View angle perception vehicle re-identification method and system based on multi-task learning

Abstract

The invention discloses a multi-task learning-based visual angle perception vehicle re-identification method and system, and relates to the technical field of intelligent traffic, and the method comprises the following steps: receiving a 3D target detection data set, inputting the LiDAR annotation in the 3D target detection data set into a pre-established visual angle predictor, and obtaining a trained visual angle predictor; generating a visual angle label of the ReID data set by using the trained visual angle predictor, inputting the visual angle label of the ReID data set into a pre-established multi-task feature extractor, and outputting to obtain a visual angle feature vector and an identity feature vector; and comprehensive loss function calculation is carried out by using the visual angle feature vector and the identity feature vector to obtain a total loss function, so that the vehicle re-identification performance can be improved, and the identity and visual angle features can be decoupled by using a network, so that the vehicle re-identification task performance is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a perspective-aware vehicle re-identification method and system based on multi-task learning. Background Art

[0002] Vehicle Re-Identification (V-ReID) aims to match vehicle images across cameras and is an important research topic in intelligent transportation systems. However, due to the rapid changes in shooting angles of different cameras, the appearance of the vehicle will change significantly, which poses a huge challenge to re-identification. Summary of the invention

[0003] In order to address the deficiencies mentioned in the above background technology, the purpose of the present invention is to provide a perspective-aware vehicle re-identification method and system based on multi-task learning, which can utilize the network to decouple identity and perspective features, thereby improving the performance of vehicle re-identification tasks.

[0004] In a first aspect, the objective of the present invention can be achieved by the following technical solution: a perspective-aware vehicle re-identification method based on multi-task learning, the method comprising the following steps:

[0005] receiving a 3D object detection dataset, and inputting LiDAR annotations in the 3D object detection dataset into a pre-established viewpoint predictor to obtain a trained viewpoint predictor;

[0006] The trained viewpoint predictor is used to generate the viewpoint labels of the ReID dataset, and the viewpoint labels of the ReID dataset are input into the pre-established multi-task feature extractor, and the viewpoint feature vector and the corresponding predicted viewpoint and identity feature vector and the corresponding identity ID are output;

[0007] The comprehensive loss function is calculated using the view feature vector and the identity feature vector to obtain the total loss function, which can improve the performance of vehicle re-identification.

[0008] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the 3D target detection dataset crops the target image according to the 2D bounding box, filters out images smaller than preset pixels, and then uses the ESRGAN method for super-resolution processing.

[0009] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the pre-established view predictor uses ResNet-101 as a backbone network, and the original classification layer is replaced by a view prediction layer.

[0010] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the loss function L of the pre-established view predictorpredictor as follows:

[0011]

[0012] Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value.

[0013] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the pre-established multi-task feature extractor adopts ResNet50-IBN-a as the backbone network, and combines the SE channel attention module to enhance the feature representation capability.

[0014] In combination with the first aspect, in certain implementations of the first aspect, the method also includes: the comprehensive loss function includes cross entropy loss for identity classification, cross entropy loss for perspective prediction, triple loss, center loss and orthogonal loss.

[0015] In combination with the first aspect, in some implementations of the first aspect, the method further includes: a calculation formula for the comprehensive loss function L:

[0016] L=L id +λ1L viewpoint +λ2L triplet +λ3L center +λ4L orth ,

[0017] Among them, λ1, λ2, λ3 and λ4 are hyperparameters that control the relative importance of each loss term, L id is the cross entropy loss for identity classification, L viewpoint is the cross entropy loss of view prediction, L triplet is the triplet loss, L center is the center loss, L orth is the orthogonal loss.

[0018] In combination with the first aspect, in some implementations of the first aspect, the method further includes: the orthogonal loss:

[0019]

[0020] where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation of |B| is the number of samples in the batch.

[0021] In a second aspect, in order to achieve the above-mentioned object, the present invention discloses a perspective-aware vehicle re-identification system based on multi-task learning, comprising:

[0022] A model training module is used to receive a 3D object detection dataset, input the LiDAR annotations in the 3D object detection dataset into a pre-established viewpoint predictor, and obtain a trained viewpoint predictor;

[0023] A feature extraction module is used to generate the view labels of the ReID dataset using the trained view predictor, input the view labels of the ReID dataset into a pre-established multi-task feature extractor, and output the view feature vector and the corresponding predicted view and identity feature vector and the corresponding identity ID;

[0024] The comprehensive loss module is used to calculate the comprehensive loss function using the view feature vector and the identity feature vector to obtain the total loss function, thereby improving the performance of vehicle re-identification.

