CN113780066B - Pedestrian re-recognition method and device, electronic equipment and readable storage medium - Google Patents
Pedestrian re-recognition method and device, electronic equipment and readable storage medium Download PDFInfo
- Publication number
- CN113780066B CN113780066B CN202110864392.6A CN202110864392A CN113780066B CN 113780066 B CN113780066 B CN 113780066B CN 202110864392 A CN202110864392 A CN 202110864392A CN 113780066 B CN113780066 B CN 113780066B
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- recognition
- initial model
- classification
- expert network
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 57
- 230000006870 function Effects 0.000 claims description 49
- 238000012549 training Methods 0.000 claims description 33
- 238000010586 diagram Methods 0.000 claims description 21
- 238000012545 processing Methods 0.000 claims description 15
- 239000013598 vector Substances 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 8
- 238000000638 solvent extraction Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 5
- 238000012546 transfer Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000001627 detrimental effect Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000003116 impacting effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The application discloses a pedestrian re-identification method, a device, electronic equipment and a readable storage medium. The method comprises the steps of constructing an expert network and a pedestrian re-recognition initial model in advance, and performing knowledge supervision on a classification label and an embedded layer of the pedestrian re-recognition initial model by utilizing the expert network, so that knowledge obtained by the expert network is transferred to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; and deleting the classification layer of the pedestrian re-recognition initial model to obtain a pedestrian re-recognition model. Inputting the image to be identified into a pedestrian re-identification model to obtain the embedded features of the image to be identified, and comparing the embedded features with the embedded features of each search image in the database to obtain the identification result of the image to be identified. The pedestrian re-identification method and the pedestrian re-identification device can realize accurate and efficient pedestrian re-identification on the premise of not increasing the parameter quantity and the calculated quantity of the network model.
Description
Technical Field
The present disclosure relates to the field of computer vision, and in particular, to a pedestrian re-recognition method, device, electronic apparatus, and readable storage medium.
Background
The Person Re-identification (Re-ID) is also called as pedestrian Re-identification, and is a technology for judging whether a specific pedestrian exists in an image or a video sequence by using a computer vision technology, and whether the pedestrians in the fields of view of different cameras are the same pedestrian is determined by searching cameras distributed at different positions.
In the related technology, a pedestrian re-recognition model is obtained based on deep learning training, and the trained pedestrian re-recognition model is utilized to carry out image recognition on a video or an image, so that in order to improve the precision of the pedestrian re-recognition model, the network performance is generally improved by improving the network structure. However, as deep learning progresses, the network structure changes more and more. For different training data sets, selecting and training an optimal network model to adapt to the current pedestrian re-recognition task becomes more and more complex, and even cannot obtain the optimal. In particular, the related art may continue to improve network model performance by building more complex network structures. Deeper, wider or more complex networks often lead to a proliferation of parameters that are detrimental to the storage and deployment of portable devices. For example, implementing the deployment of real-time pedestrian detection recognition programs in a webcam requires a network with a smaller number of parameters and higher recognition accuracy. In addition, deeper, wider or more complex networks often lead to increased computational effort, which is detrimental to real-time demanding scene applications. Large computational delays can miss the best opportunity for the entire system, negatively impacting system functionality.
In view of this, how to improve the accuracy of pedestrian re-recognition without increasing the number of parameters and the amount of calculation of the network model is a technical problem that needs to be solved by those skilled in the art.
Disclosure of Invention
The application provides a pedestrian re-recognition method, a device, electronic equipment and a readable storage medium, which can realize accurate and efficient pedestrian re-recognition on the premise of not increasing the parameter quantity and the calculated quantity of a network model.
In order to solve the technical problems, the embodiment of the invention provides the following technical scheme:
in one aspect, the embodiment of the invention provides a pedestrian re-identification method, which comprises the following steps:
pre-constructing an expert network and a pedestrian re-recognition initial model, and performing knowledge supervision on a classification label and an embedding layer of the pedestrian re-recognition initial model by utilizing the expert network so as to transfer knowledge obtained by the expert network to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model;
inputting an image to be identified into the pedestrian re-identification model to obtain embedded features of the image to be identified;
and comparing the embedded features of the image to be identified with the embedded features of each search image in the database to obtain an identification result of the image to be identified.
Optionally, after the constructing the expert network and the initial model for pedestrian re-recognition, the method further includes:
obtaining an output characteristic diagram of the expert network;
partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams;
performing downsampling operation on each sub-output feature map to obtain local features of the input image of the expert network;
carrying out maximum pooling treatment on the output feature map to obtain global features of the input image;
and taking the local features and the global features as embedded features of the expert network.
Optionally, the pre-constructing the expert network and the pedestrian re-recognition initial model includes:
and constructing a pedestrian re-identification initial model by adopting a lightweight neural network structure.
Optionally, the performing knowledge supervision on the classification label and the embedded layer of the pedestrian re-recognition initial model by using the expert network includes:
obtaining corresponding classification characteristics according to the embedded characteristics of the expert network and the pedestrian re-recognition initial model;
processing each classification characteristic by using a classification function to obtain corresponding classification probability;
the classification probability of the pedestrian re-recognition initial model and the loss function of each branch of the expert network are calculated in sequence to complete knowledge supervision of the classification labels of the pedestrian re-recognition initial model;
And carrying out knowledge supervision on the embedded layer of the pedestrian re-recognition initial model according to the balance coefficient of the embedded layer characteristics of the expert network and the pedestrian re-recognition initial model, the embedded characteristics of the expert network and the embedded characteristics of the pedestrian re-recognition initial model.
