Image classification automatic annotation method based on unknown pre-training annotation data

A technology for automatically labeling and labeling data, applied in the field of deep learning and computer vision, can solve the problems of long training time, inconvenient storage and transmission, and large space occupied by pre-training data sets, so as to reduce storage and transmission costs and automatically label The effect of improving accuracy and saving labeling costs

Active Publication Date: 2021-07-30
ZHEJIANG LAB
View PDF3 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, there are two problems with this method: 1. There may be a large distribution difference between the pre-training data and the data to be labeled, and the generalization performance of the model is difficult to guarantee; 2. Th

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
  • Image classification automatic annotation method based on unknown pre-training annotation data
  • Image classification automatic annotation method based on unknown pre-training annotation data
  • Image classification automatic annotation method based on unknown pre-training annotation data

Examples

Experimental program
Comparison scheme
Effect test

Example Embodiment

[0054]The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, but not to limit the present invention.

[0055] like figure 1 , 2 As shown, the present invention discloses an automatic labeling method for image classification based on pre-training labeling data agnostic, comprising the following steps:

[0056] Step 1: Obtain the image to be labeled X i (i=1, 2...N), the number is N. Offline collection of pre-trained image classification tasks corresponding to performance SOTA image classification models. Specifically, refer to the following but not limited to the following model selections: VGG, ResNet, DenseNet, Inception.

[0057] As an optional implementation manner, a pre-trained image classification model M is obtained, and the label space C corresponding to M inc...

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 an image classification automatic annotation method based on unknown pre-training annotation data, and the method comprises the following steps: S1, obtaining a to-be-annotated image, and collecting a pre-training image classification model; s2, splitting the pre-trained image classification model into a feature extraction model and a label prediction model, initializing and fixing parameters of the label prediction model, and not participating in subsequent migration training; s3, constraining the feature extraction model, so that the output specific category of the automatic labeling model is determined, and the overall distribution is discrete; s4, clustering output features of the feature extraction model; s5, screening out clusters with the sizes exceeding a threshold value, and forming a to-be-labeled image label space by corresponding categories; s6, labeling all the to-be-labeled images with pseudo labels; s7, re-clustering and allocating pseudo labels, and performing supervised training on the feature extraction model; s8, iterating S3 to S7; and S9, reasoning the to-be-labeled image by using the migrated automatic labeling model to obtain a labeling result.

Description

technical field [0001] The invention relates to the fields of deep learning and computer vision, in particular to an automatic labeling method for image classification based on agnostic pre-training labeling data. Background technique [0002] With the rapid development of deep learning and computer vision research, its related applications have affected every aspect of our lives. As a typical task in the field of computer vision, image classification, its methods have been widely used in tasks such as face recognition, automatic driving and scene recognition. However, training a good image classification model relies on a large amount of labeling data, and using manual labeling of images often consumes a lot of manpower and time. Therefore, how to perform efficient image annotation has received more and more attention, and automatic annotation can effectively alleviate the above problems. [0003] Existing automatic labeling methods often require semi-supervised training ...

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): G06K9/62G06N3/04G06N3/08
CPCG06N3/088G06N3/045G06F18/2155G06F18/23G06F18/24
Inventor 钟昊文陈岱渊单海军杨非傅家庆俞再亮
Owner ZHEJIANG LAB
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
Try Eureka
PatSnap group products