[0025] In combination with the second aspect, in some implementations of the second aspect, the system further includes: the 3D target detection data set of the model training module crops the target image according to the 2D bounding box, filters out images smaller than preset pixels, and then performs super-resolution processing using the ESRGAN method;

[0026] Or the pre-built viewpoint predictor of the model training module uses ResNet-101 as the backbone network, and the original classification layer is replaced by the viewpoint prediction layer;

[0027] Or the loss function L of the pre-established viewpoint predictor of the model training module predictor as follows:

[0028]

[0029] Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value;

[0030] The pre-built multi-task feature extractor of the feature extraction module uses ResNet50-IBN-a as the backbone network and combines the SE channel attention module to enhance the feature representation capability;

[0031] The comprehensive loss function of the comprehensive loss module includes cross entropy loss for identity classification, cross entropy loss for view prediction, triplet loss, center loss, and orthogonal loss;

[0032] The calculation formula of the comprehensive loss function L of the comprehensive loss module is:

[0033] L=L id +λ1Lviewpoint +λ2L triplet +λ3L center +λ4L orth ,

[0034] Among them, λ1, λ2, λ3 and λ4 are hyperparameters that control the relative importance of each loss term, L id is the cross entropy loss for identity classification, L viewpoint is the cross entropy loss of view prediction, L triplet is the triplet loss, L center is the center loss, L orth is the orthogonal loss;

[0035] Orthogonal loss of the comprehensive loss module:

[0036]

[0037] where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation of |B| is the number of samples in the batch.

[0038] Beneficial effects of the present invention:

[0039] The present invention utilizes LiDAR annotations in 3D object detection datasets to train a viewpoint predictor, automatically generates high-quality viewpoint labels, and provides rich supervisory signals for training. Multi-task learning decouples viewpoint features from identity features, ensuring that the learned features have robust viewpoint invariance. The loss function ensures that identity features and viewpoint features do not interfere with each other, and can use the network to decouple identity and viewpoint features, thereby improving the performance of vehicle re-identification tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 It is a schematic flow chart of the method of the present invention;

[0042] Figure 2 It is a schematic diagram of the overall framework of the multi-task learning perspective modeling VMM of the present invention;

[0043] Figure 3is a schematic diagram of the angle between the vehicle's driving direction and the camera's optical axis of the present invention;

[0044] Figure 4 Schematic diagram of the perspective-aware V-ReID network structure of this embodiment;

[0045] Figure 5 It is a schematic diagram of the system structure of the present invention;

[0046] Figure 6 2 is a diagram of experimental results of an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Embodiment 1:

[0049] The following is an introduction to the relevant terms involved in the embodiments of the present application:

[0050] Feature extraction: In machine learning, pattern recognition, and image processing, feature extraction starts with an initial set of measurements and builds derived values ​​(features) that are intended to be informative and non-redundant, thereby facilitating subsequent learning and generalization steps and, in some cases, leading to better interpretability. Feature extraction is related to dimensionality reduction. The quality of features has a crucial impact on generalization capabilities.

[0051] Loss function: A loss function or cost function is a function that maps the value of a random event or its related random variables to a non-negative real number to represent the "risk" or "loss" of the random event. In applications, the loss function is usually associated with the optimization problem as a learning criterion, that is, the model is solved and evaluated by minimizing the loss function.

[0052] like Figure 1 As shown, a perspective-aware vehicle re-identification method based on multi-task learning includes the following steps:

[0053] S101: receiving a 3D object detection dataset, inputting LiDAR annotations in the 3D object detection dataset into a pre-established viewpoint predictor, and obtaining a trained viewpoint predictor;

[0054] Among them, the 3D object detection dataset crops the target image according to the 2D bounding box, filters out images smaller than 50x50 pixels, and then uses the ESRGAN method for super-resolution processing to make the resolution closer to the ReID dataset;

[0055] The pre-built viewpoint predictor is the key innovation of VMM, which automatically obtains viewpoint supervision signals by leveraging LiDAR annotations in 3D object detection datasets. This approach effectively addresses the challenge of estimating vehicle viewpoints from a single image, which is crucial for learning viewpoint-invariant features in V-ReID.

[0056] In the KITTI dataset, the observation angle α (α∈[0,2π)) is defined as the angle between the vehicle's direction of travel and the camera's optical axis ( Figure 3 ). The left sub-image shows the top view of the car and the object coordinate system, while the right sub-image shows the perspective view of the object coordinate system. The formula for calculating the viewing angle α is α=-[π+r y +π+β]. This angle reflects the angle at which the object is detected from the camera's viewpoint, making it an accurate estimate of the object's orientation.

