CN111738172A - Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering - Google Patents

Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering Download PDF

Info

Publication number
CN111738172A
CN111738172A CN202010592282.4A CN202010592282A CN111738172A CN 111738172 A CN111738172 A CN 111738172A CN 202010592282 A CN202010592282 A CN 202010592282A CN 111738172 A CN111738172 A CN 111738172A
Authority
CN
China
Prior art keywords
domain
loss
data set
training
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010592282.4A
Other languages
Chinese (zh)
Other versions
CN111738172B (en
Inventor
郭海云
王金桥
唐明
刘松岩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhongke Zidong Taichu Beijing Technology Co ltd
Original Assignee
Institute of Automation of Chinese Academy of Science
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Automation of Chinese Academy of Science filed Critical Institute of Automation of Chinese Academy of Science
Priority to CN202010592282.4A priority Critical patent/CN111738172B/en
Publication of CN111738172A publication Critical patent/CN111738172A/en
Application granted granted Critical
Publication of CN111738172B publication Critical patent/CN111738172B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/20—Scenes; Scene-specific elements in augmented reality scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention belongs to the field of computer vision and pattern recognition, and particularly relates to a cross-domain target re-recognition method, a system and a device based on feature countermeasure learning and self-similarity clustering, aiming at solving the problem that the existing target re-recognition method limits the discrimination of feature expression due to unfixed number of clustering centers and causes poor robustness of recognition results. The system method comprises the following steps: acquiring an image to be identified as an input image; extracting the characteristics of the input image through a pre-trained characteristic extraction network to serve as first characteristics; and calculating the Euclidean distance between the first feature and the corresponding feature of each image in the image library, sequencing the first feature and the corresponding feature of each image in the image library, and outputting a sequencing result. The invention improves the robustness of cross-domain target re-identification.

