TWI769635B - Network training pedestrian re-identification method and storage medium - Google Patents

Abstract

The embodiments of the present disclosure relate to a network training, a pedestrian re-identification method and a storage medium. The cloud server includes a first pedestrian re-identification network, and the method includes: sending to multiple side-end servers corresponding to the first pedestrian re-identification network The first network parameter; receiving the second network parameters returned by multiple side-end servers, the second network parameter is that each side-end server re-weights the second pedestrians included in itself according to the local image data set, the identity classification network and the first network parameter The recognition network is obtained by training; according to the second network parameters returned by multiple side-end servers, the first pedestrian re-recognition network is updated to obtain the updated first pedestrian re-recognition network.

Description

Network training, pedestrian re-identification method, electronic device and computer-readable storage medium

The present invention relates to the field of computer technology, in particular to a network training, a pedestrian re-identification method, an electronic device and a computer-readable storage medium.

Pedestrian re-identification, also known as pedestrian re-identification, is a technology that uses computer vision technology to determine whether there is a specific pedestrian in an image or video sequence. At present, pedestrian re-identification technology has been widely used in many fields and industries, such as smart video detection, smart security, etc. Since person re-identification technology involves private data such as face, human body, and personal identity in the process of processing images or video frame sequences, there is an urgent need for a pedestrian re-identification method that can avoid the leakage of private data.

The embodiments of the present invention provide a network training, a pedestrian re-identification method, an electronic device, and a technical solution of a computer-readable storage medium.

According to an aspect of the embodiments of the present invention, a network training method is provided, the method is applied to a cloud server, and the cloud server includes a first pedestrian re-identification network, and the method includes: sending to a plurality of The side server sends first network parameters corresponding to the first pedestrian re-identification network; receives second network parameters returned by the plurality of side servers, wherein for any of the side servers The edge server includes a second person re-identification network, an identity classification network and a local image data set, and the second person re-identification network and the first person re-identification network have The same network structure, the second network parameter is that the edge server re-weights the second pedestrian according to the local image data set, the identity classification network and the first network parameter. obtained after the identification network is trained; according to the second network parameters returned by the multiple edge servers, the first pedestrian re-identification network is updated to obtain the updated first pedestrian re-identification network. Pedestrian re-identification network.

In a possible implementation manner, the first pedestrian re-identification network is updated according to the second network parameters returned by the plurality of edge servers to obtain the updated first pedestrian re-identification network. A pedestrian re-identification network includes: receiving weights corresponding to the second network parameters returned by the plurality of edge servers, wherein, for any of the edge servers, the second network The weights corresponding to the parameters are determined by the edge server according to the second pedestrian re-identification network before training and the second pedestrian re-identification network after training; according to the multiple edge servers The weights corresponding to the returned second network parameters are weighted and averaged on the second network parameters returned by the plurality of edge servers to obtain the updated first network parameters; according to the updated first network parameters; and updating the first pedestrian re-identification network to obtain the updated first pedestrian re-identification network.

In a possible implementation manner, the method further includes: sending a common image data set to the multiple edge servers; receiving pseudo tags returned by the multiple edge servers, wherein for any of the edge servers the side server, the pseudo label is generated by the side server according to the shared image data set and the trained second person re-identification network; according to the shared image data set and The pseudo labels returned by the plurality of side servers are used to train the updated first pedestrian re-identification network to obtain the trained first pedestrian re-identification network.

In a possible implementation manner, the updated first pedestrian re-identification network is trained according to the shared image data set and the pseudo labels returned by the plurality of side servers, to obtain The trained first pedestrian re-identification network includes: determining an average pseudo-label according to the pseudo-labels returned by the plurality of edge servers; according to the shared image data set and the average pseudo-label, The updated first pedestrian re-identification network is trained to obtain the trained first pedestrian re-identification network.

According to an aspect of the embodiments of the present invention, a network training method is provided. The method is applied to an edge server, and the edge server includes a second pedestrian re-identification network, an identity classification network, and a local map. Like a data set, the method includes: receiving a first network parameter corresponding to a first pedestrian re-identification network sent by a cloud server, wherein the first pedestrian re-identification network and the second pedestrian re-identification network The recognition network has the same network structure; according to the local image data set, the identity classification network and the first network parameters, the second pedestrian re-identification network is trained, and the trained The second pedestrian re-identification network of , wherein the second pedestrian re-identification network corresponds to a second network parameter; the second network parameter is sent to the cloud server.

In a possible implementation manner, the second pedestrian re-identification network is trained according to the local image data set, the identity classification network and the first network parameters, and the trained The second pedestrian re-identification network includes: training the second pedestrian re-identification network and the identity classification network according to the local image data set and the first network parameters, The trained second person re-identification network and the trained identity classification network are obtained.

In a possible implementation manner, the method further includes: storing the trained identity classification network in the edge server.

In a possible implementation manner, the local image data set includes image data corresponding to multiple identities; the dimension of the identity classification network is related to the number of the multiple identities.

In a possible implementation manner, the method further includes: receiving a shared image data set sent by the cloud server; according to the shared image data set and the trained second pedestrian re-identification network, Generate a pseudo-label; send the pseudo-label to the cloud server.

In a possible implementation manner, the method further includes: determining a first feature vector according to the second person re-identification network before training and the local image data set, and determining a first feature vector according to the second person re-identification network after training and the local image data set; The pedestrian re-identification network and the local image data set determine the second feature vector; determine the cosine distance between the first feature vector and the second feature vector; determine the first feature vector according to the cosine distance. The weight corresponding to the second network parameter; sending the weight corresponding to the second network parameter to the cloud server.

In a possible implementation manner, the edge server is an image acquisition device; the local image data set is acquired according to the image acquisition device.

In a possible implementation manner, the edge server is connected to at least one image acquisition device, and the edge server and the at least one image acquisition device are located in the same geographical area; the local image data The set is obtained by the edge server from the at least one image acquisition device.

According to an aspect of the embodiments of the present invention, there is provided a pedestrian re-identification method, comprising: performing pedestrian re-identification processing on at least one frame of an image to be identified obtained within a target geographical area through a target pedestrian re-identification network, and determining Pedestrian re-identification result; wherein, the target pedestrian re-identification network is obtained by training the above-mentioned network training method.

In a possible implementation manner, the target pedestrian re-identification network is an updated first-person-re-identification network or a trained first-person-re-identification network.

In a possible implementation manner, in the case that an edge server is included in the target geographical area, and the edge server includes a trained second pedestrian re-identification network, the target pedestrian re-identification network The recognition network is the second person re-identification network after training.

According to an aspect of the embodiments of the present invention, a network training device is provided, the network training device is applied to a cloud server, the cloud server includes a first pedestrian re-identification network, and the device includes: The sending part is configured to send the first network parameters corresponding to the first pedestrian re-identification network to the plurality of side servers; the receiving part is configured to receive the first network parameters returned by the plurality of side servers. Two network parameters, wherein, for any of the edge servers, the edge server includes a second pedestrian re-identification network, an identity classification network and a local image data set, and the second pedestrian re-identification network The identification network and the first pedestrian re-identification network have the same network structure, and the second network parameter is that the edge server classifies the network according to the local image data set and the identity. obtained after training the second pedestrian re-identification network with the first network parameters; the update part is configured to, according to the second network parameters returned by the plurality of edge servers, The first pedestrian re-identification network is updated to obtain the updated first pedestrian re-identification network.