Optionally, after the constructing the expert network and the pedestrian re-recognition initial model, the method includes:
determining training loss functions of the expert network and the pedestrian re-recognition initial model according to the classification characteristics, the classification labels and the embedded characteristics of the expert network and the pedestrian re-recognition initial model; the training loss function L is:
L＝L p +L c +L e ；
wherein L is p ＝L(p r ，y hard )＝cross_entropy(softmax(c r )，y hard )，L p Classifying loss function, p, of the initial model for re-identifying the pedestrian r Classifying probability, y, of the initial model for re-recognition of the pedestrian hard C, for inputting the true label of the image r Identifying classification features of an initial model again for the pedestrian, wherein cross_entropy represents a cross entropy loss function; l (L) c As a function of the tag loss,p (b) represents the b-th classification header of the expert network, N is the total number of classification headers of the expert network, p (l) represents the classification header output of the pedestrian re-recognition initial model, C is the vector dimensions of p (b) and p (l), and C represents C-th element of the corresponding vector is shown, and N is the total number of classification heads of the expert network; l (L) e For embedding loss function->e (k) is the embedded feature of the expert network, e (l) is the embedded feature of the pedestrian re-recognition initial model, alpha k And re-identifying the balance coefficient of the embedded layer characteristics of the initial model for the expert network and the pedestrian, wherein K is the kth branch of the expert network, and K is the total number of branches of the expert network.
Optionally, before determining the training loss function of the expert network and the pedestrian re-recognition initial model according to the classification features, the classification labels and the embedded features of the expert network and the pedestrian re-recognition initial model, the method further includes:
calling a regression network relation to calculate the balance coefficient, wherein the regression network relation is alpha k ＝ReLU(W k [e(k),e(l)])；
Wherein ReLU represents a linear rectification function, W k [e(k)，e(l)]The embedded feature e (k) and the embedded feature e (l) of the kth branch of the expert network are subjected to feature stitching by the full connection layer.
Another aspect of the embodiment of the present invention provides a pedestrian re-recognition apparatus, including:
the model training module is used for pre-constructing an expert network and a pedestrian re-recognition initial model, and performing knowledge supervision on a classification label and an embedding layer of the pedestrian re-recognition initial model by utilizing the expert network so as to transfer knowledge obtained by the expert network to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model;
The feature extraction module is used for inputting the image to be identified into the pedestrian re-identification model to obtain the embedded features of the image to be identified;
and the image recognition module is used for obtaining a recognition result of the image to be recognized by comparing the embedded features of the image to be recognized with the embedded features of each search image in the database.
Optionally, the model training module includes a feature output unit, configured to obtain an output feature map of the expert network; partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams; performing downsampling operation on each sub-output feature map to obtain local features of the input image of the expert network; carrying out maximum pooling treatment on the output feature map to obtain global features of the input image; and taking the local features and the global features as embedded features of the expert network.
The embodiment of the invention also provides electronic equipment, which comprises a processor, wherein the processor is used for realizing the steps of the pedestrian re-identification method when executing the computer program stored in the memory.
Finally, an embodiment of the present invention provides a readable storage medium, where a computer program is stored, where the computer program, when executed by a processor, implements the steps of the pedestrian re-identification method according to any one of the preceding claims.
The technical scheme provided by the application has the advantages that the expert network generally has stronger learning ability than the pedestrian re-recognition initial model, and the learned global features and detail features with more discrimination can be transmitted to the pedestrian re-recognition initial model through the action of knowledge supervision so as to be used for the pedestrian re-recognition initial model to imitate learning, so that the trained pedestrian re-recognition initial model is obtained. The classification layer of the pedestrian re-recognition initial model is deleted to serve as a model for executing the pedestrian re-recognition task, network potential can be mined maximally on the premise of not increasing the number of reasoning parameters and the calculated amount, the performance of the deep learning network is improved, the accuracy of the pedestrian re-recognition model is further improved, and accurate and efficient pedestrian re-recognition is realized.
In addition, the embodiment of the invention also provides a corresponding implementation device, electronic equipment and a readable storage medium for the pedestrian re-identification method, so that the method is more practical, and the device, the electronic equipment and the readable storage medium have corresponding advantages.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the related art, the drawings that are required to be used in the embodiments or the description of the related art will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort to those of ordinary skill in the art.
Fig. 1 is a schematic flow chart of a pedestrian re-recognition method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a model training process for an exemplary application scenario provided by an embodiment of the present invention;
fig. 3 is a schematic diagram of a pedestrian re-recognition process of an exemplary application scenario provided in an embodiment of the present invention;
FIG. 4 is a block diagram of a pedestrian re-recognition device according to an embodiment of the present invention;
fig. 5 is a block diagram of an embodiment of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to better understand the aspects of the present invention, the present invention will be described in further detail with reference to the accompanying drawings and detailed description. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims of this application and in the above-described figures, are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements but may include other steps or elements not expressly listed.
Having described the technical solutions of embodiments of the present invention, various non-limiting implementations of the present application are described in detail below.
Referring first to fig. 1, fig. 1 is a flow chart of a pedestrian re-recognition method according to an embodiment of the present invention, where the embodiment of the present invention may include the following:
s101: pre-constructing an expert network and a pedestrian re-recognition initial model, and performing knowledge supervision on a classification label and an embedded layer of the pedestrian re-recognition initial model by using the expert network so as to transfer knowledge obtained by the expert network to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; and deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model.