Description

Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering
Technical Field
The invention belongs to the field of computer vision and pattern recognition, and particularly relates to a cross-domain target re-recognition method, a system and a device based on feature countermeasure learning and self-similarity clustering.
Background
Object re-identification belongs to a sub-problem in the field of image retrieval. Given an image of an object, typically a pedestrian image or a vehicle image, the object re-recognition task aims to find the object image in other scenes. In recent years, vehicle re-identification and pedestrian re-identification have become the focus of research in the field of computer vision, and many methods based on deep learning have achieved good results. However, most of these methods based on deep learning require a large amount of labeled training data, and due to the existence of domain differences, the model trained on one data set is tested with another data set, and the performance is greatly reduced. Since labeling a large number of samples on a large target data set is time-consuming and labor-consuming, the domain adaptive method becomes a more economical choice.
The goal of unsupervised domain adaptation is to migrate discriminant information learned from one fully labeled data domain (source domain) to another unlabeled data domain (target domain). Conventional unsupervised domain adaptive methods map samples onto a domain-independent, discriminative-power-preserving feature representation, thereby migrating the discriminative information learned on the source domain to the target domain. Many approaches limit the domain independence of feature expression from different perspectives, such as statistics-based or counterlearning-based approaches. The former thought that in the target re-identification task, the identity information of the source domain and the target domain are completely different, i.e. there is no overlap of the labels, so the above-mentioned method cannot be applied to the re-identification task. However, the present invention recognizes that using these domain-independent methods does not necessarily require strict agreement between the label spaces of the source domain and the target domain.
The existing cross-Domain target Re-recognition method mainly focuses on Image translation from a source Domain to a target Domain by using antagonistic learning, and then trains a Re-recognition model by using images after translation, such as Image-Image Domain adaptive with preset Self-Similarity and Domain-Similarity for Person Re-Identification and Person Transfer GAN to Bridge Domain Gap for Person Re-Identification, which respectively improve a classical Image translation algorithm CycleGAN from different angles. In addition, A Simple unscuperved Cross Domain AdaptationAppcroach for Person Re-identification A text uses a model pre-trained on a source Domain to extract features of a target Domain sample, and then uses a clustering to generate pseudo-labels for training a Re-recognition model. However, in this method, the number of cluster centers is varied, which results in that it cannot use other objective functions that require a fixed number of classes, such as cross entropy loss and center loss.
Disclosure of Invention
In order to solve the above problems in the prior art, that is, to solve the problem that the existing target re-identification method has poor robustness of the identification result due to the fact that the number of clustering centers is not fixed and the discrimination power of feature expression is limited, the first aspect of the present invention provides a cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering, the method comprising:
step S100, acquiring an image to be identified as an input image;
step S200, extracting the characteristics of the input image through a pre-trained characteristic extraction network to serve as first characteristics; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network is as follows;
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
In some preferred embodiments, the feature extraction network, when trained using the source domain training dataset, has a corresponding loss function of:
Figure BDA0002556058630000031
Figure BDA0002556058630000032
wherein,
Figure BDA0002556058630000033
representing the loss value, x, corresponding to the feature extraction networksRepresenting each training sample in a source domain training dataset, ysRepresenting the true label, n, corresponding to each training sample in the source domain training datasetsRepresenting the number of training samples, X, in the source domain training datasetsA set of source domain training data is represented,
Figure BDA0002556058630000034
which represents the cross-entropy loss in the entropy domain,
Figure BDA0002556058630000035
a loss of a triplet is represented as,
Figure BDA0002556058630000036
denotes the center loss, λtri、λcenAnd representing the weight value corresponding to each loss function.
In some preferred embodiments, the self-similarity clustering is a K-means clustering algorithm.
In some preferred embodiments, the pseudo tag alignment algorithm is performed by the following method.
Step B10, establishing a corresponding relation between each cluster of the K-1 th cluster and the cluster with the same pseudo label number and the maximum pseudo label number of the K-th cluster; wherein K is a positive integer;
step B20, if the corresponding relations are many-to-one, keeping the corresponding relation of the cluster with the most number of the same pseudo labels in the K-1 th cluster, and deleting other corresponding relations;
and B30, deleting the clusters corresponding to the K-1 th cluster, establishing a corresponding relationship between the clusters which have the same pseudo label quantity as the K-1 th cluster and have the largest number of pseudo labels and do not have the corresponding relationship, and jumping to the step B20 until the clusters of the K-1 th cluster and the clusters of the K-1 th cluster establish a one-to-one relationship.
In some preferred embodiments, the countermeasure loss is constructed by:
Figure BDA0002556058630000041
Figure BDA0002556058630000042
wherein x istRepresenting each training sample in the target domain training dataset, gamma representing the balance weight of the gradient penalty term, fwA mapping function corresponding to the domain discriminant network for mapping an input d-dimensional feature vector into a real number, fg(xs) Representing extracted source domain features, fg(xt) Representing extracted target Domain features, ntRepresenting the number of training samples, X, in the target field training datasettA set of target domain training data is represented,
Figure BDA0002556058630000043
a gradient penalty term is represented that is,
Figure BDA0002556058630000044
representing the source domain features, the target domain features and a random point on the determined straight line of the source domain features and the target domain features, thetag、θwRespectively representing the learning parameters of the feature extraction network and the domain discrimination network.
In some preferred embodiments, the gradient penalty term is obtained by:
Figure BDA0002556058630000045
the invention provides a cross-domain target re-identification system based on feature countermeasure learning and self-similarity clustering, which comprises an acquisition module and an identification module;
the acquisition module is configured to acquire an image to be identified as an input image;
the recognition module is configured to extract features of the input image through a pre-trained feature extraction network to serve as first features; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network is as follows;
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
In a third aspect of the present invention, a storage device is provided, in which a plurality of programs are stored, the programs being loaded and executed by a processor to implement the above-mentioned cross-domain object re-identification method based on feature-confrontation learning and self-similarity clustering.
In a fourth aspect of the present invention, a processing apparatus is provided, which includes a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is suitable for being loaded and executed by a processor to realize the cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering.
The invention has the beneficial effects that:
the invention improves the robustness of cross-domain target re-identification. The invention introduces feature counterlearning into the cross-domain target re-identification problem, reduces the distance of the feature distribution of the source domain and the target domain through the counterlearning to map the feature distribution to a domain-independent feature expression space, and eliminates the limitation of discrimination of feature expression. Secondly, the invention designs a self-similarity clustering module, generates pseudo labels by mining the internal relation of the non-label target domain samples, fixes the number of clustering centers, and aligns the pseudo labels, thereby training with the source domain data to be suitable for more effective supervision loss functions. And finally, integrating the counterlearning and the self-similarity clustering into a unified framework for the cross-domain target re-identification task.
Meanwhile, the method guides the feature extraction network to extract the features which are irrelevant to the field and have discrimination power from the input target image, and improves the performance of target retrieval.
Drawings
Other features, objects and advantages of the present application will become more apparent upon reading of the following detailed description of non-limiting embodiments thereof, made with reference to the accompanying drawings.