According to an aspect of the embodiments of the present invention, a network training apparatus is provided, the apparatus is applied to a side server, and the side server includes a second pedestrian re-identification network, an identity classification network and a local map Like a data set, the device includes: a receiving part configured to receive a first network parameter corresponding to the first pedestrian re-identification network sent by the cloud server, wherein the first pedestrian-re-identification network and The second pedestrian re-identification network has the same network structure; the network training part is configured to, according to the local image data set, the identity classification network and the first network parameters, perform a The second pedestrian re-identification network is trained, and the trained second pedestrian re-identification network is obtained, wherein the second pedestrian re-identification network corresponds to the second network parameter; the sending part is configured to send The cloud server sends the second network parameter.

According to an aspect of the embodiments of the present invention, there is provided a pedestrian re-identification device, comprising: a pedestrian re-identification part configured to perform at least one frame of a to-be-identified image acquired within a target geographic area through a target pedestrian re-identification network. For example, the pedestrian re-identification process is performed, and the pedestrian re-identification result is determined; wherein, the target pedestrian re-identification network is trained by the above-mentioned network training method.

According to an aspect of the embodiments of the present invention, an electronic device is provided, including: a processor; a memory configured to store instructions executable by the processor; wherein the processor is configured to invoke the instructions stored in the memory , to perform the above method.

According to an aspect of the embodiments of the present invention, there is provided a computer-readable storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the above method is implemented.

According to an aspect of the embodiments of the present invention, a computer program is provided, including computer-readable code, when the computer-readable code is executed in an electronic device, the processor in the electronic device implements the above method when executed.

In the embodiment of the present invention, in the cloud server including the first person re-identification network, the first network parameters corresponding to the first person re-identification network are sent to a plurality of side servers, and the The second network parameters returned by the plurality of edge servers, wherein, for any edge server, the edge server includes a second pedestrian re-identification network having the same network structure as the first pedestrian re-identification network The network, the identity classification network and the local image data set, the second network parameter is that the edge server performs the second pedestrian re-identification network according to the local image data set, the identity classification network and the first network parameters. After the training, the first pedestrian re-identification network is updated according to the second network parameters returned by the multiple side servers, and the updated first pedestrian re-identification network is obtained. The cloud server cooperates with multiple side servers to train the pedestrian re-identification network. During the training process, the image data set is still stored in the side server and does not need to be uploaded to the cloud server, so that pedestrian re-identification can be effectively trained. The network also protects data privacy.

It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not limiting of embodiments of the present invention. Other features and aspects of embodiments of the present invention will become apparent from the following detailed description of exemplary embodiments with reference to the accompanying drawings.

Various exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numbers in the figures denote components that have the same or similar functions. While various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

The term "and/or" in this article is only a relationship to describe related objects, which means that there can be three relationships, for example, A and/or B, which can mean that A exists alone, A and B exist at the same time, and B exists alone. three conditions. In addition, the term "at least one" herein refers to any combination of any one of a plurality or at least two of a plurality, for example, including at least one of A, B, and C, and may mean including those composed of A, B, and C. Any one or more elements selected in the collection.

In addition, in order to better illustrate the embodiments of the present invention, numerous specific details are given in the following detailed description. It should be understood by those skilled in the art that the embodiments of the present invention can also be implemented without certain specific details. In some instances, methods, means, components and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the subject matter of the embodiments of the present invention.

FIG. 1 shows a flowchart of a network training method according to an embodiment of the present invention. The network training method can be executed by a cloud server, and the cloud server includes a first person re-identification network. In some possible implementations, the network training method can be implemented by the cloud server calling computer-readable instructions stored in the memory. As shown in Figure 1, the method may include: In step S11, a first network parameter corresponding to the first pedestrian re-identification network is sent to a plurality of edge servers. In step S12, the second network parameters returned by the plurality of edge servers are received, wherein, for any edge server, the edge server includes a second pedestrian re-identification network, an identity classification network and a local map Like the dataset, the second person re-identification network and the first person re-identification network have the same network structure. A network parameter obtained after training the second person re-identification network. In step S13, the first pedestrian re-identification network is updated according to the second network parameters returned by the plurality of edge servers to obtain the updated first pedestrian re-identification network.

The cloud server cooperates with multiple side servers to train the pedestrian re-identification network. During the training process, the image data set is still stored in the side server and does not need to be uploaded to the cloud server, so that pedestrian re-identification can be effectively trained. The network also protects data privacy. In addition, since there is no need to upload the image data set to the cloud server, communication bandwidth can be effectively saved.

When the cloud server cooperates with multiple side servers to train the pedestrian re-identification network, the cloud server can combine multiple side servers to perform network training based on the federated learning algorithm. For example, multiple communities jointly train a pedestrian re-recognition network, each community is set up with a side server, through the federated learning algorithm, the image data set (set in the community or nearby the image acquisition equipment collected Image datasets) are still stored in the community (local side server) and do not need to be uploaded to other communities (other side servers), thus protecting data privacy.

In practical applications, since the data amounts of the local image data sets in different edge servers are different, the data between different edge servers is heterogeneous. When the traditional federated learning algorithm uses multiple side servers for network training, the weights of the second network parameters obtained by network training in the side servers are set according to the amount of data in different side servers. However, since the amount of data cannot directly reflect the training effect of network training, the cloud server uses the weight of the second network parameter obtained based on this weight determination method in the side server to give weight to the first pedestrian. Updating the identification network will result in a lower accuracy of the updated first pedestrian re-identification network.

In a possible implementation manner, the first pedestrian re-identification network is updated according to the second network parameters returned by multiple side servers, and the updated first pedestrian re-identification network is obtained, including: Receive weights corresponding to the second network parameters returned by multiple side servers, wherein, for any side server, the weights corresponding to the second network parameters are the weights corresponding to the side servers based on the second pedestrian re-identification network before training. It is determined by the road and the trained second pedestrian re-identification network; according to the weights corresponding to the second network parameters returned by the multiple side servers, the second network parameters returned by the multiple side servers are weighted After averaging, the updated first network parameters are obtained; according to the updated first network parameters, the first pedestrian re-identification network is updated to obtain the updated first pedestrian re-identification network.

Since the weight of the second network parameter sent by the side server is determined by the side server according to the second pedestrian re-identification network before training and the second pedestrian re-identification network after training, that is to say, the first The weights of the second network parameters are determined according to the training effect of the side servers, so that the cloud server can use the weights corresponding to the second network parameters returned by each side server to the second network parameters returned by the multiple side servers. After the network parameters are weighted and averaged, the updated first network parameters with higher accuracy are obtained, and then the first pedestrian re-identification network is updated according to the updated first network parameters, which effectively improves the post-update. The accuracy of the first pedestrian re-identification network.

In practical applications, since the local image data sets in different side servers are collected under different scenes (lighting, angles), the data between different side servers is heterogeneous, which leads to The performance of the second pedestrian re-identification network trained by the end server based on the local image data set, the identity classification network and the first network parameters is better than that obtained by the cloud server combined with multiple end servers. The updated first pedestrian re-identification network. Therefore, based on the knowledge distillation algorithm, the second pedestrian re-identification network after training in each side server can be used as the teacher network, and the updated first pedestrian re-identification network in the cloud server can be used as the student network. Road, use the teacher network to train the student network (use the updated second person re-identification network to train the updated first-person re-identification network) to improve the first-person re-identification network Stability and convergence of the training process.

In a possible implementation manner, the method further includes: sending a common image data set to multiple edge servers; receiving pseudo tags returned by multiple edge servers, wherein, for any edge server, the pseudo tag It is generated by the side server according to the shared image data set and the trained second pedestrian re-identification network; according to the shared image data set and the pseudo labels returned by multiple side servers, the updated first line The person re-identification network is trained, and the first person re-identification network after training is obtained.

