Processing

Please wait...

Settings

Settings

Goto Application

1. WO2022001489 - UNSUPERVISED DOMAIN ADAPTATION TARGET RE-IDENTIFICATION METHOD

Publication Number WO/2022/001489
Publication Date 06.01.2022
International Application No. PCT/CN2021/095647
International Filing Date 25.05.2021
IPC
G06K 9/62 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
G06N 3/08 2006.1
GPHYSICS
06COMPUTING; CALCULATING OR COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
CPC
G06K 9/6256
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
6256Obtaining sets of training patterns; Bootstrap methods, e.g. bagging, boosting
G06K 9/6262
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
9Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
62Methods or arrangements for recognition using electronic means
6217Design or setup of recognition systems and techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
6262Validation, performance evaluation or active pattern learning techniques
G06N 3/0454
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
04Architectures, e.g. interconnection topology
0454using a combination of multiple neural nets
G06N 3/08
GPHYSICS
06COMPUTING; CALCULATING; COUNTING
NCOMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
3Computer systems based on biological models
02using neural network models
08Learning methods
Applicants
  • 北京交通大学 BEIJING JIAOTONG UNIVERSITY [CN]/[CN]
Inventors
  • 金一 JIN, Yi
  • 高雅君 GAO, Yajun
  • 李浥东 LI, Yidong
  • 王涛 WANG, Tao
  • 郎丛妍 LANG, Congyan
  • 冯松鹤 FENG, Songhe
Agents
  • 北京市商泰律师事务所 BEIJING SUN LIGHT LAW FIRM
Priority Data
202010597480.X28.06.2020CN
Publication Language Chinese (zh)
Filing Language Chinese (ZH)
Designated States
Title
(EN) UNSUPERVISED DOMAIN ADAPTATION TARGET RE-IDENTIFICATION METHOD
(FR) PROCÉDÉ DE RÉ-IDENTIFICATION DE CIBLE D’ADAPTATION DE DOMAINE NON SUPERVISÉE
(ZH) 一种无监督域适应的目标重识别方法
Abstract
(EN) An unsupervised domain adaptation target re-identification method. The method comprises: constructing a multiscale domain adaptation attention learning network utilizing a source domain dataset and a target domain dataset to train the multiscale domain adaptation attention learning network, calculating multitasking losses of the multiscale domain adaptation attention learning network, and when the values of the multitasking losses converge, producing a trained multiscale domain adaptation attention learning network; utilizing the trained multiscale domain adaptation attention learning network to construct an unsupervised domain adaptation target re-identification model, and utilizing the unsupervised domain adaptation target re-identification model for target re-identification processing of an inputted image. The method reduces domain differences by splitting a feature map into a target-related feature map and a domain-related feature map, maps the feature maps in different scales, performs splitting in multiple scales, and allows multiscale feature expression that is only domain-related to be learned, thus achieving optimal performance.
(FR) Le procédé comporte les étapes consistant à: construire un réseau d’apprentissage multi-échelles d’attention d’adaptation de domaine en utilisant un jeu de données de domaine d’origine et un jeu de données de domaine de destination pour entraîner le réseau d’apprentissage multi-échelles d’attention d’adaptation de domaine, calculer des pertes en mode multitâches du réseau d’apprentissage multi-échelles d’attention d’adaptation de domaine, et lorsque les valeurs des pertes en mode multitâches convergent, produire un réseau entraîné d’apprentissage multi-échelles d’attention d’adaptation de domaine; utiliser le réseau entraîné d’apprentissage multi-échelles d’attention d’adaptation de domaine pour construire un modèle de ré-identification de cible d’adaptation de domaine non supervisée, et utiliser le modèle de ré-identification de cible d’adaptation de domaine non supervisée pour un traitement de ré-identification de cible d’une image introduite. Le procédé réduit les différences de domaine en divisant une carte d’attributs en une carte d’attributs liée à la cible et une carte d’attributs liée au domaine, met en correspondance les cartes d’attributs à des échelles différentes, effectue une division à des échelles multiples, et permet d’apprendre une expression d’attributs multi-échelles qui est uniquement liée au domaine, atteignant ainsi des performances optimales.
(ZH) 一种无监督域适应的目标重识别方法,该方法包括:构建多尺度域适应注意力学习网络;利用源域数据集和目标域数据集对多尺度域适应注意力学习网络进行训练,计算多尺度域适应注意力学习网络的多任务损失,在多任务损失的值收敛后,得到训练好的多尺度域适应注意力学习网络;利用训练好的多尺度域适应注意力学习网络构建无监督域适应的目标重识别模型,利用无监督域适应的目标重识别模型对输入的图像进行目标重识别处理。上述方法通过将特征图分割成与目标相关的特征图和与域相关特征图来减少域差异,将特征图映射在不同尺度下,在多个尺度下进行分割,可以学习到仅仅与域相关的、多尺度的特征表示,从而达到了最优的性能。
Related patent documents
Latest bibliographic data on file with the International Bureau