FIG. 1 is a schematic flow chart of a cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering according to an embodiment of the present invention;
FIG. 2 is a block diagram of a cross-domain object re-recognition system based on feature-confrontation learning and self-similarity clustering in accordance with an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a training process of a cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering according to an embodiment of the present invention;
FIG. 4 is a schematic flow chart diagram of a pseudo tag alignment algorithm according to an embodiment of the present invention;
FIG. 5 is a schematic block diagram of a computer system suitable for use with the electronic device to implement an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The present application will be described in further detail with reference to the following drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the related invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict.
The invention relates to a cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering, which comprises the following steps:
step S100, acquiring an image to be identified as an input image;
step S200, extracting the characteristics of the input image through a pre-trained characteristic extraction network to serve as first characteristics; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network is as follows;
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
In order to more clearly describe the cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering, the following will expand the detailed description of each step in one embodiment of the method of the present invention with reference to the drawings.
In the following preferred embodiment, the training process of the feature extraction network is detailed first, and then the identification result obtained by the cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering is detailed.
1. Training process for feature extraction network
A100, acquiring a source domain training data set and a target domain training data set; extracting a network based on training sample pre-training characteristics in the source domain training data set;
for the cross-domain target re-identification problem, the following is defined in this embodiment: there are two data sets, one containing nsIs from the source domain
Figure BDA0002556058630000081
Wherein all data are labeled:
Figure BDA0002556058630000082
referred to as source domain training dataThe collection of the data is carried out,
Figure BDA0002556058630000083
a representation of a training sample is shown,
Figure BDA0002556058630000084
representing a real label corresponding to the training sample; the other data set contains ntIs from the target domain
Figure BDA0002556058630000085
Wherein all data are not labeled:
Figure BDA0002556058630000086
referred to as a target domain training data set,
Figure BDA0002556058630000087
representing a training sample. Assuming that the two image domains have the same feature space but obey different data distributions
Figure BDA0002556058630000091
And
Figure BDA0002556058630000092
the purpose of the cross-domain target re-identification problem is to learn a feature expression space which is discriminative on a target domain.
Firstly, a feature extraction network is pre-trained on source domain data, and the feature extraction network is constructed based on a convolutional neural network. Order to
Figure BDA0002556058630000093
The purpose of the feature extraction network is to learn a parameter θ for a sample over the source domaingFunction of (2)
Figure BDA0002556058630000094
Which can map this sample to a d-dimensional representation of the feature. During the training process, the network overall recognition loss function is a weighted sum of cross-entropy loss, triplet loss, and center loss. As shown in equation (1):
Figure BDA0002556058630000095
Figure BDA0002556058630000096
wherein,
Figure BDA0002556058630000097
representing the loss value, x, corresponding to the feature extraction networksRepresenting each training sample in a source domain training dataset, ysRepresenting labels, n, corresponding to training samples in the source domain training datasetsRepresenting the number of training samples, X, in the source domain training datasetsA set of source domain training data is represented,
Figure BDA0002556058630000098
which represents the cross-entropy loss in the entropy domain,
Figure BDA0002556058630000099
a loss of a triplet is represented as,
Figure BDA00025560586300000910
denotes the center loss, λtri、λcenAnd representing the weight value corresponding to each loss function.
Step A200, extracting the characteristics of each training sample in the target domain training data set through a trained characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
in this embodiment, in order to mine an implicit semantic relationship between unlabeled samples in a target domain, the present invention provides a self-similarity clustering module. By clustering the samples into several clusters using features extracted by the feature extraction network, pseudo-identity labels can be obtained, which allows unlabeled samples of the target domain to be trained together with labeled samples in the source domain. The invention fixes the number of cluster centers so that more discriminating loss functions can be applied, such as cross entropy loss, triplet loss, and center loss, which are not only more discriminating but also faster in convergence than using only triplet loss.
As shown in fig. 3, for all training samples in the target domain, their features after the global convergence layer (global convergence layer) are extracted, and then the features are clustered using the K-means algorithm. Since the number of clustering centers can be fixed by K-means clustering, the generated pseudo identity label can directly calculate the identification loss. At this time, for any sample in the target domain
Figure BDA0002556058630000101
A false identity label can be obtained by clustering
Figure BDA0002556058630000102
This clustering process may be performed at the beginning of each generation (epoch) training, and then the pseudo-identity label of each sample in the target domain is updated.
Step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the target domain training data set and the loss of the corresponding first label as a first loss;
because the initial value selection in K-means clustering is different, the same class may be assigned different pseudo-identity labels in two adjacent clustering operations, which is detrimental to the calculation of cross-entropy loss and center loss. In order to solve the problem, a one-to-one matching must be constructed on the clustering results of two times, so the invention provides a pseudo label alignment algorithm, which specifically comprises the following steps:
giving a label-free target domain data set
Figure BDA0002556058630000103
Self-similarity clustering can cluster each generation of training into several different clusters at the beginning of the training. Define the result of the k-th clustering as
Figure BDA0002556058630000104
Wherein
Figure BDA0002556058630000105
N is the number of clusters, which is a list of all samples in the qth cluster. The result of the k-1 th clustering is
Figure BDA0002556058630000106
The purpose of the pseudo label alignment algorithm is to construct a slave Ck-1To CkOne-to-one matching graph of
Figure BDA0002556058630000107
Making the labels between two clusters invariant as many samples as possible, wherein
Figure BDA0002556058630000108
Is CkNeutralization of
Figure BDA0002556058630000109
The serial number of the corresponding cluster.
The pseudo label alignment algorithm is shown in fig. 4, where each circle represents a cluster in the cluster, and the number therein represents the number of samples contained in this cluster. First, establish Ck-1To each of CkThe corresponding relationship of the cluster having the most number of identical samples in (a) of fig. 4. Then, if there is more than one Ck-1Cluster of (5) and CkOne cluster of the clusters establishes a corresponding relationship (i.e., a many-to-one relationship), and only the cluster with the largest number of samples, i.e., the dotted line (dotted line with larger interval) in fig. 4(a), is retained, while the other corresponding relationship, i.e., the dotted line with smaller interval in fig. 4(a), is discarded. Then, the corresponding relation of the clusters which are discarded are processed in CkFind the cluster with the maximum number of the same samples but no corresponding relationship at present, and establish the corresponding relationship, as shown in fig. 4 (b). Finally, this process is repeated until the entire matching graph becomes one-to-one, as shown in fig. 4 (c).
And calculating the loss of the feature extracted from each training sample in the target domain training data set and the corresponding first label as a first loss.
Step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
in this embodiment, the countermeasure loss is calculated by the feature extraction network and the domain discrimination network based on the features of the source domain and the target domain, and the countermeasure learning is performed. The domain discrimination network is constructed based on a convolutional neural network, and the function of the domain discrimination network is similar to that of a discriminator. The calculation of the resistance loss is specifically as follows:
the antagonistic learning of the invention is similar to the WDGRL algorithm proposed by Wasserstein Distance Guided responsiveness learning for Domain Adaptation to realize the characteristic antagonistic learning, and aims to reduce the difference of the characteristic distribution of a source Domain and a target Domain. The present invention uses a domain discrimination network whose purpose is to estimate the Wasserstein distance expressed by the source domain and target domain features. Feature expression h ═ f for giving one feature extraction network outputg(x) The domain discriminant network learns a parameter thetawFunction of (2)