The expert network and the pedestrian re-recognition initial model in the embodiment can be built based on any deep learning network structure, the pedestrian re-recognition initial model is used as a learning network, the expert network is responsible for guiding training, and the pedestrian re-recognition initial model can obtain better results than learning without guidance through the expert network. The expert network may choose a model that is more powerful than the learning network, but is not completely limited. The expert network has stronger capability and can learn richer local detail features and global features with more discrimination. The learning network is a layer of a shallow layer generally, has fewer parameters and calculation amount, and is easier to deploy in an actual environment. In order to more easily deploy the pedestrian re-recognition model for executing the pedestrian re-recognition task, a lightweight neural network structure can be adopted to build a pedestrian re-recognition initial model. For example, the expert network may be constructed based on the ResNet50 basic structure, and those skilled in the art may refer to the specific structure of ResNet50 described in the related art, and will not be described here again. The initial model of pedestrian re-recognition may be constructed based on the CMSNet basic structure. CMSNet is a stack of multi-layer neural network layer structures including convolutional layers conv2d, max-pooling layers, average-pooling layers, full-connectivity layers fc and CMS block (a special network structure composed of multiple convolutional layers unique to CMSNet). Those skilled in the art can refer to the specific structure of CMSNet described in the related art, and the description thereof will be omitted here.
In the training process of the pedestrian re-recognition model, knowledge supervision is performed on the classification labels and the embedded layers of the pedestrian re-recognition initial model by utilizing the expert network, knowledge learned by the expert network is transferred to the pedestrian re-recognition model, and the pedestrian re-recognition model performs model training by combining a large amount of training image sample data based on the knowledge, so that the effects of uncomplicated network structure and good network performance are obtained. After the initial model of pedestrian re-recognition is trained, the model can execute the task of pedestrian re-recognition, and as the trained initial model of pedestrian re-recognition directly extracts the characteristics of the input image, the characteristics can be compared with the images in the database without classifying the characteristics, the classifying layer of the initial model of pedestrian re-recognition can be deleted, and the initial model of pedestrian re-recognition after deletion is the pedestrian re-recognition model for executing the task of recognition.
According to the embodiment, the obtained pedestrian re-recognition model can be deployed in physical equipment needing to execute the pedestrian re-recognition task, such as a network camera, and the generalization capability and performance of the pedestrian re-recognition model can be improved on the basis of not increasing the parameter quantity and the calculation quantity of the network of the pedestrian re-recognition model in reasoning by migrating knowledge in an expert network to the reasoning network.
S102: and inputting the image to be identified into the pedestrian re-identification model to obtain the embedded characteristics of the image to be identified.
The image to be identified in the step is the image of the pedestrian to be identified currently, and the embedded features are the global features and the local features of the extracted image to be identified.
S103: and comparing the embedded features with the embedded features of each search image in the database to obtain the identification result of the image to be identified.
The database is a gallery for finding whether there is a pedestrian image identical to the pedestrian in the image to be recognized, and for convenience of description and distinction, each image in the database may be referred to as a search image. And the pedestrian re-identification model can be utilized to correspondingly process each search image of the database to obtain the embedded feature of each search image, whether a target image of the pedestrian of the image to be identified exists in the database is judged by calculating the similarity or the difference between the embedded feature of the image to be identified and the embedded feature of each search image, and the target image is output to serve as the identification result of the image to be identified. The similarity between the image to be identified and the retrieval image can be determined by calculating the Euclidean distance or cosine distance.
In the technical scheme provided by the embodiment of the invention, the expert network generally has stronger learning ability than the pedestrian re-recognition initial model, and the learned global features and detailed features with better discrimination can be transmitted to the pedestrian re-recognition initial model through the action of knowledge supervision so as to be used for the pedestrian re-recognition initial model to simulate learning, thereby obtaining the trained pedestrian re-recognition initial model. The classification layer of the pedestrian re-recognition initial model is deleted to serve as a model for executing the pedestrian re-recognition task, network potential can be mined maximally on the premise of not increasing the number of reasoning parameters and the calculated amount, the performance of the deep learning network is improved, the accuracy of the pedestrian re-recognition model is further improved, and accurate and efficient pedestrian re-recognition is realized.
It should be noted that, in the present application, the steps may be executed simultaneously or in a certain preset order as long as the steps conform to the logic order, and fig. 1 is only a schematic manner and does not represent only such an execution order.
It can be appreciated that a key issue of pedestrian re-recognition is how to extract rich and discernable features from a pedestrian image, while observations from multiple different perspectives of the same image can be mutually aided, thereby exploiting the intelligence of the population to obtain more accurate results. Wherein the plurality of results includes both the final result and the intermediate result. In order to further improve performance of the pedestrian re-recognition model, in the process of performing model training on the pedestrian re-recognition model and the expert network, global features and local features of an input image are extracted, and the input image is any pedestrian image in a model training sample set. The global features and the local features have strong authenticability, and the combination of the two features can better serve a classification network to authenticate pedestrians. That is, the output features of the embedded layer of the expert network of the present embodiment are composed of local features and global features, and the generation manners of the local features and the global features may include:
Acquiring an output characteristic diagram of the expert network; partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams; and carrying out downsampling operation on each sub-output feature map to obtain local features of the input image of the expert network. Carrying out maximum pooling treatment on the output feature map to obtain global features of the input image; the global features are the integral features of the image, and the local features and the global features are used as embedded features of the expert network.