The cloud server receives the pseudo-label returned by each side server. Since the pseudo-label is generated by the side-end server according to the shared image data set and the trained second person re-identification network, the pseudo-label can be used to represent The network characteristics of the trained second person re-identification network. Therefore, the updated first person re-identification network is trained according to the shared image data set and the pseudo-labels returned by multiple side servers. It is equivalent to synthesizing the network characteristics of each side server to train the updated first person re-identification network, which can effectively improve the stability and convergence of the first person re-identification network training process. Among them, the shared image data set refers to the image data set that both the cloud server and each side server can use for network training.

，其中，

為多個邊端伺服器的個數，

。雲端伺服器中更新後的第一行人重識別網路構成學生網路。教師網路1利用共用圖像資料集生成偽標籤

，以及將偽標籤

發送至雲端伺服器；將教師網路2利用共用圖像資料集生成偽標籤

，以及將偽標籤

發送至雲端伺服器；……；教師網路

利用共用圖像資料集生成偽標籤

，以及將偽標籤

發送至雲端伺服器。雲端伺服器根據偽標籤

、偽標籤

、……、偽標籤

、以及共用圖像資料集，對學生網路（更新後的第一行人重識別網路）進行訓練，得到訓練後的第一行人重識別網路。FIG. 2 shows a structural diagram of an exemplary network training provided according to an embodiment of the present invention. As shown in Figure 2, the second pedestrian re-identification network after training in multiple side servers constitutes teacher network 1, teacher network 2, ..., teacher network

,in,

is the number of multiple edge servers,

. The updated first pedestrian re-identification network in the cloud server constitutes the student network. Teacher Network 1 Generates Pseudo-Labels Using Shared Image Datasets

, and the pseudo-label

Send to cloud server; use teacher network 2 to generate pseudo-labels using shared image data set

, and the pseudo-label

Send to cloud server;...;Teacher network

Generating pseudo-labels from a shared image dataset

, and the pseudo-label

Sent to the cloud server. Cloud server based on pseudo tag

, pseudo tags

, ... , Pseudo tags

, and the shared image data set, train the student network (the updated first person re-identification network), and obtain the first-person re-identification network after training.

In a possible implementation, the updated first pedestrian re-identification network is trained according to the shared image data set and the pseudo-labels returned by multiple side servers, and the trained first pedestrian re-identification network is obtained. Identifying the network includes: determining the average pseudo-label according to the pseudo-labels returned by multiple side servers; training the updated first pedestrian re-identification network according to the shared image data set and the average pseudo-label, and obtaining The first person re-identification network after training.

個邊端伺服器返回的偽標籤

，其中，偽標籤

是第

個邊端伺服器根據共用圖像資料集以及第

個邊端伺服器中訓練後的第二行人重識別網路生成的。雲伺服器根據

個邊端伺服器返回的偽標籤，通過下述公式（1）確定平均偽標籤

：

（1）， 進而雲端伺服器根據共用圖像資料集和平均偽標籤

，對更新後的第一行人重識別網路進行訓練，得到訓練後的第一行人重識別網路。For example, the cloud server cooperates with N (N>1) edge servers to train the first person re-identification network, and the cloud server receives the first pedestrian re-identification network.

Pseudo tags returned by a side server

, where the pseudo-label

is the first

each side server based on the shared image data set and the

It is generated by the second person re-identification network after training in the end server. cloud server according to

For the pseudo-tags returned by each edge server, the average pseudo-tags are determined by the following formula (1)

:

(1), and then the cloud server based on the shared image data set and the average pseudo-label

, train the updated first pedestrian re-identification network to obtain the trained first pedestrian re-identification network.

In the embodiment of the present invention, in the cloud server including the first person re-identification network, the first network parameters corresponding to the first person re-identification network are sent to a plurality of side servers, and the The second network parameters returned by the plurality of edge servers, wherein, for any edge server, the edge server includes a second pedestrian re-identification network having the same network structure as the first pedestrian re-identification network The network, the identity classification network and the local image data set, the second network parameter is that the edge server performs the second pedestrian re-identification network according to the local image data set, the identity classification network and the first network parameters. After the training, the first pedestrian re-identification network is updated according to the second network parameters returned by the multiple side servers, and the updated first pedestrian re-identification network is obtained. The cloud server cooperates with multiple side servers to train the pedestrian re-identification network. During the training process, the image data set is still stored in the side server and does not need to be uploaded to the cloud server, so that pedestrian re-identification can be effectively trained. The network also protects data privacy.

FIG. 3 shows a flowchart of a network training method according to an embodiment of the present invention. The network training method can be executed by a side server, and the side server includes a second pedestrian re-identification network, an identity classification network and a local image data set. In some possible implementations, the network training method can be implemented by the side server calling computer-readable instructions stored in the memory. As shown in Figure 3, the method may include: In step S31, a first network parameter corresponding to the first pedestrian re-identification network sent by the cloud server is received, wherein the first pedestrian-re-identification network and the second pedestrian-re-identification network have the same network structure. In step S32, according to the local image data set, the identity classification network and the first network parameters, the second pedestrian re-identification network is trained to obtain a trained second pedestrian re-identification network, wherein the second pedestrian re-identification network is obtained. The pedestrian re-identification network corresponds to the second network parameter. In step S33, the second network parameter is sent to the cloud server.

The side server and the cloud server are used to train the pedestrian re-identification network. During the training process, the image data set is still stored in the side server and does not need to be uploaded to the cloud server, so that the pedestrian re-identification network can be effectively trained. while protecting data privacy.

In a possible implementation manner, the side server is an image acquisition device; the local image data set is acquired according to the image acquisition device.

In the case where the side server is an image capture device (for example, a smart camera) that communicates directly with the cloud server, the image capture device needs to have certain computing power, storage capability, and communication capability. The image acquisition device collects images to obtain a local image data set, and periodically deletes and clears invalid image data in the local image data set (for example, image data whose cache duration exceeds a preset threshold) to reduce storage pressure. The image acquisition device receives the first network parameters corresponding to the first pedestrian re-identification network sent by the cloud server, and trains the second pedestrian re-identification network according to the local image data set and the first network parameters. The trained second pedestrian re-identification network corresponding to the second network parameter is obtained, and the second network parameter is sent to the cloud server.

In a possible implementation manner, the edge server is connected to at least one image acquisition device, and the edge server and the at least one image acquisition device are located in the same geographical area; the local image data set is the edge server from at least one image acquisition device. obtained from an image acquisition device.

In the case where at least one image acquisition device is provided within the same geographical area, an edge server may be provided within the geographical area, and in this case, the at least one image acquisition device does not need to have storage capability and computing power. The side server is connected with each image acquisition device, and further acquires images from each image acquisition device to construct a local image data set. The side server receives the first network parameters corresponding to the first pedestrian re-identification network sent by the cloud server, and trains the second pedestrian re-identification network according to the local image data set and the first network parameters. The trained second pedestrian re-identification network corresponding to the second network parameter is obtained, and the second network parameter is sent to the cloud server.

In a possible implementation manner, the second person re-identification network is trained according to the local image data set, the identity classification network and the first network parameters, and the trained second person re-identification network is obtained, including : According to the local image data set and the first network parameters, the second pedestrian re-identification network and the identity classification network are trained, and the trained second pedestrian re-identification network and the trained identity classification network are obtained.

In a possible implementation manner, the local image data set includes image data corresponding to multiple identities; the dimension of the identity classification network is related to the number of multiple identities.

Since the trained pedestrian re-identification network is a network for identifying images, in the process of training the pedestrian re-identification network, it is necessary to use a local map including image data corresponding to multiple identities Like datasets, and identity classification networks, the dimensions of identity classification networks are related to the number of multiple identities included in the local image dataset. For example, if the local image data set includes image data corresponding to 100 identities, the dimension of the identity classification network is 100. That is, the identity classification network includes 100 different identity classes.