Figure BDA0002556058630000111
The method maps a d-dimensional feature expression to a real number, and the difference value of corresponding outputs of the source domain feature and the target domain feature can be used for expressing Wasserstein distance. If all the parameters of the domain discriminant network satisfy the Lipschitz condition, the Wasserstein distance can be obtained from the maximized domain discriminant loss, as shown in equation (2):
Figure BDA0002556058630000112
wherein x istRepresenting each training sample in the target field training dataset, gamma, fwA mapping function corresponding to the domain discriminant network for mapping an input d-dimensional feature vector into a real number, fg(xs) Representing extracted source domain features, fg(xt) Representing extracted target Domain features, ntRepresenting the number of training samples, X, in the target field training datasettRepresenting a target domain training data set.
In order to guarantee the Lipschitz condition, a gradient penalty term is added in the invention, as shown in formula (3):
Figure BDA0002556058630000121
wherein,
Figure BDA0002556058630000122
a gradient penalty term is represented that is,
Figure BDA0002556058630000123
and the random point on the straight line is determined by representing the source domain characteristics, the target domain characteristics and the source domain characteristics and the target domain characteristics.
The goal of the feature countermeasure learning module is to solve the minimum maximum problem as shown in equation (4):
Figure BDA0002556058630000124
where γ represents the equilibrium weight of the gradient penalty term.
The confrontation loss is taken as a second loss, and the loss of the feature extracted from each training sample in the first data set and the corresponding real label is calculated as a third loss (namely, the recognition loss).
Step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
in this embodiment, the whole unsupervised cross-domain re-identification framework proposed by the present invention is composed of feature countermeasure learning, self-similarity clustering, and pseudo tag alignment algorithm. The overall objective function is shown in equation (5):
Figure BDA0002556058630000125
Figure BDA0002556058630000126
wherein,
Figure BDA0002556058630000127
for the pseudo-identity label obtained after self-similarity clustering and pseudo-label alignment, which is updated at the beginning of each generation of training, lambdacriThe domain is weighted against the loss.
Figure BDA0002556058630000128
Consistent with the calculation method of equation (1), only the parameters are different.
And updating parameters of the feature extraction network based on the overall loss objective function.
And step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
In summary, the training of the whole network is divided into two stages, the first stage is to pre-train the feature extraction network on the source domain sample, and then the second stage is to jointly optimize the feature extraction network and the domain discrimination network by using the self-similarity clustering and feature confrontation learning module on the source domain sample and the target domain sample, so as to obtain a better feature extraction network by jointly using the source domain sample and the target domain sample for training, learn a domain-independent feature space with strong discrimination, and improve the target re-identification performance on the target domain.
2. A cross-domain target re-identification method based on feature counterstudy and self-similarity clustering is shown in FIG. 1
Step S100, acquiring an image to be identified as an input image;
in this embodiment, an image to be recognized, typically an image of a target domain, is acquired.
Step S200, extracting the characteristics of the input image through a pre-trained characteristic extraction network to serve as first characteristics; and calculating and sequencing Euclidean distances between the first features and the features corresponding to the images in the image library, and outputting a sequencing result.
In the embodiment, the euclidean distance between the image to be recognized and each target feature in the search library (i.e., the image library) is calculated based on the trained feature extraction network extraction features. And finally, the calculated distances are arranged in an ascending order, and the higher the Rank-1 (the first Rank) and the matching rate of the top Rank is, the better the effect of the learned depth features on the target re-identification task is.
A cross-domain target re-recognition system based on feature countermeasure learning and self-similarity clustering according to a second embodiment of the present invention, as shown in fig. 2, includes: the method comprises an acquisition module 100 and an identification module 200;
the acquiring module 100 is configured to acquire an image to be recognized as an input image; (ii) a An image;
the recognition module 200 is configured to extract features of the input image through a pre-trained feature extraction network as first features; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network is as follows;
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process and related description of the system described above may refer to the corresponding process in the foregoing method embodiment, and details are not described herein again.
It should be noted that, the cross-domain object re-identification system based on feature countermeasure learning and self-similarity clustering provided in the foregoing embodiment is only illustrated by the division of the functional modules, and in practical applications, the above function allocation may be completed by different functional modules according to needs, that is, the modules or steps in the embodiment of the present invention are further decomposed or combined, for example, the modules in the foregoing embodiment may be combined into one module, or may be further split into multiple sub-modules, so as to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not to be construed as unduly limiting the present invention.
A storage device according to a third embodiment of the present invention stores therein a plurality of programs adapted to be loaded by a processor and to implement the above-described cross-domain object re-recognition method based on feature-confrontation learning and self-similarity clustering.
A processing apparatus according to a fourth embodiment of the present invention includes a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; the program is adapted to be loaded and executed by a processor to implement the above-described cross-domain object re-recognition method based on feature-confrontation learning and self-similarity clustering.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes and related descriptions of the storage device and the processing device described above may refer to the corresponding processes in the foregoing method examples, and are not described herein again.
Reference is now made to FIG. 5, which is a block diagram illustrating a computer system suitable for use as a server in implementing embodiments of the present methods, systems, and apparatus. The server shown in fig. 5 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.
As shown in fig. 5, the computer system includes a Central Processing Unit (CPU)501 that can perform various appropriate actions and processes according to a program stored in a Read Only Memory (ROM) 502 or a program loaded from a storage section 508 into a Random Access Memory (RAM) 503. In the RAM503, various programs and data necessary for system operation are also stored. The CPU 501, ROM502, and RAM503 are connected to each other via a bus 504. An Input/Output (I/O) interface 505 is also connected to bus 504.
The following components are connected to the I/O interface 505: an input portion 506 including a keyboard, a mouse, and the like; an output section 507 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, a speaker, and the like; a storage portion 508 including a hard disk and the like; and a communication section 509 including a network interface card such as a LAN (Local area network) card, a modem, or the like. The communication section 509 performs communication processing via a network such as the internet. The driver 510 is also connected to the I/O interface 505 as necessary. A removable medium 511 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 510 as necessary, so that a computer program read out therefrom is mounted into the storage section 508 as necessary.
In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method illustrated in the flow chart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 509, and/or installed from the removable medium 511. The computer program performs the above-described functions defined in the method of the present application when executed by the Central Processing Unit (CPU) 501. It should be noted that the computer readable medium mentioned above in the present application may be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In this application, however, a computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, fiber optic cable, RF, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The terms "first," "second," and the like are used for distinguishing between similar elements and not necessarily for describing or implying a particular order or sequence.
The terms "comprises," "comprising," or any other similar term are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.

