04G06N3/08
CPCG06N3/088G06N3/045G06F18/241G06F18/253
Inventor 邵桂芳刘暾东李铁军黄梦高凤强
Owner XIAMEN UNIV