As shown in fig. 2, the expert network outputs a feature map at the last convolutional layer, the scale of the feature map may be the size of the fetch size×channel×height×width, abbreviated as b×c×h×w, and the feature map output at the last layer of the convolutional neural network is a higher level abstract feature of the input image. The battsize represents the number of images that are input per iteration of the training. For example, 1 image is input, and battsize=1. The BatchSize does not affect the final result analysis and calculation, so the BatchSize is omitted below. The local features of the pedestrian image have strong authenticability, for example, the head features of the pedestrian can be used for authenticating the pedestrian by the face, the trunk features consisting of the white scarf and the dark jacket can also be used for authenticating the identity of the pedestrian, and the dark trousers and shoes of the lower body have strong authenticability. The present embodiment may divide the feature map s1=b×c×h×w of the output of the expert network into blocks according to the dimension of the image height, where the division ratio may be, for example, 2:3: 5, each feature map can be divided into 3 blocks, and the dimensions of the three feature maps after the division are respectively as follows: b X C X (H/5) X W, B X C X (3H/10) X W, B X C X (H/2) X W. The three feature images after the block division are subjected to downward operation according to the dimension of B multiplied by C, for example, the three feature images can be subjected to downward sampling in a MaxPooling mode, and the sampled output features are as follows: e1 E2, e3. The dimension is as follows: and C dimension. B=1. The integral features of the pedestrian image are also highly identifiable. In order to classify pedestrians by using integral features, the method directly carries out MaxPooling on a feature map s1=B×C×H×W output by an expert network to obtain a feature representation e4 of the whole pedestrian image, wherein e4=C dimension. The integral features of the pedestrian image are also highly identifiable. In order to classify pedestrians by using the overall features, the embodiment may directly perform MaxPooling on the feature map s1=b×c×h×w output by the expert network, to obtain the overall feature representation e4 of the pedestrian image, where e4=c dimension. e1 E2, e3, e4 are collectively referred to as embedded features, which are features used for retrieval and comparison in the pedestrian re-recognition task, with which the final detection query of the pedestrian can be made.
According to the embodiment, the output characteristics of the expert network are subjected to blocking processing, the expert network can provide a plurality of view angles of the pedestrian image, the pedestrian re-recognition model learns the view angles, regularization constraint is provided among the view angles, and overfitting is prevented, so that the generalization capability of the pedestrian re-recognition model is improved.
In the above embodiment, the knowledge supervision on the initial model for re-identifying the pedestrian by using the expert network is not limited, and this embodiment provides an alternative implementation manner in conjunction with fig. 2, and may include the following steps:
obtaining corresponding classification characteristics according to the embedded characteristics of the expert network and the pedestrian re-identification initial model; processing each classification characteristic by using a classification function to obtain corresponding classification probability; the method comprises the steps of sequentially calculating the classification probability of a pedestrian re-recognition initial model and the loss function of each branch of an expert network to complete knowledge supervision on classification labels of the pedestrian re-recognition initial model; and carrying out knowledge supervision on the embedded layer of the pedestrian re-recognition initial model according to the balance coefficient of the embedded layer characteristics of the expert network and the pedestrian re-recognition initial model, the embedded characteristics of the expert network and the embedded characteristics of the pedestrian re-recognition initial model.
In this embodiment, the expert network is built based on a ResNet50 network, and the pedestrian re-recognition initial model adopts a CMSNet network. And respectively inputting the training sample images into an expert network and a pedestrian re-recognition initial model, outputting feature images by the expert network and the pedestrian re-recognition initial model, and obtaining embedded features corresponding to each output feature image by carrying out downsampling processing on the output feature images of the expert network and the pedestrian re-recognition initial model. The embedded features of the expert network comprise local features e1, e2 and e3 and global features e4, and the embedded feature of the pedestrian re-identification initial model is global features e5 of the input image. The embedded features e1, e2, e3, e4, e5 pass through a full connection layer full connected layer to obtain classification features c1, c2, c3, c4, c5. All classification features are subjected to an activation function, such as a softmax function, to obtain classification tags p1, p2, p3, p4, p5. The expert network knowledge supervision of the learning network comprises 2 parts, namely soft label softlabels-based knowledge supervision and embedded ebedding layer-based knowledge supervision. Firstly, calculating a classification Loss function los of a pedestrian re-recognition initial model, wherein a final classification feature c5 of the pedestrian re-recognition initial model passes through a softmax function to obtain classification probability based on a learning network, the classification probability can be combined with a real label (such as hard labels in the figure), a cross entropy Loss function is calculated, and a calculation formula can be as follows:
L p ＝L(p5，y hard )＝cross_entropy(softmax(c5)，y hard )。
And sequentially calculating the classification probability p5 of the pedestrian re-recognition initial model and KL loss functions of all branches of the expert network, and realizing soft-label-based knowledge supervision of the expert network on the pedestrian re-recognition initial model. The calculation formula can be as follows:
where p1 and p5 are both C-dimensional vectors, C representing the C-th element of the corresponding vector. Traversing all p1, p2, p3, p4, one can obtain:
where p (b) represents the b-th classification header in the expert network, and in this embodiment, there are a total of 4 classification headers. p (l) represents the classification head output of the learning network, in this embodiment p (l) =p5.
The expert network calculates knowledge supervision of the learning network based on the ebedding layer, and the calculation formula can be as follows:
where e (l) represents the learning network's ebadd layer feature, in this embodiment e (l) =e5. e (K) represents the embedding layer feature of the expert network, and in this embodiment, the biggest value of k=4. Alpha k And the balance coefficient representing the characteristics of the pedestrian re-identification initial model and the embedding layer of the expert network is calculated and used for balancing the importance of each branch of the expert network on the learning network guidance. For example, for different images (severely blocked, noisy, or important discriminative markers), branches corresponding to different blocks of the image should have different guiding effects on the learning network, and the part with obvious characteristics should be assigned with larger guiding weights, so that better guiding effects can be obtained. In order to obtain better network model performance, the guiding function of the characteristics of each branch embellishing layer of the expert network on the learning network can be weighed through a regression network, namely, the balance coefficient can be calculated by calling a regression network relational expression, and the regression network relational expression can be expressed as alpha k ＝ReLU(W k [e(k),e(l)])；
Wherein ReLU represents a linear rectification function, W represents a fully-connected layer, W k [e(k)， e(l)]The embedded feature e (k) and the embedded feature e (l) of the kth branch of the expert network are subjected to feature splicing by the full connection layer, wherein the feature splicing means that two features are connected end to form a vector with higher dimension. By W k The stitching feature may be projected to a discrete value that ultimately passes the ReLAnd mapping the U activation function to finally obtain the weight value of each branch of the expert network.