Claims (9)

1. A cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering is characterized by comprising the following steps:
step S100, acquiring an image to be identified as an input image;
step S200, extracting the characteristics of the input image through a pre-trained characteristic extraction network to serve as first characteristics; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network comprises the following steps:
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
2. The method for cross-domain target re-recognition based on feature countermeasure learning and self-similarity clustering according to claim 1, wherein the feature extraction network has a corresponding loss function when it is trained by using a source domain training data set as follows:
Figure FDA0002556058620000021
wherein,
Figure FDA0002556058620000022
representing the loss value, x, corresponding to the feature extraction networksRepresenting each training sample in a source domain training dataset, ysRepresenting the true label, n, corresponding to each training sample in the source domain training datasetsRepresenting the number of training samples, X, in the source domain training datasetsA set of source domain training data is represented,
Figure FDA0002556058620000023
which represents the cross-entropy loss in the entropy domain,
Figure FDA0002556058620000024
a loss of a triplet is represented as,
Figure FDA0002556058620000025
denotes the center loss, λtri、λcenAnd representing the weight value corresponding to each loss function.
3. The method of claim 2, wherein the self-similarity clustering is a K-means clustering algorithm.
4. The cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering according to claim 1, wherein the pseudo label alignment algorithm comprises the following steps:
step B10, establishing a corresponding relation between each cluster of the K-1 th cluster and the cluster with the same pseudo label number and the maximum pseudo label number of the K-th cluster; wherein K is a positive integer;
step B20, if the corresponding relations are many-to-one, keeping the corresponding relation of the cluster with the most number of the same pseudo labels in the K-1 th cluster, and deleting other corresponding relations;
and B30, deleting the clusters corresponding to the K-1 th cluster, establishing a corresponding relationship between the clusters which have the same pseudo label quantity as the K-1 th cluster and have the largest number of pseudo labels and do not have the corresponding relationship, and jumping to the step B20 until the clusters of the K-1 th cluster and the clusters of the K-1 th cluster establish a one-to-one relationship.
5. The cross-domain target re-identification method based on feature countermeasure learning and self-similarity clustering according to claim 2, wherein the countermeasure loss is constructed by:
Figure FDA0002556058620000031
Figure FDA0002556058620000032
wherein x istRepresenting each training sample in the target domain training dataset, gamma representing the balance weight of the gradient penalty term, fwA mapping function corresponding to the domain discriminant network for mapping an input d-dimensional feature vector into a real number, fg(xs) Representing extracted source domain features, fg(xt) Representing extracted target Domain features, ntRepresenting the number of training samples, X, in the target field training datasettA set of target domain training data is represented,
Figure FDA0002556058620000033
a gradient penalty term is represented that is,
Figure FDA0002556058620000034
representing the source domain features, the target domain features and a random point on the determined straight line of the source domain features and the target domain features, thetag、θwRespectively representing the learning parameters of the feature extraction network and the domain discrimination network.
6. The method for cross-domain target re-identification based on feature-confronted learning and self-similarity clustering according to claim 5, wherein the gradient penalty term is obtained by:
Figure FDA0002556058620000035
7. a cross-domain object re-recognition system based on feature-confrontation learning and self-similarity clustering, the system comprising: the device comprises an acquisition module and an identification module;
the acquisition module is configured to acquire an image to be identified as an input image;
the recognition module is configured to extract features of the input image through a pre-trained feature extraction network to serve as first features; calculating and sequencing Euclidean distances between the first features and corresponding features of each image in an image library, and outputting sequencing results;
the training method of the feature extraction network comprises the following steps:
a100, acquiring a first data set and a second data set; pre-training a feature extraction network based on the first data set; the first data set is a source domain training data set which comprises training samples and corresponding real labels; the second data set is a target domain training data set;
step A200, extracting the characteristics of each training sample in the second data set through a pre-training characteristic extraction network, and acquiring a corresponding pseudo label through self-similarity clustering;
step A300, aligning each pseudo label through a preset pseudo label alignment algorithm, and taking the aligned label as a first label; respectively calculating the loss of the features extracted from each training sample in the second data set and the loss of the corresponding first label as a first loss;
step A400, calculating Wasserstein distances among feature distributions extracted from training samples in a first data set and a second data set through a pre-constructed domain discrimination network, and constructing a countermeasure loss as a second loss based on the distances; calculating the loss of the features extracted from each training sample in the first data set and the real label corresponding to the features as a third loss; the domain discrimination network is constructed based on a convolutional neural network;
step A500, updating parameters of the feature extraction network based on the first loss, the second loss and the third loss;
and step A600, the steps A200 to A500 are circulated until a trained feature extraction network is obtained.
8. A storage device having stored therein a plurality of programs, wherein said program applications are loaded and executed by a processor to implement the method of cross-domain object re-recognition based on feature-confronted learning and self-similarity clustering of any one of claims 1-6.
9. A processing device comprising a processor, a storage device; a processor adapted to execute various programs; a storage device adapted to store a plurality of programs; characterized in that the program is adapted to be loaded and executed by a processor to implement the method of cross-domain object re-recognition based on feature-confrontation learning and self-similarity clustering according to any of claims 1 to 6.
CN202010592282.4A 2020-06-24 2020-06-24 Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering Active CN111738172B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010592282.4A CN111738172B (en) 2020-06-24 2020-06-24 Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010592282.4A CN111738172B (en) 2020-06-24 2020-06-24 Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering

Publications (2)

Publication Number Publication Date
CN111738172A true CN111738172A (en) 2020-10-02
CN111738172B CN111738172B (en) 2021-02-12

Family

ID=72651089

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010592282.4A Active CN111738172B (en) 2020-06-24 2020-06-24 Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering

Country Status (1)

Country Link
CN (1) CN111738172B (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112347995A (en) * 2020-11-30 2021-02-09 中国科学院自动化研究所 Unsupervised pedestrian re-identification method based on fusion of pixel and feature transfer
CN112712570A (en) * 2020-12-22 2021-04-27 北京字节跳动网络技术有限公司 Image processing method, image processing apparatus, electronic device, and medium
CN112784674A (en) * 2020-11-13 2021-05-11 北京航空航天大学 Cross-domain identification method of key personnel search system based on class center self-adaption
CN112861705A (en) * 2021-02-04 2021-05-28 东北林业大学 Cross-domain pedestrian re-identification method based on hybrid learning
CN112990152A (en) * 2021-05-10 2021-06-18 中国科学院自动化研究所 Vehicle weight identification method based on key point detection and local feature alignment
CN113128411A (en) * 2021-04-22 2021-07-16 深圳市格灵精睿视觉有限公司 Cross-domain capture identification method and device, electronic equipment and storage medium
CN113221916A (en) * 2021-05-08 2021-08-06 哈尔滨工业大学（深圳） Visual sorting method and device based on cross-domain rapid migration
CN113505642A (en) * 2021-06-04 2021-10-15 北京大学 Method, device, equipment and storage medium for improving target re-identification generalization
CN114429436A (en) * 2022-01-25 2022-05-03 山东大学 Image migration method and system for reducing domain difference
CN114842256A (en) * 2022-01-25 2022-08-02 南京大学 Domain generalization image classification method based on semi-supervised learning
CN115171203A (en) * 2022-09-05 2022-10-11 珠海翔翼航空技术有限公司 Automatic identification method, system and equipment for pilot instrument monitoring execution degree
CN117153161A (en) * 2023-10-31 2023-12-01 中国传媒大学 Cross-domain voice authentication method and system based on domain invariant feature learning