The edge server constructs the local second person re-identification network and the identity classification network as a combined network, and uses the first network parameters received from the cloud server and the local image data set to perform the combined network. After training, the combined network after training is obtained, wherein the combined network after training includes the second pedestrian re-identification network after training and the identity classification network after training, and the second pedestrian re-identification network after training. Corresponding to the second network parameter. Then, the edge server sends the second network parameter to the cloud server. Since the first person re-identification network and the second person-re-identification network have the same network structure, the first person-re-identification network can be updated by using the second network parameters.

In a possible implementation manner, the method further includes: storing the trained identity classification network in the edge server.

Since the first pedestrian re-identification network trained in the cloud server does not need to use the classifier network in the actual pedestrian re-identification process, in order to save the communication bandwidth and ensure the performance based on the federated learning algorithm During the joint training process, the network structure in the cloud server and the side server is consistent. The side server only sends the second network parameters corresponding to the trained second person re-identification network to the cloud server, while Store the trained identity classification network locally on the edge server.

In a possible implementation manner, the method further includes: receiving a shared image data set sent by the cloud server; generating a pseudo-label according to the shared image data set and the trained second pedestrian re-identification network; The device sends a pseudo tag.

Still taking the above Figure 2 as an example, as shown in Figure 2, the edge server receives the shared image data set sent by the cloud server, and uses the shared image data set and the locally trained second person re-identification network to generate pseudo-tag, and then the side server sends the pseudo-tag to the cloud server, because the pseudo-tag can be used to represent the network characteristics of the second pedestrian re-identification network after training, so that the cloud server can identify the cloud based on the pseudo-tag. The updated first person re-identification network in the server is the trained first person re-identification network obtained after network training, so that the network performance of the trained first-person re-identification network is the same as that of the first person re-identification network. The second person re-identification network after training in the side server is closer, thereby effectively improving the stability and convergence of the training process of the first person re-identification network.

In a possible implementation manner, the method further includes: determining the first feature vector according to the second person re-identification network before training and the local image data set, and determining the first feature vector according to the second person re-identification network and local image data set after training Image data set, determine the second feature vector; determine the cosine distance between the first feature vector and the second feature vector; determine the weight corresponding to the second network parameter according to the cosine distance; send the second network parameter to the cloud server The weights corresponding to the parameters.

。邊端伺服器根據從雲端伺服器接收到的第一網路參數進行網路訓練得到訓練後的第二行人重識別網路。邊端伺服器根據訓練後的第二行人重識別網路和本地圖像資料集生成第二特徵向量

。邊端伺服器確定第一特徵向量

和第二特徵向量

之間的餘弦距離

。邊端伺服器根據餘弦距離

確定訓練後的第二行人重識別網路對應的第二網路參數的權重。餘弦距離

越大，表示本次網路訓練產生的變化越大，訓練效果較好，則分配較大的權重；餘弦距離

越小，表示本次網路訓練產生的變化越小，訓練效果較差，則分配較小的權重。進而邊端伺服器將確定好的第二網路參數對應的權重發送至雲端伺服器，由於第二網路參數的權重是根據邊端伺服器中的網路訓練效果確定的，使得雲端伺服器基於該權重更新第一行人重識別網路後，可以有效提高更新後的第一行人重識別網路的精度。FIG. 4 shows an exemplary schematic diagram of determining a weight of a second network parameter according to an embodiment of the present invention. As shown in Figure 4, the edge server generates the first feature vector according to the second pedestrian re-identification network before training and the local image data set

. The side server performs network training according to the first network parameters received from the cloud server to obtain a trained second pedestrian re-identification network. The side server generates the second feature vector according to the trained second person re-identification network and the local image data set

. The edge server determines the first eigenvector

and the second eigenvector

cosine distance between

. Edge server based on cosine distance

Determine the weight of the second network parameter corresponding to the trained second pedestrian re-identification network. cosine distance

The larger the value, the greater the change of the network training, the better the training effect, and the larger the weight is assigned; the cosine distance

The smaller the value, the smaller the change produced by this network training and the poorer the training effect, the smaller the weight will be assigned. Furthermore, the side server sends the determined weight corresponding to the second network parameter to the cloud server. Since the weight of the second network parameter is determined according to the network training effect in the side server, the cloud server After updating the first person re-identification network based on the weight, the accuracy of the updated first person re-identification network can be effectively improved.

和第二特徵向量

的可以為本地圖像資料集的全部，也可以為本地圖像資料集的部分，本發明對此不做具體限定。In a possible implementation, for determining the first feature vector

and the second eigenvector

It can be the whole of the local image data set, or it can be a part of the local image data set, which is not specifically limited in the present invention.

In the embodiment of the present invention, in the edge server including the second person re-identification network, the identity classification network and the local image data set, the corresponding the first network parameters, wherein the first pedestrian re-identification network and the second pedestrian re-identification network have the same network structure, and the network and the first network parameters are classified according to the local image data set, the identity, The second pedestrian re-identification network is trained, and after the trained second pedestrian re-identification network corresponding to the second network parameters is obtained, the second network parameters are sent to the cloud server. The side server and the cloud server are used to train the pedestrian re-identification network. During the training process, the image data set is still stored in the side server and does not need to be uploaded to the cloud server, so that the pedestrian re-identification network can be effectively trained. while protecting data privacy.

In a possible implementation, when the cloud server cooperates with multiple side servers to train the pedestrian re-identification network, the multiple side servers may all be image acquisition devices that communicate directly with the cloud server ( e.g. Smart Camera). FIG. 5 shows a network structure diagram of an exemplary cloud server-edge server provided according to an embodiment of the present invention. As shown in Figure 5, the five side servers connected to the cloud server are all image capture devices (image capture device 1, image capture device 2, image capture device 3, image capture device 4 and Fig. like acquisition device 5). Under this network structure, each image acquisition device needs to have certain computing power, storage capacity and communication capacity. Each image acquisition device collects images to obtain a local image data set, and periodically deletes and clears invalid image data in the local image data set (for example, image data whose cache duration exceeds a preset threshold) to reduce storage pressure. The cloud server cooperates with 5 image acquisition devices serving as side servers to train the pedestrian re-identification network. During the training process, the image data set is still saved locally on each image acquisition device, and does not need to be uploaded to the cloud server. The data privacy can be protected while effectively training the pedestrian re-identification network.

In a possible implementation manner, when the cloud server cooperates with multiple side servers to train the pedestrian re-identification network, the multiple side servers may all be side servers connected to at least one image acquisition device. , each side server is located in the same geographic area with at least one image acquisition device connected to it. FIG. 6 shows an exemplary network structure diagram of a cloud server-edge server-terminal device according to an embodiment of the present invention. As shown in Figure 6, the cloud server communicates directly with the side server A and the side server B. The side server A is connected to the terminal device 1 and the terminal device 2, and the terminal device 1 and the terminal device 2 are image capture devices (image capture device 1 and image capture device 2, for example, the image capture device is a camera) , edge server A, image acquisition device 1, and image acquisition device 2 are set in the same geographic area (for example, the same community, or the same company), and edge server A collects images from image acquisition device 1 and image acquisition device 2 respectively. The acquisition device 2 acquires images to construct a local image dataset. The side server B is connected to terminal equipment 3, terminal equipment 4 and terminal equipment 5, and terminal equipment 3, terminal equipment 4 and terminal equipment 5 are image acquisition equipment (image acquisition equipment 3, image acquisition equipment 4 and image acquisition equipment The acquisition device 5, for example, the image acquisition device is a camera), the edge server B, the image acquisition device 3, the image acquisition device 4 and the image acquisition device 5 are set in the same geographical area (for example, the same community, or the same company), the side server B obtains images from the image capture device 3, the image capture device 4, and the image capture device 5 respectively to construct a local image data set. The cloud server cooperates with 2 side servers (side server A and side server B) to train the pedestrian re-identification network. During the training process, the image data set is still saved locally on each side server, without the need for Uploaded to the cloud server, which can effectively train the pedestrian re-identification network while protecting data privacy.

