Supercharge Your Innovation With Domain-Expert AI Agents!

Semantic segmentation model training method, computer equipment and storage medium

A computer equipment and model training technology, applied in the field of image processing, can solve problems such as network training instability, and achieve the effect of avoiding gradient instability and preventing category imbalance

Active Publication Date: 2019-11-05
XIAMEN UNIV
View PDF9 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to provide a semantic segmentation model training method, computer equipment and storage media to solve the problem of network training instability caused by using Dice_loss in the prior art

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
  • Semantic segmentation model training method, computer equipment and storage medium
  • Semantic segmentation model training method, computer equipment and storage medium
  • Semantic segmentation model training method, computer equipment and storage medium

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0022] Such as figure 1 Shown is the model training flowchart for semantic segmentation. The parameters of the model can be used to calculate the degree of deviation between the predicted value and the target value, and a loss function can be formed based on these parameters. In the present invention, the loss function is defined and optimized, the model parameters are updated, and the model is finally converged in the training segmentation model (ie, S2).

[0023] Define R b × R n → Functions on R Among them, α is called the adjustment factor, and its value range is between It can be proved (R n ,D) form the distance space and are used in the backpropagation algorithm.

[0024] First, prove that the function D(P,T) is the distance on Rn:

[0025] obviously, yes There exists D(P,T)=D(T,P), D(P,T)≧0, and the equality sign is achieved only when P=T.

[0026] Next, prove that All exist D(P,T)≤D(P,Z)+D(Z,T).

[0027] Note that the latter part of D(P,T) conforms to t...

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 semantic segmentation model training method. A loss function of the method is as follows: in the formula, P is a prediction vector, and T is a label vector of one-hot coding;wherein a and b are weights, a is increased along with the increase of the training period, b is decreased along with the increase of the training period, and a + b = 1; alpha is a regulatory factor,wherein the value range is located in a function defined by the invention; combining with a Dice coefficient; a new loss function is obtained, the problem of gradient explosion or disappearance is not prone to occurring in the back propagation process, meanwhile, whether classification of a single pixel point is correct or not and whether overall classification is correct or not are judged, gradient instability occurring when a Dice coefficient is independently used as the loss function is avoided, and the problem of category imbalance can be prevented.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a semantic segmentation model training method, computer equipment and storage media. Background technique [0002] In deep learning semantic segmentation tasks, the Dice coefficient is a commonly used measurement method for evaluating segmentation results. Its derived loss function Dice_loss directly aims to optimize the intersection ratio, which is not easily affected by analog imbalance, and is widely used in network training. However, Dice_loss sometimes has a sharp gradient change during network training, which leads to unstable network training. Therefore, this application analyzes the reasons for its occurrence and proposes an improvement plan. [0003] For the binary classification task, record the prediction vector as P, and the one-hot encoded label vector as T. Dice_loss can be expressed as follows: [0004] [0005] In the process of backpropagation, the loss funct...

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): G06K9/62G06T7/11
CPCG06T7/11G06F18/214
Inventor 周联昱
Owner XIAMEN UNIV
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