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180253596A1 (en) * 2017-03-06 2018-09-06 Conduent Business Services, Llc System and method for person re-identification using overhead view images
CN108875487A (en) * 2017-09-29 2018-11-23 北京旷视科技有限公司 Pedestrian is identified the training of network again and is identified again based on its pedestrian
CN109271895A (en) * 2018-08-31 2019-01-25 西安电子科技大学 Pedestrian's recognition methods again based on Analysis On Multi-scale Features study and Image Segmentation Methods Based on Features
CN109919246A (en) * 2019-03-18 2019-06-21 西安电子科技大学 Pedestrian's recognition methods again based on self-adaptive features cluster and multiple risks fusion
US20190213399A1 (en) * 2018-01-08 2019-07-11 Samsung Electronics Co., Ltd. Apparatuses and methods for recognizing object and facial expression robust against change in facial expression, and apparatuses and methods for training
CN110135294A (en) * 2019-04-29 2019-08-16 中国科学院西安光学精密机械研究所 Pedestrian based on unsupervised cross-view metric learning recognition methods and system again
CN110555390A (en) * 2019-08-09 2019-12-10 厦门市美亚柏科信息股份有限公司 pedestrian re-identification method, device and medium based on semi-supervised training mode
CN110555428A (en) * 2019-09-12 2019-12-10 腾讯科技（深圳）有限公司 Pedestrian re-identification method, device, server and storage medium
CN110837850A (en) * 2019-10-23 2020-02-25 浙江大学 Unsupervised domain adaptation method based on counterstudy loss function
CN110942025A (en) * 2019-11-26 2020-03-31 河海大学 Unsupervised cross-domain pedestrian re-identification method based on clustering
CN111191793A (en) * 2019-12-18 2020-05-22 同济大学 Regularization-based method for solving problem of gradient disappearance of anti-residual transformation network

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180253596A1 (en) * 2017-03-06 2018-09-06 Conduent Business Services, Llc System and method for person re-identification using overhead view images
CN108875487A (en) * 2017-09-29 2018-11-23 北京旷视科技有限公司 Pedestrian is identified the training of network again and is identified again based on its pedestrian
US20190213399A1 (en) * 2018-01-08 2019-07-11 Samsung Electronics Co., Ltd. Apparatuses and methods for recognizing object and facial expression robust against change in facial expression, and apparatuses and methods for training
CN109271895A (en) * 2018-08-31 2019-01-25 西安电子科技大学 Pedestrian's recognition methods again based on Analysis On Multi-scale Features study and Image Segmentation Methods Based on Features
CN109919246A (en) * 2019-03-18 2019-06-21 西安电子科技大学 Pedestrian's recognition methods again based on self-adaptive features cluster and multiple risks fusion
CN110135294A (en) * 2019-04-29 2019-08-16 中国科学院西安光学精密机械研究所 Pedestrian based on unsupervised cross-view metric learning recognition methods and system again
CN110555390A (en) * 2019-08-09 2019-12-10 厦门市美亚柏科信息股份有限公司 pedestrian re-identification method, device and medium based on semi-supervised training mode
CN110555428A (en) * 2019-09-12 2019-12-10 腾讯科技（深圳）有限公司 Pedestrian re-identification method, device, server and storage medium
CN110837850A (en) * 2019-10-23 2020-02-25 浙江大学 Unsupervised domain adaptation method based on counterstudy loss function
CN110942025A (en) * 2019-11-26 2020-03-31 河海大学 Unsupervised cross-domain pedestrian re-identification method based on clustering
CN111191793A (en) * 2019-12-18 2020-05-22 同济大学 Regularization-based method for solving problem of gradient disappearance of anti-residual transformation network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
TONGGUANG NI ET AL.: "Relative Distance Metric Leaning Based on Clustering Centralization and Projection Vectors Learning for Person Re-Identification", 《 IEEE ACCESS》 *
崔鹏，范志旭: "基于域鉴别网络和域自适应的行人重识别", 《光电子·激光》 *