According to the above discussion, the present invention proposes two training architectures combining federated learning and person re-identification: cloud-edge architecture and device-edge-cloud architecture.

Cloud-side architecture: The cloud server communicates directly with the smart camera, and the cloud coordinates multiple smart cameras to train at the same time. The smart camera caches pictures on the edge, and deletes and cleans them regularly to reduce the storage pressure on the edge server. And this architecture requires the smart camera to have certain computing power, storage and communication capabilities.

Cloud-side-end architecture: The edge gateway (that is, the above-mentioned side-end server) is connected to multiple smart cameras, the cloud server is connected to multiple edge gateways, and the pedestrian re-identification training image is transmitted from the smart camera to the edge gateway, and Cached in the edge gateway, the edge gateway and the cloud server are trained for federated learning. During this process, the data remains in the edge gateway, and data privacy can still be protected. Among them, typical application scenarios, such as multiple communities jointly training a pedestrian re-identification model, each community has an edge gateway connected to multiple smart cameras, through federated learning, the data is still retained in the community, not transmitted to other communities or cloud servers to protect data privacy.

In a possible implementation manner, when the cloud server cooperates with multiple side servers to train the pedestrian re-identification network, the multiple side servers may also be partly image acquisition devices that communicate directly with the cloud server. (For example, a smart camera), part of which is an edge server connected to at least one image capture device, which is not specifically limited in this embodiment of the present invention.

FIG. 7 shows a structural diagram of an exemplary network training provided according to an embodiment of the present invention. As shown in Fig. 7, the cloud server can communicate with multiple side servers (side server 1, side server 2, ..., side server N), and the first server included in the cloud server The person re-identification network and the second person re-identification network included in the side server have the same network structure. Each edge server also includes a local image data set and an identity classification network. The cloud server sends the first network parameters corresponding to the first pedestrian re-identification network to the multiple side servers. After receiving the first network parameters, each side server uses the local image data set and identity classification The network trains the second person re-identification network to obtain a trained second person re-identification network corresponding to the parameters of the second network, and a trained identity classification network. In order to ensure that the network structure jointly trained by the cloud server and each side server is consistent, each side server only sends the second network parameters corresponding to the trained second person re-identification network to the cloud server. The cloud server updates the first pedestrian re-identification network according to the received second network parameters returned by the plurality of side servers, and obtains the updated first pedestrian re-identification network. Then, the first network parameters corresponding to the updated first pedestrian re-identification network are sent to multiple side servers for loop training until the updated first pedestrian re-identification network is identified in the cloud server. When the accuracy reaches the threshold, or the number of loop training reaches the preset number of times, the training ends.

The general federated learning algorithm (Federated Averaging, FedAvg) requires that the multi-party models that are synchronized (the deep learning model for person re-identification, that is, the above-mentioned person re-identification network) must be exactly the same. The classifier layer of the deep learning model for person re-identification (that is, the above-mentioned identity classification network) depends on how many different pedestrians each party's data contains, so the classifier layers of the multi-party models participating in training may be different, resulting in participation in federated learning. The multi-party models may be different, so the federated learning algorithm FedAvg is not applicable in the above application scenarios. According to the above content, because the federated learning algorithm is improved in the present invention, that is, the models of multiple parties participating in the federated learning are partially different, so the federated learning can be better applied to the training of person re-identification.

Since the data amount of the local image data sets in different edge servers is different, the data between different edge servers is heterogeneous. When combining multiple side servers to train the first person re-identification network, in order to reduce the influence of data heterogeneity on the accuracy of the updated first person re-identification network, each side server The weight determination method based on the training effect can be used to determine the weight of the second network parameter corresponding to the trained second pedestrian re-identification network, so that the second network parameter returned by the cloud server in conjunction with each side server After updating the first pedestrian re-identification network, an updated first-person re-identification network with higher accuracy is obtained. The specific steps of the weight determination method based on the training effect are as described in the relevant parts of the foregoing embodiments, and are not repeated here.

Since the local image data sets in different side servers are collected under different scenes (lighting, angles), the data between different side servers is heterogeneous, which leads to The performance of the trained second person re-identification network obtained by training the local image data set and the first network parameters is better than the updated first person re-identification network trained by the cloud server combined with multiple side servers. Identify the network. In order to improve the stability and convergence of the first person re-identification network training process, a knowledge distillation algorithm can be used, based on the updated second person re-identification network and shared image data set in each side server, The updated first person re-identification network in the cloud server is trained, thereby effectively improving the stability and convergence of the training process of the first person re-identification network. The specific training process based on the knowledge distillation algorithm is described in the relevant part of the above-mentioned embodiment, which is not repeated here.

Based on the above content, it can be seen that the present invention proposes a method of using knowledge distillation. The local model of multiple parties participating in federated learning is regarded as the teacher model, and the model of the cloud server is regarded as the student model. The knowledge distillation method is used to better transfer the knowledge of the teacher model. to the student model, thereby improving the stability and convergence of model training.

In the process of training the pedestrian re-identification network based on the network structure shown in Figure 7, the weight determination method based on the training effect and the network training based on the knowledge distillation algorithm can be used separately or in combination. This embodiment of the invention does not specifically limit this.

In an application scenario, for example, when multiple companies or institutions want to jointly train the pedestrian re-identification network to improve the robustness of the trained pedestrian re-identification network, in order to avoid aggregating data from multiple parties into For the problem of data privacy leakage on the same server, the pedestrian re-identification network can be jointly trained based on the network structure shown in Figure 7. Or the organization communicates directly with the same cloud server, and the data is still stored locally during the training process, and there is no need to upload it to the cloud server, so that the pedestrian re-identification network can be obtained through effective training in the cloud server. Data privacy of a company or organization.

In an application scenario, for example, company A provides company B with a pedestrian re-identification network training service. If the image data of each image acquisition device (for example, a smart camera) of company B is uploaded to company A, There will be data privacy issues. At this time, company A can jointly train the pedestrian re-identification network based on the network structure shown in Figure 7, company A can be used as a cloud server, and each image acquisition device in company B can be used as multiple side servers During the training process, the data is still stored locally in company B and does not need to be uploaded to company A, so that the pedestrian re-identification network can be obtained through effective training in company B while protecting the privacy of company A’s data.

The embodiment of the present invention also provides a pedestrian re-identification method. The pedestrian re-identification method can be executed by a terminal device or other processing device, wherein the terminal device can be an image acquisition device (for example, a smart camera), a user equipment (User Equipment, UE), a mobile device, a user terminal, Terminals, cellular phones, wireless phones, Personal Digital Assistants (PDAs), handheld devices, computing devices, in-vehicle devices, wearable devices, etc. Other processing devices may be servers, cloud servers, or the like. In some possible implementations, the pedestrian re-identification method may be implemented by the processor calling computer-readable instructions stored in the memory. The method can include: The target pedestrian re-identification network is used to perform pedestrian re-identification processing on at least one frame of the image to be recognized obtained within the target geographical area, and the pedestrian re-identification result is determined; wherein, the target pedestrian re-identification network adopts the network of the foregoing embodiment. The road training method is trained.

The target pedestrian re-identification network can perform pedestrian re-identification processing on at least one frame of the to-be-identified image within the target geographic area, and determine whether there is a characteristic pedestrian in the at least one frame of the to-be-identified image.

