Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Pixel-level label automatic generation model construction and automatic generation method and device

A technology for automatic generation and construction of methods, applied in biological neural network models, neural architectures, character and pattern recognition, etc., can solve the problems of lack of a large number of training labels, weakly supervised semantic segmentation, etc., to expand the scope of application, improve Semantic segmentation effect, the effect of improving segmentation accuracy

Active Publication Date: 2019-12-20
NORTHWEST UNIV(CN)
View PDF6 Cites 3 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a pixel-level label automatic generation model construction and automatic generation method to solve the existing strong supervised semantic segmentation lacks a large number of training labels and weakly supervised semantic segmentation.

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
  • Pixel-level label automatic generation model construction and automatic generation method and device
  • Pixel-level label automatic generation model construction and automatic generation method and device
  • Pixel-level label automatic generation model construction and automatic generation method and device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0075] In this embodiment, a method for constructing a pixel-level automatic label generation model is disclosed, which is used to obtain an automatic label generation model for an image to be labeled.

[0076] Such as figure 1 shown, follow the steps below:

[0077] Step 1. Obtain an existing image set that is semantically similar to the image to be labeled, and obtain a semantically similar image set;

[0078] or

[0079] Obtain multiple single background images containing the object to be labeled in the image to be labeled, and obtain a simple image set;

[0080] Using the semantically similar image set or simple image set to train the deep neural network to obtain a pre-labeled model;

[0081] In this embodiment, one of two methods is used to obtain the pre-marking model. When the existing image set has an image set similar to the object to be marked in the image to be marked, directly use the existing image set to train the deep neural network. , to obtain a pre-annot...

Embodiment 2

[0113] A method for automatically generating pixel-level labels, which is performed in the following steps:

[0114] Step A, obtaining the image to be labeled;

[0115] Step B, using the pixel-level label automatic generation model construction method in Embodiment 1 to construct an automatic label generation model for the image to be labeled;

[0116] Step C. Input the image to be labeled into the automatic label generation model obtained in step B, and output image pixel-level labels.

[0117] In this embodiment, images of different plant species in the "Orchid" family are collected as images to be labeled such as figure 2 shown.

[0118] When using the pixel-level label automatic generation model construction method in Embodiment 1 to construct the label automatic generation model of the image to be marked, the pre-labeled model is obtained using the existing image set PASCAL VOC 2012 data set as a semantically similar image set, PASCAL VOC 2012 The dataset contains 20 ...

Embodiment 3

[0124] In this embodiment, the branch image in the complex background image is obtained as the image to be labeled such as Figure 4 shown.

[0125] When using the pixel-level label automatic generation model construction method in Example 1 to construct the label automatic generation model of the image to be labeled, the pre-labeled model was obtained because the domain "tree branch" semantics was not found on the public dataset with pixel-level labels similar categories, so a simple image set is obtained first by acquiring a single background image, such as Figure 5 As shown, the ResNet101 in DeepLabV3+ is selected as the deep neural network to obtain the pre-labeled model, and then the automatic label generation model is obtained through pre-labeled model training. A batch size of 8 is used for training, and the initial learning rate is set to 0.007, divided by 10 every 5 epochs. Weight decay and momentum are set to 0.0002 and 0.9 respectively.

[0126] Will Figure 4 ...

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 pixel-level label automatic generation model construction and automatic generation method and device. The pixel-level label automatic generation model construction and automatic generation method comprises the steps: firstly learning segmentation knowledge from a source domain through a pixel-level label, then transferring the knowledge to a target domain to generate a coarse label of an image, carrying out the reasoning of the coarse label through a guide filter, and generating a fine label. On the basis of refining the label, the segmentation network is optimized, and the fine-grained target label with a detailed pixel-level structure / boundary is generated, and the semantic segmentation effect is improved.

Description

technical field [0001] The invention relates to an image label generation method, in particular to a pixel-level label automatic generation model construction, automatic generation method and device. Background technique [0002] Semantic image segmentation is an important task in computer vision. It assigns a specific semantic label to each pixel in the image, that is, each pixel in the image will have a label, for example, when segmenting the image , segment the foreground object and the background image into two categories, where the label of the foreground object is 1, the label of the background image is 0, and each pixel will have a pixel-level label of 0 or 1, then from the image point of view, an image is processed A binarized result is achieved, which realizes image segmentation. [0003] In recent years, deep convolutional neural networks have shown excellent performance in semantic image segmentation and achieved remarkable results, in which the success of deep 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 Applications(China)
IPC IPC(8): G06K9/62G06N3/04
CPCG06N3/045G06F18/22G06F18/29G06F18/214
Inventor 范建平张翔赵万青罗迒哉彭进业李展胡琦瑶艾娜樊萍王琳
Owner NORTHWEST UNIV(CN)
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products