Based on the above, the training loss function of the expert network and the pedestrian re-recognition initial model can be determined according to the classification characteristics, the classification labels and the embedded characteristics of the expert network and the pedestrian re-recognition initial model; the training loss function L can be expressed as:
L＝L p +L c +L e ；
wherein L is p ＝L(p r ，y hard )＝cross_entropy(softmax(c r )，y hard )，L p Classification loss function, p, for pedestrian re-recognition initial model r Classification probability, y, of the initial model for pedestrian re-recognition hard C, for inputting the true label of the image r Identifying classification features of the initial model again for pedestrians, wherein cross_entropy represents a cross entropy loss function; l (L) c As a function of the tag loss,p (b) represents the b-th classification head of the expert network, N is the total number of classification heads of the expert network, p (l) represents the classification head output of the pedestrian re-recognition initial model, C is the vector dimension of p (b) and p (l), C represents the C-th element of the corresponding vector, and N is the total number of classification heads of the expert network; l (L) e In order to embed the loss function,e (k) is the embedded feature of the expert network, e (l) is the embedded feature of the pedestrian re-recognition initial model, alpha k And re-identifying the balance coefficient of the embedded layer characteristic of the initial model for the expert network and the pedestrian, wherein K is the kth branch of the expert network, and K is the total number of branches of the expert network.
After training to obtain an initial pedestrian re-recognition model, deleting the classification layer to obtain a pedestrian re-recognition model, as shown in fig. 3, and the reasoning process flow chart is as follows:
and (3) sending any test image into a learning network, namely a pedestrian re-identification model, and obtaining embedded features of the test image. The embedded features and the embedded features of the images in the database, the database or referred to as a comparison library, are usually established in advance, and the image features are generally extracted and constructed by using a trained learning network. In the comparison, the euclidean distance or cosine distance between vectors is usually obtained. Thus, the image feature label closest to the feature of the test image in the database is the final label of the test image. From the above, it can be seen that the embedded features are applied in the final reasoning process, so that it is necessary to constrain the embedded features in the training.
Therefore, the embodiment can realize accurate and efficient pedestrian re-recognition on the premise of not increasing the parameter quantity and the calculated quantity of the network model.
The embodiment of the invention also provides a corresponding device for the pedestrian re-identification method, so that the method has higher practicability. Wherein the device may be described separately from the functional module and the hardware. The pedestrian re-recognition device provided by the embodiment of the invention is introduced below, and the pedestrian re-recognition device described below and the pedestrian re-recognition method described above can be referred to correspondingly.
Based on the angles of the functional modules, referring to fig. 4, fig. 4 is a structural diagram of a pedestrian re-recognition device provided by an embodiment of the present invention under a specific implementation manner, where the device may include:
the model training module 401 is configured to pre-construct an expert network and a pedestrian re-recognition initial model, and perform knowledge supervision on a classification tag and an embedded layer of the pedestrian re-recognition initial model by using the expert network so as to migrate knowledge obtained by the expert network to the pedestrian re-recognition initial model, thereby obtaining a trained pedestrian re-recognition initial model; and deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model.
The feature extraction module 402 is configured to input an image to be identified into the pedestrian re-identification model, and obtain an embedded feature of the image to be identified.
The image recognition module 403 is configured to obtain a recognition result of the image to be recognized by comparing the embedded feature with the embedded feature of each search image in the database.
Optionally, in some implementations of the present embodiment, the model training module 401 may include a feature output unit, configured to obtain an output feature map of the expert network; partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams; downsampling is carried out on each sub-output feature map to obtain local features of the input image of the expert network; carrying out maximum pooling treatment on the output feature map to obtain global features of the input image; the local features and the global features are used as embedded features of the expert network.
As an alternative implementation manner of this embodiment, the model training module 401 may be a module for building the pedestrian re-recognition initial model by adopting a lightweight neural network structure.
As another optional implementation manner of this embodiment, the model training module 401 may include a knowledge supervision unit, configured to obtain corresponding classification features according to the expert network and the embedded features of the pedestrian re-recognition initial model; processing each classification characteristic by using a classification function to obtain corresponding classification probability; the method comprises the steps of sequentially calculating the classification probability of a pedestrian re-recognition initial model and the loss function of each branch of an expert network to complete knowledge supervision on classification labels of the pedestrian re-recognition initial model; and carrying out knowledge supervision on the embedded layer of the pedestrian re-recognition initial model according to the balance coefficient of the embedded layer characteristics of the expert network and the pedestrian re-recognition initial model, the embedded characteristics of the expert network and the embedded characteristics of the pedestrian re-recognition initial model.
Optionally, in other implementations of the present embodiment, the apparatus may further include a loss function determining module configured to determine a training loss function of the expert network and the pedestrian re-recognition initial model according to classification features, classification labels, and embedded features of the expert network and the pedestrian re-recognition initial model; the training loss function L is:
L＝L p +L c +L e ；
wherein L is p ＝L(p r ，y hard )＝cross_entropy(softmax(c r )，y hard )，L p Classification loss function, p, for pedestrian re-recognition initial model r Classification probability, y, of the initial model for pedestrian re-recognition hard C, for inputting the true label of the image r Identifying classification features of the initial model again for pedestrians, wherein cross_entropy represents a cross entropy loss function; l (L) c As a function of the tag loss,p (b) represents the b-th classification head of the expert network, N is the total number of classification heads of the expert network, p (l) represents the classification head output of the pedestrian re-recognition initial model, C is the vector dimension of p (b) and p (l), C represents the C-th element of the corresponding vector, and N is the total number of classification heads of the expert network; l (L) e In order to embed the loss function,e (k) is the embedded feature of the expert network, e (l) is the embedded feature of the pedestrian re-recognition initial model, alpha k And re-identifying the balance coefficient of the embedded layer characteristic of the initial model for the expert network and the pedestrian, wherein K is the kth branch of the expert network, and K is the total number of branches of the expert network.