In a possible implementation manner, the target person re-identification network is an updated first person re-identification network or a trained first-person re-identification network.

Since the updated first person re-identification network or the trained first person re-identification network in the cloud server is universal, that is, it can be applied to any application scenario, therefore, the updated first person re-identification network in the cloud server can be used. The first pedestrian re-identification network or the trained first pedestrian re-identification network can realize the pedestrian re-recognition processing of at least one frame of the image to be recognized obtained within the target geographic area, so as to obtain the pedestrian re-identification process. Identify the results.

In a possible implementation manner, in the case that a side server is included within the target geographical area, and the side server includes a trained second pedestrian re-identification network, the target pedestrian re-recognition network is the post-training network. The second pedestrian re-identification network.

Combining the above-mentioned embodiments of the network training method of the cloud server and the side server, it can be seen that since the local image data sets in different side servers are collected under different scenes (lighting, angles), different side servers are obtained. The data between the servers is heterogeneous. The trained second person re-identification network obtained by different side servers based on the local image data set is personalized and more suitable for the local scene, which leads to each side server The performance of the second person re-identification network trained in the server is better than that of the updated first person re-identification network trained by the cloud server combined with multiple side servers. Therefore, in the case where the end server is included in the target geographical area, and the second pedestrian re-recognition network after training is included in the end server, the trained re-identification network of the local scene that is more suitable for the target geographical area can be used. The second pedestrian re-identification network performs pedestrian re-identification processing on at least one frame of the image to be identified, so as to improve the accuracy of the processing result.

In the present invention, after one deployment, further training and repeated operations can be performed according to the data generated by the side server, so that a low-cost model can be continuously updated and repeated operations can be achieved.

It can be understood that the foregoing method embodiments mentioned in the embodiments of the present invention can be combined with each other to form a combined embodiment without violating the principle and logic. Those skilled in the art can understand that, in the above method of the specific embodiment, the specific execution order of each step should be determined by its function and possible internal logic.

In addition, the embodiments of the present invention also provide network training/pedestrian re-identification devices, electronic devices, computer-readable storage media, and programs, all of which can be used to implement any of the network training/pedestrian re-identification methods provided by the embodiments of the present invention , the corresponding technical solutions and descriptions, and refer to the corresponding records in the method section, which will not be repeated.

FIG. 8 shows a block diagram of a network training apparatus according to an embodiment of the present invention. The network training device is applied to a cloud server, and the cloud server includes a first pedestrian re-identification network. As shown in FIG. 8, the apparatus 80 includes: The sending part 81 is configured to send the first network parameters corresponding to the first pedestrian re-identification network to the plurality of side servers; The receiving part 82 is configured to receive the second network parameters returned by the plurality of edge servers, wherein, for any edge server, the edge server includes the second pedestrian re-identification network, the identity classification network and the Local image data set, the second pedestrian re-identification network and the first pedestrian re-identification network have the same network structure, and the second network parameter is that the edge server classifies the network according to the local image data set and identity. obtained after training the second pedestrian re-identification network with the first network parameters; The updating part 83 is configured to update the first pedestrian re-identification network according to the second network parameters returned by the plurality of edge servers, so as to obtain the updated first pedestrian-re-identification network.

In one possible implementation, the update section 83 includes: The receiving subsection is configured to receive the weights corresponding to the second network parameters returned by the plurality of side servers, wherein, for any side server, the weights corresponding to the second network parameters are the weights corresponding to the side servers according to the pre-training Determined by the second person re-identification network and the trained second person re-identification network; The first update subsection is configured to perform a weighted average on the second network parameters returned by the multiple edge servers according to the weights corresponding to the second network parameters returned by the multiple edge servers to obtain the updated No. a network parameter; The second update subsection is configured to update the first pedestrian re-identification network according to the updated first network parameters to obtain the updated first pedestrian re-identification network.

In a possible implementation manner, the sending part 81 is further configured to send a common image data set to a plurality of edge servers; The receiving part 82 is further configured to receive pseudo labels returned by a plurality of edge servers, wherein, for any edge server, the pseudo labels are the second pedestrian weights after training according to the shared image data set and the edge server. Identify network generated; Device 80, further comprising: The network training part is configured to train the updated first pedestrian re-identification network according to the shared image data set and the pseudo-labels returned by multiple side servers, and obtain the first pedestrian re-identification network after training. Identify the network.

In a possible implementation, the network training part is also configured as: Determine the average pseudo-tag according to the pseudo-tags returned by multiple edge servers; According to the shared image dataset and the average pseudo-label, the updated first person re-identification network is trained to obtain the trained first person re-identification network.

FIG. 9 shows a block diagram of a network training apparatus according to an embodiment of the present invention. The network training device is applied to a side server, and the side server includes a second pedestrian re-identification network, an identity classification network and a local image data set. As shown in FIG. 9, the apparatus 90 includes: The receiving part 91 is configured to receive the first network parameter corresponding to the first pedestrian re-identification network sent by the cloud server, wherein the first pedestrian-re-identification network and the second pedestrian-re-identification network have the same parameters. network structure; The network training part 92 is configured to train the second pedestrian re-identification network according to the local image data set, the identity classification network and the first network parameters to obtain the trained second pedestrian re-identification network, Wherein, the second pedestrian re-identification network corresponds to the second network parameter; The sending part 93 is configured to send the second network parameter to the cloud server.

In a possible implementation, the network training part 92 is further configured to: According to the local image data set and the first network parameters, the second person re-identification network and the identity classification network are trained, and the trained second person re-identification network and the trained identity classification network are obtained.

In a possible manner, the device 90 further includes: The storage part is configured to store the trained identity classification network in the edge server.

In a possible manner, the local image data set includes image data corresponding to multiple identities; the dimension of the identity classification network is related to the number of multiple identities.

In one possible way: The receiving part 91 is further configured to receive the common image data set sent by the cloud server; Device 90, further comprising: The pseudo-label generation part is configured to generate pseudo-labels according to the shared image data set and the trained second pedestrian re-identification network; The sending part 93 is further configured to send the pseudo tag to the cloud server.

In a possible manner, the device 90 further includes: The first determining part is configured to determine the first feature vector according to the second pedestrian re-identification network and the local image data set before training, and to determine the first feature vector according to the second pedestrian re-identification network and the local image data set after training, determine the second eigenvector; a second determining part configured to determine a cosine distance between the first feature vector and the second feature vector; The third determination part is configured to determine the weight corresponding to the second network parameter according to the cosine distance; The sending part 93 is further configured to send the weight corresponding to the second network parameter to the cloud server.

In a possible manner, the side server is an image acquisition device; the local image data set is acquired according to the image acquisition device.

In a possible manner, the edge server is connected with at least one image acquisition device, and the edge server and the at least one image acquisition device are located in the same geographical area; the local image data set is the edge server from at least one image acquisition device. obtained from the image acquisition device.

An embodiment of the present invention further provides a pedestrian re-identification device, comprising: a pedestrian re-identification part configured to perform pedestrian re-identification on at least one frame of an image to be identified obtained within a target geographical area through a target pedestrian re-identification network process, and determine the pedestrian re-identification result; wherein, the target pedestrian re-identification network is trained by the above-mentioned network training method.

In a possible implementation manner, the target person re-identification network is an updated first person re-identification network or a trained first-person re-identification network.

In a possible implementation manner, in the case that a side server is included within the target geographical area, and the side server includes a trained second pedestrian re-identification network, the target pedestrian re-recognition network is the post-training network. The second pedestrian re-identification network.

