Supercharge Your Innovation With Domain-Expert AI Agents!

Establishment method and application of representative graph structure model and visual understanding model

A technology of model building and visual understanding, applied in the field of visual understanding, can solve the problems of limiting long-distance dependent capture efficiency and effect, low accuracy of visual understanding tasks, and high computational complexity, so as to improve application prospects, reduce computational complexity, The effect of enhancing representational power

Active Publication Date: 2020-11-24
HUAZHONG UNIV OF SCI & TECH
View PDF8 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] Aiming at the defects and improvement needs of the prior art, the present invention provides a representative graph structure model, a method for establishing a visual understanding model and its application. , which limits the efficiency and effectiveness of long-distance dependent capture, leading to the technical problem of low accuracy in visual understanding tasks

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
  • Establishment method and application of representative graph structure model and visual understanding model
  • Establishment method and application of representative graph structure model and visual understanding model
  • Establishment method and application of representative graph structure model and visual understanding model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] A method for establishing a representative graph structure model, comprising:

[0053] Build a representative graph structure model for capturing long-distance dependency information of input feature images;

[0054] Such as figure 1 As shown, the representative graph structure model includes: feature mapping module, sampling module, long-distance dependency information capture module and feature demapping module;

[0055] A feature mapping module for extracting a value branch, a key-value branch and a sequence branch from the input feature image, and generating an offset matrix for indicating coordinates of sampling points;

[0056] The sampling module is used to sample the neighbor nodes of each node in the value branch and the key-value branch respectively according to the offset matrix, so as to obtain representative features of the value branch and representative features of the key-value branch;

[0057] The long-distance dependent information capture module is ...

Embodiment 2

[0072] A method for establishing a representative graph structure model. This embodiment is similar to the above-mentioned embodiment 1. The difference is that the representative graph structure model provided in this embodiment is a bottleneck-shaped representative graph structure model. Its structure is as follows Figure 4 shown;

[0073] Such as Figure 4 As shown, in this embodiment, the feature mapping module includes: the sixth convolutional layer, the first batch of normalization layers, the first activation layer and the seventh convolutional layer, the sixth convolutional layer and the seventh convolutional layer The convolution kernel size is 1×1;

[0074] The sixth convolution layer, the first batch normalization layer and the first activation layer are used to sequentially perform convolution operation, batch normalization operation and activation operation on the input feature image to obtain the value branch, key value branch and sequence branch ;

[0075] T...

Embodiment 3

[0081] A method for establishing a representative graph structure model, this embodiment is similar to the above-mentioned embodiment 1, the difference is that, as Figure 5 As shown, in this embodiment, a node represents an image grid;

[0082] Specifically, the feature mapping module rasterizes the input feature image according to space, divides the positions in the input feature into different groups, and the upper left position element in each group is the anchor position, and uses the average pooling to aggregate the information to regress the partial shift matrix; each grid acts as a node; the learned shift matrix is ​​applied to all anchor positions to sample its representative nodes for each grid;

[0083] Such as Figure 5 As shown, in this embodiment, the grid size is 3×3. Specifically, as shown in the center box in the 3×3 input and its corresponding rasterized feature p, the anchor coordinates of each group are grid The coordinates of the pixel position in the up...

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 representative graph structure model, and a visual understanding model establishment method and application, and belongs to the field of visual understanding, and the methodcomprises the steps: establishing a representative graph structure model, wherein the representative graph structure model comprises a feature mapping module used for extracting a value branch, a keyvalue branch and a sequence branch from an input feature image and generating an offset matrix, a sampling module is used for sampling nodes (pixels or image grids) in the value branches and the key value branches according to the offset matrix to obtain representative features, a long-distance dependence information capture module is used for performing matrix multiplication on the representativecharacteristics of the key value branches and the sequence branches and then performing Softmax operation to obtain a relation matrix, and performing matrix multiplication on the representative characteristics of the value branches and the relation matrix to obtain a long-distance dependence matrix; and a feature reflection module is used for encoding the long-distance dependent information intothe input feature image. According to the method, more refined long-distance dependence information can be learned, and the accuracy of a visual understanding task is improved.

Description

technical field [0001] The invention belongs to the field of visual understanding, and more specifically relates to a representative graph structure model, a method for establishing a visual understanding model, and an application thereof. Background technique [0002] Long-distance dependence is the semantic relationship between regions or pixels that are far apart in the image. The work of modeling long-distance dependencies is of great significance for visual understanding tasks such as semantic segmentation, target detection, and target segmentation. Factors that influence the judgment result are included. Previous mainstream methods rely on deep stacks of local operations, such as convolution operations. However, this method is computationally inefficient, difficult to optimize and has a small receptive field. [0003] To solve the above problems, non-local methods are proposed to capture long-distance dependencies. For each position, the non-local operation takes t...

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
IPC IPC(8): G06K9/62G06K9/46G06N3/04
CPCG06V10/40G06N3/045G06F18/214
Inventor 吴东岳余昌黔高常鑫桑农
Owner HUAZHONG UNIV OF SCI & TECH
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More

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

© 2025 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com