As another alternative implementation manner of this embodiment, the apparatus may further include a balance coefficient calculating module, configured to invoke a regression network relation to calculate the balance coefficient, where the regression network relation is α k ＝ReLU(W k [e(k),e(l)])；
Wherein ReLU represents a linear rectification function, W k [e(k)，e(l)]Representing that the full connection layer performs feature stitching on the embedded feature e (k) and the embedded feature e (l) of the kth branch of the expert network.
The functions of each functional module of the pedestrian re-recognition device according to the embodiment of the present invention may be specifically implemented according to the method in the embodiment of the method, and the specific implementation process may refer to the related description of the embodiment of the method, which is not repeated herein.
Therefore, the embodiment can realize accurate and efficient pedestrian re-recognition on the premise of not increasing the parameter quantity and the calculated quantity of the network model.
The pedestrian re-recognition device is described from the perspective of a functional module, and further, the application also provides an electronic device, which is described from the perspective of hardware. Fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application in an implementation manner. As shown in fig. 5, the electronic device comprises a memory 50 for storing a computer program; the processor 51 is configured to implement the steps of the pedestrian re-recognition method as mentioned in any one of the above embodiments when executing the computer program.
Processor 51 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and processor 51 may also be a controller, microcontroller, microprocessor, or other data processing chip, among others. The processor 51 may be implemented in at least one hardware form of DSP (Digital Signal Processing ), FPGA (Field-Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array ). The processor 51 may also include a main processor, which is a processor for processing data in an awake state, also called a CPU (Central Processing Unit ), and a coprocessor; a coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 51 may be integrated with a GPU (Graphics Processing Unit, image processor) for taking care of rendering and drawing of the content that the display screen is required to display. In some embodiments, the processor 51 may also include an AI (Artificial Intelligence ) processor for processing computing operations related to machine learning.
Memory 50 may include one or more computer-readable storage media, which may be non-transitory. Memory 50 may also include high-speed random access memory as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. The memory 50 may in some embodiments be an internal storage unit of the electronic device, such as a hard disk of a server. The memory 50 may in other embodiments also be an external storage device of the electronic device, such as a plug-in hard disk provided on a server, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card) or the like. Further, the memory 50 may also include both internal storage units and external storage devices of the electronic device. The memory 50 may be used to store not only application software installed in an electronic device, but also various types of data, such as: code of a program that executes the vulnerability processing method, or the like, may also be used to temporarily store data that has been output or is to be output. In this embodiment, the memory 50 is at least used for storing a computer program 501, where the computer program, when loaded and executed by the processor 51, can implement the relevant steps of the pedestrian re-recognition method disclosed in any one of the foregoing embodiments. In addition, the resources stored in the memory 50 may also include an operating system 502, data 503, and the like, where the storage mode may be transient storage or permanent storage. Operating system 502 may include Windows, unix, linux, among other things. The data 503 may include, but is not limited to, data corresponding to the pedestrian re-recognition result, and the like.
In some embodiments, the electronic device may further include a display 52, an input/output interface 53, a communication interface 54, or network interface, a power supply 55, and a communication bus 56. Among other things, the display 52, input output interface 53 such as a Keyboard (Keyboard) pertain to a user interface, which may optionally also include standard wired interfaces, wireless interfaces, etc. Alternatively, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch, or the like. The display may also be referred to as a display screen or display unit, as appropriate, for displaying information processed in the electronic device and for displaying a visual user interface. Communication interface 54 may optionally include a wired interface and/or a wireless interface, such as a WI-FI interface, a bluetooth interface, etc., typically used to establish a communication connection between an electronic device and other electronic devices. The communication bus 56 may be a peripheral component interconnect standard (peripheral component interconnect, PCI) bus or an extended industry standard architecture (extended industry standard architecture, EISA) bus, among others. The bus may be classified as an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is shown in fig. 5, but not only one bus or one type of bus.
Those skilled in the art will appreciate that the configuration shown in fig. 5 is not limiting of the electronic device and may include more or fewer components than shown, for example, may also include sensors 57 to perform various functions.
The functions of each functional module of the electronic device according to the embodiment of the present invention may be specifically implemented according to the method in the embodiment of the method, and the specific implementation process may refer to the related description of the embodiment of the method, which is not repeated herein.
Therefore, the embodiment can realize accurate and efficient pedestrian re-recognition on the premise of not increasing the parameter quantity and the calculated quantity of the network model.
It will be appreciated that if the pedestrian re-recognition method in the above embodiment is implemented in the form of a software functional unit and sold or used as a separate product, it may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution contributing to the prior art, or in a software product stored in a storage medium, performing all or part of the steps of the methods of the various embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), an electrically erasable programmable ROM, registers, a hard disk, a multimedia card, a card-type Memory (e.g., SD or DX Memory, etc.), a magnetic Memory, a removable disk, a CD-ROM, a magnetic disk, or an optical disk, etc., that can store program code.
Based on this, an embodiment of the present invention further provides a readable storage medium storing a computer program, which when executed by a processor, performs the steps of the pedestrian re-recognition method according to any one of the above embodiments.