In some embodiments, the functions or included parts of the network training/person re-identification apparatus provided in the embodiments of the present invention may be configured to execute the methods described in the above method embodiments, and the specific implementation may be implemented with reference to the above methods For the sake of brevity, the description of the example will not be repeated here.

In the embodiments of the present invention and other embodiments, a "part" may be a part of a circuit, a part of a processor, a part of a program or software, etc., of course, a unit, a module or a non-modular form.

An embodiment of the present invention further provides a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the above-mentioned method is implemented. The computer-readable storage medium may be a non-volatile computer-readable storage medium.

An embodiment of the present invention further provides an electronic device, including: a processor; a memory configured to store instructions executable by the processor; wherein the processor is configured to call the instructions stored in the memory to execute the above method .

Embodiments of the present invention also provide a computer program product, including computer-readable codes. When the computer-readable codes are run on a device, a processor in the device executes the network training for implementing the network training provided in any of the above embodiments. /Directive for pedestrian re-identification methods.

Embodiments of the present invention also provide another computer program product configured to store computer-readable instructions, which, when executed, cause the computer to execute the operations of the network training/pedestrian re-identification method provided by any of the above embodiments.

The electronic device may be provided as a terminal, server or other form of device.

FIG. 10 shows a block diagram of an electronic device according to an embodiment of the present invention. As shown in FIG. 10, the electronic device 800 may be an image capture device (eg, a smart camera), a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal Terminals such as digital assistants.

10, an electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input/output (I/O) interface 812, a sensing server component 814, and communication component 816.

The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing component 802 can include one or more processors 820 to execute instructions to perform all or some of the steps of the methods described above. Additionally, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

The memory 804 is configured to store various types of data to support the operation of the electronic device 800 . Examples of such data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, pictures, videos, and the like. Memory 804 may be implemented by any type of volatile or non-volatile storage device or combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Magnetic Memory, Flash Memory, Disk or CD.

Power supply assembly 806 provides power to various components of electronic device 800 . Power supply components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800 .

Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touch, swipe, and gestures on the touch panel. The touch sensor may not only sense the boundaries of a touch or swipe action, but also detect the duration and pressure associated with the touch or swipe action. In some embodiments, the multimedia component 808 includes a front-facing camera and/or a rear-facing camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each of the front and rear cameras can be a fixed optical lens system or have focal length and optical zoom capability.

Audio component 810 is configured to output and/or input audio signals. For example, audio component 810 includes a microphone (MIC) that is configured to receive external audio signals when electronic device 800 is in operating modes, such as call mode, recording mode, and voice recognition mode. The received audio signal may be further stored in memory 804 or transmitted via communication component 816 . In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing element 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, and the like. These buttons may include, but are not limited to: home button, volume buttons, start button, and lock button.

Sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for electronic device 800 . For example, the sensor assembly 814 can detect the open/closed state of the electronic device 800, the relative positioning of the components, such as the display and keypad of the electronic device 800, the sensor assembly 814 can also detect the electronic device 800 or Changes in the position of a component of the electronic device 800 , presence or absence of user contact with the electronic device 800 , orientation or acceleration/deceleration of the electronic device 800 and changes in the temperature of the electronic device 800 . Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects in the absence of any physical contact. The sensor assembly 814 may also include a light sensor, such as a complementary metal oxide semiconductor (CMOS) or charge coupled device (CCD) image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices. The electronic device 800 may access a wireless network based on a communication standard, such as wireless network (WiFi), second generation mobile communication technology (2G) or third generation mobile communication technology (3G), or a combination thereof. In an exemplary embodiment, the communication component 816 receives broadcast signals or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, Infrared Data Association (IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (BT) technology and other technologies.

In an exemplary embodiment, electronic device 800 may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor or other electronic component implementation for performing the above method.

In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 804 including computer program instructions executable by the processor 820 of the electronic device 800 to accomplish the above method.

FIG. 11 shows a block diagram of an electronic device according to an embodiment of the present invention. As shown in FIG. 11, the electronic device 1900 may be provided as a server. 11, the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions executable by the processing component 1922, such as applications. An application program stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Additionally, the processing component 1922 is configured to execute instructions to perform the above-described methods.

The electronic device 1900 may also include a power supply assembly 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input output (I/O) Interface 1958. The electronic device 1900 can operate an operating system based on the memory 1932, such as Microsoft Server Operating System (Windows Server ^TM ), a graphical user interface based operating system (Mac OS X ^TM ) introduced by Apple Inc. Process Computer Operating System (Unix ^™ ), Free and Open Source Unix-like Operating System (Linux ^™ ), Open Source Unix-like Operating System (FreeBSD ^™ ) or the like.

In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions executable by the processing component 1922 of the electronic device 1900 to accomplish the above method.

Embodiments of the present invention may be systems, methods and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the embodiments of the present invention.

A computer-readable storage medium may be a tangible device that can hold and store instructions for use by the instruction execution device. The computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (non-exhaustive list) of computer readable storage media include: portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable Design read only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read only memory (CD-ROM), digital versatile disk (DVD), memory stick , a floppy disk, a mechanically encoded device, such as a punched card or a raised structure in a groove with instructions stored thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (eg, light pulses through fiber optic cables), or Electrical signals carried by wires.

The computer-readable program instructions described herein may be downloaded from computer-readable storage media to various computing/processing devices, or downloaded to external computers or external storage over a network, such as the Internet, a local area network, a wide area network, and/or a wireless network device. Networks may include copper transmission cables, fiber optic transmissions, wireless transmissions, routers, firewalls, switches, gateway computers and/or edge servers. A network interface card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for computer-readable storage stored in each computing/processing device in the medium.

Computer program instructions for carrying out the operations of embodiments of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, or in one or more programming languages Source or object code written in any combination of programming languages including object-oriented programming languages - such as Smalltalk, C++, etc., and conventional procedural programming languages - such as the "C" language or similar programming language. The computer readable program instructions 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 remotely. run on a client computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any kind of network—including a local area network (LAN) or a wide area network (WAN)—or, can be connected to an external computer (for example, using the Internet road service provider to connect via the Internet). In some embodiments, electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer readable program instructions. Electronic circuits may execute computer-readable program instructions to implement various aspects of the embodiments of the present invention.

Aspects of embodiments of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.

These computer readable program instructions may be provided to the processor of a general purpose computer, special purpose computer or other programmable data processing device to produce a machine for execution of the instructions by the processor of the computer or other programmable data processing device When, means are created that implement the functions/acts specified in one or more of the blocks in the flowchart and/or block diagrams. These computer readable program instructions may also be stored on a computer readable storage medium, the instructions causing the computer, programmable data processing device and/or other equipment to operate in a particular manner, so that the computer readable medium storing the instructions Included is an article of manufacture comprising instructions for implementing various aspects of the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.

Computer readable program instructions can also be loaded into a computer, other programmable data processing device, or other equipment, so that a series of operational steps are performed on the computer, other programmable data processing device, or other equipment to generate a computer Processes of implementation such that instructions executing on a computer, other programmable data processing apparatus, or other device implement the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.

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 invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions that contains one or more logic for implementing the specified logic Executable instructions for the function. In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks 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 is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by dedicated hardware-based systems that perform the specified functions or actions. implementation, or may be implemented in a combination of special purpose hardware and computer instructions.

The computer program product can be implemented by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), etc. Wait.