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112784674A (en) * 2020-11-13 2021-05-11 北京航空航天大学 Cross-domain identification method of key personnel search system based on class center self-adaption
CN112347995A (en) * 2020-11-30 2021-02-09 中国科学院自动化研究所 Unsupervised pedestrian re-identification method based on fusion of pixel and feature transfer
CN112347995B (en) * 2020-11-30 2022-09-23 中国科学院自动化研究所 Unsupervised pedestrian re-identification method based on fusion of pixel and feature transfer
CN112712570A (en) * 2020-12-22 2021-04-27 北京字节跳动网络技术有限公司 Image processing method, image processing apparatus, electronic device, and medium
CN112712570B (en) * 2020-12-22 2023-11-24 抖音视界有限公司 Image processing method, device, electronic equipment and medium
CN112861705B (en) * 2021-02-04 2022-07-05 东北林业大学 Cross-domain pedestrian re-identification method based on hybrid learning
CN112861705A (en) * 2021-02-04 2021-05-28 东北林业大学 Cross-domain pedestrian re-identification method based on hybrid learning
CN113128411A (en) * 2021-04-22 2021-07-16 深圳市格灵精睿视觉有限公司 Cross-domain capture identification method and device, electronic equipment and storage medium
CN113221916A (en) * 2021-05-08 2021-08-06 哈尔滨工业大学（深圳） Visual sorting method and device based on cross-domain rapid migration
CN113221916B (en) * 2021-05-08 2023-07-07 哈尔滨工业大学（深圳） Vision picking method and device based on cross-domain rapid migration
CN112990152A (en) * 2021-05-10 2021-06-18 中国科学院自动化研究所 Vehicle weight identification method based on key point detection and local feature alignment
CN112990152B (en) * 2021-05-10 2021-07-30 中国科学院自动化研究所 Vehicle weight identification method based on key point detection and local feature alignment
CN113505642A (en) * 2021-06-04 2021-10-15 北京大学 Method, device, equipment and storage medium for improving target re-identification generalization
CN113505642B (en) * 2021-06-04 2023-10-24 北京大学 Method, device, equipment and storage medium for improving target re-identification generalization
CN114842256A (en) * 2022-01-25 2022-08-02 南京大学 Domain generalization image classification method based on semi-supervised learning
CN114429436A (en) * 2022-01-25 2022-05-03 山东大学 Image migration method and system for reducing domain difference
CN115171203A (en) * 2022-09-05 2022-10-11 珠海翔翼航空技术有限公司 Automatic identification method, system and equipment for pilot instrument monitoring execution degree
CN117153161A (en) * 2023-10-31 2023-12-01 中国传媒大学 Cross-domain voice authentication method and system based on domain invariant feature learning

Also Published As

Publication number Publication date
CN111738172B (en) 2021-02-12

Similar Documents

Publication Publication Date Title
CN111738172B (en) Cross-domain target re-identification method based on feature counterstudy and self-similarity clustering
CN113378632B (en) Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method
CN110175527B (en) Pedestrian re-identification method and device, computer equipment and readable medium
CN112015859A (en) Text knowledge hierarchy extraction method and device, computer equipment and readable medium
WO2021253510A1 (en) Bidirectional interactive network-based pedestrian search method and system, and device
CN111914156A (en) Cross-modal retrieval method and system for self-adaptive label perception graph convolution network
CN113158815B (en) Unsupervised pedestrian re-identification method, system and computer readable medium
CN111542841A (en) System and method for content identification
CN111046275A (en) User label determining method and device based on artificial intelligence and storage medium
CN114692732B (en) Method, system, device and storage medium for updating online label
CN116910571B (en) Open-domain adaptation method and system based on prototype comparison learning
CN112560823B (en) Adaptive variance and weight face age estimation method based on distribution learning
US11526807B2 (en) Machine learning systems and methods with source-target adaptation
CN117611932A (en) Image classification method and system based on double pseudo tag refinement and sample re-weighting
CN113870863B (en) Voiceprint recognition method and device, storage medium and electronic equipment
CN115062709A (en) Model optimization method, device, equipment, storage medium and program product
CN111898528B (en) Data processing method, device, computer readable medium and electronic equipment
CN115222047A (en) Model training method, device, equipment and storage medium
CN112149623B (en) Self-adaptive multi-sensor information fusion system, method and storage medium
CN116777814A (en) Image processing method, apparatus, computer device, storage medium, and program product
CN114387465A (en) Image recognition method and device, electronic equipment and computer readable medium
CN112733925A (en) Method and system for constructing light image classification network based on FPCC-GAN
CN111401112A (en) Face recognition method and device
CN117574179B (en) Method and device for constructing multi-task learning model
CN117152467B (en) Image recognition method, device, medium and electronic equipment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20240619

Address after: 200-19, 2nd Floor, Building B, Wanghai Building, No.10 West Third Ring Middle Road, Haidian District, Beijing, 100036

Patentee after: Zhongke Zidong Taichu (Beijing) Technology Co.,Ltd.

Country or region after: China

Address before: 100190 No. 95 East Zhongguancun Road, Beijing, Haidian District

Patentee before: INSTITUTE OF AUTOMATION, CHINESE ACADEMY OF SCIENCES

Country or region before: China

TR01 Transfer of patent right