In this specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, so that the same or similar parts between the embodiments are referred to each other. For the hardware including the device and the electronic equipment disclosed in the embodiments, the description is relatively simple because the hardware includes the device and the electronic equipment corresponding to the method disclosed in the embodiments, and relevant places refer to the description of the method.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative elements and steps are described above generally in terms of functionality in order to clearly illustrate the interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The above describes in detail a pedestrian re-recognition method, a device, an electronic device and a readable storage medium provided by the application. The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to facilitate an understanding of the method of the present invention and its core ideas. It should be noted that it would be obvious to those skilled in the art that various improvements and modifications can be made to the present application without departing from the principles of the present invention, and such improvements and modifications fall within the scope of the claims of the present application.
Claims (9)
1. A pedestrian re-recognition method, characterized by comprising:
pre-constructing an expert network and a pedestrian re-recognition initial model, and performing knowledge supervision on a classification label and an embedding layer of the pedestrian re-recognition initial model by utilizing the expert network so as to transfer knowledge obtained by the expert network to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model;
inputting an image to be identified into the pedestrian re-identification model to obtain embedded features of the image to be identified;
Comparing the embedded features of the image to be identified with the embedded features of each search image in the database to obtain an identification result of the image to be identified;
the step of performing knowledge supervision on the classification labels and the embedded layers of the pedestrian re-recognition initial model by using the expert network comprises the following steps:
obtaining corresponding classification characteristics according to the embedded characteristics of the expert network and the pedestrian re-recognition initial model;
processing each classification characteristic by using a classification function to obtain corresponding classification probability;
the classification probability of the pedestrian re-recognition initial model and the loss function of each branch of the expert network are calculated in sequence to complete knowledge supervision of the classification labels of the pedestrian re-recognition initial model;
and carrying out knowledge supervision on the embedded layer of the pedestrian re-recognition initial model according to the balance coefficient of the embedded layer characteristics of the expert network and the pedestrian re-recognition initial model, the embedded characteristics of the expert network and the embedded characteristics of the pedestrian re-recognition initial model.
2. The pedestrian re-recognition method according to claim 1, wherein after the constructing of the expert network and the initial model for pedestrian re-recognition, further comprising:
Obtaining an output characteristic diagram of the expert network;
partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams;
performing downsampling operation on each sub-output feature map to obtain local features of the input image of the expert network;
carrying out maximum pooling treatment on the output feature map to obtain global features of the input image;
and taking the local features and the global features as embedded features of the expert network.
3. The pedestrian re-recognition method according to claim 2, wherein the constructing the expert network and the pedestrian re-recognition initial model includes:
and constructing a pedestrian re-identification initial model by adopting a lightweight neural network structure.
4. A pedestrian re-recognition method according to any one of claims 1 to 3, wherein after constructing the expert network and pedestrian re-recognition initial model, comprising:
determining training loss functions of the expert network and the pedestrian re-recognition initial model according to the classification characteristics, the classification labels and the embedded characteristics of the expert network and the pedestrian re-recognition initial model; the training loss functionLThe method comprises the following steps:
L=L p +L c +L e ；
Wherein,,L p =L(p r ，y hard )=cross_entropy(softmax(c r )，y hard )，L p a classification loss function of an initial model is re-identified for the pedestrian,p r the classification probability of the initial model is re-identified for the pedestrian,y hard for the real label of the input image,c r identifying classification features of the initial model again for the pedestrian,cross_entropyrepresenting a cross entropy loss function;L c as a function of the tag loss,，p(b) Representing the first of the expert networkbThe number of sorting heads is one,Nfor the total number of classification heads of the expert network,p(l) Watch (watch)The classification head output of the pedestrian re-recognition initial model is shown,Cis thatp(b) Andp(l) Is used for the vector dimension of (a),crepresenting the first of the corresponding vectorscThe number of elements to be added to the composition,Na total number of classification heads for the expert network;L e for embedding loss function->；e(k) For the embedded features of the expert network,e(l) The embedded features of the initial model are re-identified for the pedestrian,α k the equilibrium coefficients of the embedded layer features of the initial model are re-identified for the expert network and the pedestrian,kis the first of the expert networkkThe number of branches is chosen such that,Kis the total number of branches of the expert network.
5. The pedestrian re-recognition method of claim 4, wherein before the determining training loss functions of the expert network and the pedestrian re-recognition initial model from the classification features, classification labels, and embedded features of the expert network and the pedestrian re-recognition initial model, further comprises:
Calling a regression network relation to calculate the balance coefficient, wherein the regression network relation is thatα k =ReLU(W k [e(k)，e(l)])；
Where, reLU represents a linear rectification function,W k [e(k)，e(l)]representing that the full connectivity layer will be the expert networkkEmbedding features for individual branchese(k) Embedding featurese(l) And performing characteristic splicing.
6. A pedestrian re-recognition device, characterized by comprising:
the model training module is used for pre-constructing an expert network and a pedestrian re-recognition initial model, and performing knowledge supervision on a classification label and an embedding layer of the pedestrian re-recognition initial model by utilizing the expert network so as to transfer knowledge obtained by the expert network to the pedestrian re-recognition initial model to obtain a trained pedestrian re-recognition initial model; deleting the classification layer of the trained pedestrian re-recognition initial model to obtain a pedestrian re-recognition model;
the feature extraction module is used for inputting the image to be identified into the pedestrian re-identification model to obtain the embedded features of the image to be identified;
the image recognition module is used for obtaining a recognition result of the image to be recognized by comparing the embedded features of the image to be recognized with the embedded features of each search image in the database;
wherein the model training module is further to:
Obtaining corresponding classification characteristics according to the embedded characteristics of the expert network and the pedestrian re-recognition initial model; processing each classification characteristic by using a classification function to obtain corresponding classification probability; the classification probability of the pedestrian re-recognition initial model and the loss function of each branch of the expert network are calculated in sequence to complete knowledge supervision of the classification labels of the pedestrian re-recognition initial model; and carrying out knowledge supervision on the embedded layer of the pedestrian re-recognition initial model according to the balance coefficient of the embedded layer characteristics of the expert network and the pedestrian re-recognition initial model, the embedded characteristics of the expert network and the embedded characteristics of the pedestrian re-recognition initial model.
