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

Deformable convolution method and device and storable medium

A convolution method and convolution technology, applied in the field of computer vision, can solve problems such as limiting the expressive ability of deformable convolution and destroying the consistency of different channel positions of convolution input features, so as to improve the accuracy of target detection and ensure consistent positions Sexuality, the effect of enhancing expressive ability

Pending Publication Date: 2022-07-05
NANJING UNIV +1
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the number of sampling points in most existing deformable convolution methods is fixed, which limits the expressive ability of deformable convolution; and the existing deformable convolution methods that use multiple sets of sampling points will convolve the input feature Each channel is divided into multiple groups, and different sampling points are used in each group, which destroys the position consistency of different channels of convolution input 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
  • Deformable convolution method and device and storable medium
  • Deformable convolution method and device and storable medium
  • Deformable convolution method and device and storable medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0035] like figure 1 As shown, the deformable convolution method of the present invention includes the following steps:

[0036] Step 1, perform conventional convolution on the input feature map to obtain offset vectors of multiple sets of sampling point coordinates;

[0037] Step 1: Sampling the features on the input feature map, and use the convolution kernel weight to weight and sum the sampled features, including:

[0038] Step 1-1, for the position p on the output feature map y 0 , sample features using a 3x3 grid R on the input feature map x, where R={(-1,-1),(-1,0),...,(0,1),(1,1) }.

[0039] Step 1-2, use the convolution kernel weight w for the sampled features o The weighted summation is used to obtain the offset value vector o of the coordinates of each group of sampling points, using the following formula:

[0040]

[0041] Steps 1-3, divide the offset vector o into g groups, and obtain the offset vector o of each group 1 ,o 2 ,...,o g . where each set o...

Embodiment 2

[0061] An embodiment of the present invention further provides an apparatus including a processor and a memory; wherein, the memory stores programs or instructions, and the programs or instructions are loaded and executed by the processor to implement the variable convolution method of Embodiment 1.

Embodiment 3

[0063] The present invention also provides a computer-readable storage medium. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium may also be a volatile computer-readable storage medium. The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to execute the variable convolution method of Embodiment 1.

[0064] Those skilled in the art can clearly understand that the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, the computer software product. Stored in a storage medium, it includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the pr...

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 deformable convolution method and device and a storage medium. The deformable convolution method comprises the following steps of 1, performing conventional convolution on an input feature map to obtain offset vectors of multiple groups of sampling point coordinates; 2, extracting the features of the shifted sampling points from the input feature map for the coordinates of each group of sampling points, and performing weighted summation by using a convolution kernel to obtain the output features of each group; and step 3, averaging the convolution output features of each group to obtain a final output feature. According to the method, through multiple groups of sampling points, the expression capability of deformable convolution is enhanced, the position consistency of channels of convolution input features is ensured, and experiments on a target detection data set COCO prove that the target detection performance is effectively improved.

Description

technical field [0001] The invention relates to a deformable convolution method for multiple groups of sampling points, belonging to the technical field of computer vision. Background technique [0002] With the development of deep learning technology, object detection is widely used in many fields. For example, in the industrial quality inspection, the defects of industrial products are detected; in the traffic field, pedestrians and vehicles in front of the vehicle are detected; in the security field, the abnormal events in public places such as elevators are detected, and so on. [0003] Deformable convolution is a commonly used technology in target detection. Compared with ordinary convolution, deformable convolution can adaptively adjust the sampling points of convolution according to the input features, so as to adaptively realize the change of shape and scale. . However, most of the existing deformable convolution methods have a fixed number of sampling points, whic...

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): G06V10/774G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/214
Inventor 路通成晓龙黄建武曹阳
Owner NANJING 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