Various embodiments of the present invention have been described above, and the foregoing descriptions are exemplary, not exhaustive, and not limiting of the disclosed embodiments. Numerous modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or improvement over the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Industrial Applicability Embodiments of the present invention relate to a network training, pedestrian re-identification method and device, storage medium, and computer program. The cloud server includes a first pedestrian re-identification network, and the method includes: sending a message to a plurality of side servers. The first pedestrian re-identifies the first network parameters corresponding to the network; and receives the second network parameters returned by multiple side servers, wherein, for any side server, the side server includes the second pedestrian weight. Identification network, identity classification network and local image data set, the second pedestrian re-identification network and the first pedestrian re-identification network have the same network structure, and the second network parameter is the edge server according to the local The image data set, the identity classification network and the first network parameters are obtained after training the second pedestrian re-identification network; The identification network is updated, and the updated first pedestrian re-identification network is obtained. Since the cloud server cooperates with multiple side servers to train the pedestrian re-identification network, the image data set is still stored in the side server during the training process, and does not need to be uploaded to the cloud server, so that the pedestrian re-identification network can be effectively trained. Identify networks while protecting data privacy.

80: Network training device 81: Send part 82: Receive part 83: Update section 90: Network training device 91: Receive part 92: Network training part 93: Send part 800: Electronics 802: Process component 804: memory 806: Power Components 808: Multimedia Components 810: Audio Components 812: Input/Output Interface 814: Sensor Assembly 816: Communication Components 820: Processor 1900: Electronic equipment 1922: Electronic Devices 1926: Power Components 1932: Memory 1950: Web Interface 1958: Input and output interface S11~S13: Steps S31~S33: Steps

The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, and these drawings illustrate embodiments consistent with the present invention, and together with the description, serve to explain the technical solutions of the embodiments of the present invention. 1 shows a flowchart of a network training method according to an embodiment of the present invention; FIG. 2 shows a structural diagram of an exemplary network training provided according to an embodiment of the present invention; 3 shows a flowchart of a network training method according to an embodiment of the present invention; FIG. 4 shows an exemplary schematic diagram of determining a weight of a second network parameter according to an embodiment of the present invention; FIG. 5 shows a network structure diagram of an exemplary cloud server-side server provided according to an embodiment of the present invention; 6 shows an exemplary network structure diagram of a cloud server-side server-terminal device according to an embodiment of the present invention; FIG. 7 shows a structural diagram of an exemplary network training provided according to an embodiment of the present invention; 8 shows a block diagram of a network training apparatus according to an embodiment of the present invention; FIG. 9 shows a block diagram of a network training apparatus according to an embodiment of the present invention; 10 shows a block diagram of an electronic device according to an embodiment of the present invention; FIG. 11 shows a block diagram of an electronic device according to an embodiment of the present invention.

S11~S13: Steps

Claims (17)

A network training method, the method is applied to a cloud server, the cloud server includes a first pedestrian re-identification network, the method includes: sending the first pedestrian to a plurality of side servers Re-identifying the first network parameters corresponding to the network; receiving the second network parameters returned by the plurality of edge servers, wherein, for any of the edge servers, the edge server includes the first network parameter. Two person re-identification network, identity classification network and local image data set, the second person re-identification network and the first person re-identification network have the same network structure, the second network The road parameters are obtained by the side server after training the second pedestrian re-identification network according to the local image data set, the identity classification network and the first network parameters; weights corresponding to the second network parameters returned by the plurality of side servers; according to the weights corresponding to the second network parameters returned by the plurality of side servers; weighted average of the second network parameters returned by the controller to obtain the updated first network parameters; according to the updated first network parameters, perform the first pedestrian re-identification network update to obtain the updated first pedestrian re-identification network. The method according to claim 1, wherein, for any of the side servers, the weight corresponding to the second network parameter is the weight of the side server based on the second pedestrian re-identification network before training. road and after training The second pedestrian re-identification network is determined to be obtained. The method according to claim 1 or 2, further comprising: sending a common image data set to the multiple edge servers; receiving pseudo tags returned by the multiple edge servers, wherein for any one the side server, the pseudo label is generated by the side server according to the shared image data set and the trained second person re-identification network; according to the shared image data set and The pseudo labels returned by the plurality of side servers are used to train the updated first pedestrian re-identification network to obtain the trained first pedestrian re-identification network. The method according to claim 3, wherein the updated first pedestrian re-identification network is performed according to the shared image data set and the pseudo tags returned by the plurality of edge servers. training to obtain the trained first pedestrian re-identification network, including: determining an average pseudo-label according to the pseudo-labels returned by the multiple side servers; according to the shared image data set and the average Pseudo label, the updated first pedestrian re-identification network is trained to obtain the trained first pedestrian re-identification network. A network training method, wherein the method is applied to an edge server, and the edge server includes a second pedestrian re-identification network, an identity classification network and a local image data set, and the method includes: Receive the first network parameter corresponding to the first pedestrian re-identification network sent by the cloud server, wherein the first pedestrian re-identification network and the second line The person re-identification network has the same network structure; according to the local image data set, the identity classification network and the first network parameters, the second pedestrian re-identification network is trained to obtain The trained second pedestrian re-identification network, wherein the second pedestrian re-identification network corresponds to a second network parameter; sending the second network parameter to the cloud server; sending the second network parameter to the cloud The server sends the weight corresponding to the second network parameter; the weight corresponding to the second network parameter is used to perform a weighted average on the second network parameter to obtain the updated first network parameter; the updated The latter first network parameter is used to update the first pedestrian re-identification network to obtain the updated first pedestrian re-identification network. The method according to claim 5, wherein the second pedestrian re-identification network is trained according to the local image data set, the identity classification network and the first network parameters, Obtaining the trained second pedestrian re-identification network includes: re-identifying the second pedestrian and the identity classification network according to the local image data set and the first network parameters Perform training to obtain the trained second pedestrian re-identification network and the trained identity classification network. The method according to claim 6, further comprising: storing the trained identity classification network in the edge server. The method according to claim 6 or 7, wherein the local image data set includes image data corresponding to multiple identities; The dimension of the share classification network is related to the number of the plurality of identities. The method according to any one of claim 5 to 7, further comprising: receiving a shared image data set sent by the cloud server; Identify the network, generate a pseudo-label; send the pseudo-label to the cloud server. According to the method according to any one of request items 5 to 7, the sending the weight corresponding to the second network parameter to the cloud server includes: re-identifying the network and the network according to the second pedestrian before training. The local image data set determines a first feature vector, and determines a second feature vector according to the trained second pedestrian re-identification network and the local image data set; determines the first feature vector and the cosine distance between the second feature vectors; determine the weight corresponding to the second network parameter according to the cosine distance; send the weight corresponding to the second network parameter to the cloud server. The method according to any one of claims 5 to 7, wherein the edge server is an image acquisition device; and the local image data set is acquired according to the image acquisition device. The method according to any one of claims 5 to 7, wherein the side server is connected to at least one image acquisition device, the The edge server and the at least one image acquisition device are located in the same geographical area; the local image data set is obtained by the edge server from the at least one image acquisition device. A pedestrian re-identification method, comprising: performing pedestrian re-identification processing on at least one frame of an image to be identified obtained within a target geographical area through a target pedestrian re-identification network, and determining a pedestrian re-identification result; wherein the target pedestrian The re-identification network is trained by using the network training method described in any one of request items 1 to 12. The method according to claim 13, wherein the target pedestrian re-identification network is an updated first-person-re-identification network or a trained first-person-re-identification network. The method according to claim 13, wherein, in the case that an edge server is included within the target geographic area, and the edge server includes a trained second pedestrian re-identification network, the The target person re-identification network is the second person re-identification network after training. An electronic device, comprising: a processor; a memory configured to store instructions executable by the processor; wherein the processor is configured to invoke the instructions stored in the memory to execute any one of request items 1 to 15 method described in item. A computer-readable storage medium on which computer program instructions are stored, and when the computer program instructions are executed by a processor, implement the method described in any one of claim 1 to 15.

Images