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

Training method, system, device and storage medium for domain-adapted image classification network

A classification network and training method technology, applied in biological neural network models, instruments, calculations, etc., can solve the problems of large model parameters, long training time, training efficiency limitations, etc., achieve accurate image classification results, improve accuracy, The effect of improving overall performance

Active Publication Date: 2022-07-15
UNIV OF SCI & TECH OF CHINA
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the above methods often improve the accuracy of the classification model by adding additional modules, so the number of parameters of the model is large, the training time is long, and the training efficiency is also limited.

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
  • Training method, system, device and storage medium for domain-adapted image classification network
  • Training method, system, device and storage medium for domain-adapted image classification network
  • Training method, system, device and storage medium for domain-adapted image classification network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0036] The embodiment of the present invention provides a training method for a domain-adapted image classification network. Different from the existing scheme in which the feature before the classifier is directly used to calculate the contrast loss, the present invention uses the probability after the classifier to calculate the probability contrast loss, so as to facilitate the clustering process. It effectively constrains the class weights and the distance between features while keeping the features of the same class. Specifically, the present invention transfers the contrastive learning from the feature space to the probability space, and removes thel 2 Norm normalization to constrain the probability to take on a one-hot form. like figure 1 As shown, the method provided by the invention mainly comprises the following steps:

[0037] Step 1. Obtain the source domain image set, and obtain the target domain image set according to the training method; perform two different ...

Embodiment 2

[0102] The present invention also provides a training system for a domain-adapted image classification network, which is mainly implemented based on the method provided in the first embodiment, such as Figure 5 As shown, the system mainly includes:

[0103] The training data set construction unit is used to obtain the source domain image set, and obtain the target domain image set according to the training method; perform two different image transformations on each unlabeled target domain image in the target domain image set, and obtain the first The transformed image and the second transformed image form a target domain image pair, and a training data set is formed by the source domain image set, the target domain image set and all target domain image pairs;

[0104] a training data set input unit for inputting the training data set to the domain-adapted image classification network;

[0105] The baseline loss calculation unit is used to calculate one or more of the output ...

Embodiment 3

[0111] The present invention also provides a processing device, such as Image 6 As shown, it mainly includes: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the One or more processors implement the methods provided by the foregoing embodiments.

[0112] Further, the processing device further includes at least one input device and at least one output device; in the processing device, the processor, the memory, the input device, and the output device are connected through a bus.

[0113] In this embodiment of the present invention, the specific types of the memory, input device, and output device are not limited; for example:

[0114] The input device can be a touch screen, an image capture device, a physical button or a mouse, etc.;

[0115] The output device can be a display terminal;

[0116] The memory may be random access memory (Random Access Memory, RAM), or may be non...

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 training method, system, equipment and storage medium for a domain-adapted image classification network, which introduces comparative learning to cluster features of the same semantics, and solves the problem of insufficient labels in the target domain for the domain-adaptive image classification task; Feature comparison learning is improved to probability comparison learning. By performing comparison learning in probability space, the distance between the same semantic features and class weights after clustering is reduced, and the classification accuracy is improved; and only a loss of comparison learning is added. (i.e. the total probability versus loss), no complex additional modules are added, and the amount of parameters is not increased compared to previous methods. In general, the present invention improves the overall performance of the model without adding other additional modules, and can obtain more accurate image classification results.

Description

technical field [0001] The invention relates to the technical field of image classification, and in particular, to a training method, system, device and storage medium for a domain-adapted image classification network. Background technique [0002] In recent years, fully supervised learning strategies based on deep neural networks have achieved remarkable achievements in the field of image classification. Such fully supervised learning algorithms require the training data to be distributed in the same way as the test data. However, in practical applications, there are often differences between training (source domain) data and test (target domain) data. Domain adaptation methods aim to transfer the source domain knowledge to the target domain to address the above problems. [0003] In general, a classification model needs to cluster the same semantic features as much as possible while distributing them around the classification weights in the feature space. For unsupervis...

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): G06V10/774G06V10/764G06V10/82G06K9/62G06N3/04
Inventor 王子磊李俊杰
Owner UNIV OF SCI & TECH OF CHINA
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