7. The pedestrian re-recognition device of claim 6, wherein the model training module includes a feature output unit configured to obtain an output feature map of the expert network; partitioning the output characteristic diagram according to preset proportion parameters to obtain a plurality of sub-output characteristic diagrams; performing downsampling operation on each sub-output feature map to obtain local features of the input image of the expert network; carrying out maximum pooling treatment on the output feature map to obtain global features of the input image; and taking the local features and the global features as embedded features of the expert network.
8. An electronic device comprising a processor and a memory, the processor being configured to implement the steps of the pedestrian re-identification method of any one of claims 1 to 5 when executing a computer program stored in the memory.
9. A readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the pedestrian re-recognition method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110864392.6A CN113780066B (en) | 2021-07-29 | 2021-07-29 | Pedestrian re-recognition method and device, electronic equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110864392.6A CN113780066B (en) | 2021-07-29 | 2021-07-29 | Pedestrian re-recognition method and device, electronic equipment and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113780066A CN113780066A (en) | 2021-12-10 |
| CN113780066B true CN113780066B (en) | 2023-07-25 |
Family
ID=78836489
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110864392.6A Active CN113780066B (en) | 2021-07-29 | 2021-07-29 | Pedestrian re-recognition method and device, electronic equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113780066B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114821651B (en) * | 2022-04-30 | 2023-11-10 | 苏州浪潮智能科技有限公司 | Pedestrian re-recognition method, system, equipment and computer readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111061881A (en) * | 2019-12-27 | 2020-04-24 | 浪潮通用软件有限公司 | Text classification method, equipment and storage medium |
| CN111178458A (en) * | 2020-04-10 | 2020-05-19 | 支付宝(杭州)信息技术有限公司 | Training of classification model, object classification method and device |
| CN111860147A (en) * | 2020-06-11 | 2020-10-30 | 北京市威富安防科技有限公司 | Pedestrian re-identification model optimization processing method and device and computer equipment |
| CN111885000A (en) * | 2020-06-22 | 2020-11-03 | 网宿科技股份有限公司 | Network attack detection method, system and device based on graph neural network |
| CN112541448A (en) * | 2020-12-18 | 2021-03-23 | 济南博观智能科技有限公司 | Pedestrian re-identification method and device, electronic equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108399381B (en) * | 2018-02-12 | 2020-10-30 | 北京市商汤科技开发有限公司 | Pedestrian re-identification method and device, electronic equipment and storage medium |
-
2021
- 2021-07-29 CN CN202110864392.6A patent/CN113780066B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111061881A (en) * | 2019-12-27 | 2020-04-24 | 浪潮通用软件有限公司 | Text classification method, equipment and storage medium |
| CN111178458A (en) * | 2020-04-10 | 2020-05-19 | 支付宝(杭州)信息技术有限公司 | Training of classification model, object classification method and device |
| CN111860147A (en) * | 2020-06-11 | 2020-10-30 | 北京市威富安防科技有限公司 | Pedestrian re-identification model optimization processing method and device and computer equipment |
| CN111885000A (en) * | 2020-06-22 | 2020-11-03 | 网宿科技股份有限公司 | Network attack detection method, system and device based on graph neural network |
| CN112541448A (en) * | 2020-12-18 | 2021-03-23 | 济南博观智能科技有限公司 | Pedestrian re-identification method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113780066A (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948425B (en) | Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching | |
| CN111797893B (en) | Neural network training method, image classification system and related equipment | |
| Yi et al. | ASSD: Attentive single shot multibox detector | |
| CN109657533A (en) | Pedestrian recognition methods and Related product again | |
| CN112434721A (en) | Image classification method, system, storage medium and terminal based on small sample learning | |
| CN111507217A (en) | Pedestrian re-identification method based on local resolution feature fusion | |
| CN108304835A (en) | character detecting method and device | |
| CN112036322A (en) | Method, system and device for constructing cross-domain pedestrian re-identification model of multi-task network | |
| CN108399386A (en) | Information extracting method in pie chart and device | |
| CN109509222B (en) | Method and device for detecting linear object | |
| Xia et al. | An evaluation of deep learning in loop closure detection for visual SLAM | |
| CN104899579A (en) | Face recognition method and face recognition device | |
| CN113807399A (en) | Neural network training method, neural network detection method and neural network detection device | |
| CN109670517A (en) | Object detection method, device, electronic equipment and target detection model | |
| CN110472495A (en) | A kind of deep learning face identification method based on graphical inference global characteristics | |
| CN108197669A (en) | The feature training method and device of convolutional neural networks | |
| CN113762326A (en) | Data identification method, device and equipment and readable storage medium | |
| CN113515669A (en) | Data processing method based on artificial intelligence and related equipment | |
| CN114937285B (en) | Dynamic gesture recognition method, device, equipment and storage medium | |
| CN111126515A (en) | Model training method based on artificial intelligence and related device | |
| CN113780066B (en) | Pedestrian re-recognition method and device, electronic equipment and readable storage medium | |
| CN115115981A (en) | Data processing method, device, equipment, storage medium and computer program product | |
| CN113255604B (en) | Pedestrian re-identification method, device, equipment and medium based on deep learning network | |
| CN113255701B (en) | Small sample learning method and system based on absolute-relative learning framework | |
| CN109523570A (en) | Beginning parameter transform model method and device |
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 |