[0057] We use the KITTI dataset, which contains a large number of traffic scene frames and corresponding LiDAR point clouds, to use the 2D bounding box and the viewing angle of the target. We crop the target image according to the 2D bounding box and use the viewing angle in the 3D annotation as the viewpoint prediction label. This creates a dataset of target image-viewpoint pairs.

[0058] The view angle predictor uses ResNet-101 as the backbone network, and the original classification layer is replaced by the view angle prediction layer. Traditional methods usually predict the sine and cosine values ​​of the angle to handle the periodicity of the angle. However, these methods may have discontinuity problems at the boundaries (e.g., transitioning from 359° to 0°), resulting in inaccurate predictions and increased training complexity. To address these issues, we directly predict the angle itself and use a novel loss function to supervise the training:

[0059]

[0060] Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value.

[0061] This loss function constrains the accuracy of the predicted sine and cosine values ​​at the same time, avoiding discontinuity issues caused by angle periodicity. It allows the network to directly output angles in radians without additional post-processing, making predictions more direct and accurate. By directly optimizing angle predictions, our method simplifies the training process and improves the robustness of viewpoint estimation.

[0062] By leveraging these LiDAR annotations, the viewpoint predictor can automatically generate viewpoint labels for large-scale unlabeled images, providing rich viewpoint supervision information for V-ReID training. This innovative approach significantly enhances the framework's ability to model viewpoint changes and learn viewpoint-invariant features.

[0063] S102: Generate the view labels of the ReID dataset using the trained view predictor, input the view labels of the ReID dataset into a pre-established multi-task feature extractor, and output the view feature vector and the corresponding predicted view and identity feature vector and the corresponding identity ID;

[0064] The multi-task feature extractor (view-aware V-ReID feature extractor) learns robust view-invariant features by exploiting the explicit view information provided by the view predictor through multi-task learning and high-level hard sample mining.

[0065] The proposed view-aware V-ReID network structure is as follows: Figure 4 As shown in Figure 2, it adopts the ResNet50-IBN-a

[22] backbone network and uses the SE

[10] channel attention module to enhance feature representation. The network connects the view prediction and identity recognition branches, which share the same backbone network to achieve parameter sharing, information interaction, and collaborative learning. This multi-task learning method ensures that the identity recognition features are both view-invariant and discriminative.

[0066] S103: Calculate a comprehensive loss function using the view feature vector and the identity feature vector to obtain a total loss function, thereby improving the performance of vehicle re-identification.

[0067] In order to effectively learn discriminative and robust vehicle re-ID features, we designed a comprehensive loss function that combines multiple loss components. The total loss function is a weighted sum of the cross entropy loss for identity classification, the cross entropy loss for view prediction, the triplet loss, the center loss, and the orthogonal loss:

[0068] L=L id +λ1L viewpoint +λ2L triplet +λ3L center +λ4L orth , (2)

[0069] where λ1,λ2,λ3 and λ4 are hyperparameters that control the relative importance of each loss term. This design ensures that the features learned by the network are both discriminative for vehicle identity and invariant to viewpoint changes.

[0070] Next, we will introduce the different loss components in Equation 2 separately. a) L id: Cross entropy loss for identity classification. It ensures that the network correctly classifies the vehicle identity. This is crucial for the main task of vehicle re-identification as it directly optimizes the network to distinguish different vehicles; b) L viewpoint : Cross entropy loss for view prediction. To minimize the error in view labeling, we frame view prediction as a 12-class classification problem, dividing the 360-degree range into 30-degree intervals. This discretization reduces noise and inaccuracy, easing learning difficulty and facilitating mining of difficult samples by simplifying the definition of similar and different angles; c) L triplet : Triplet loss, used to minimize the intra-class distance while maximizing the inter-class distance. This loss function encourages the network to learn more discriminative features by ensuring that images of the same vehicle are closer in feature space and images of different vehicles are farther away; d) L center : Center loss, which enforces intra-class compactness by pulling features towards their class center. This loss function complements the triplet loss by further reducing the variance within each class, leading to more robust feature representation.

[0071] Specifically, the present invention is further described below by way of embodiments:

[0072] To further promote the decoupling of identity recognition features and view prediction features, we introduce an orthogonal loss:

[0073]

[0074] where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation is represented by |B|, where |B| is the number of samples in the batch. By minimizing L orth ,We ensure that identity and view features are orthogonal, thus preventing them from interfering with each other and enhancing the overall performance of the network.

[0075] The comprehensive loss function design combines the cross entropy loss for identity classification, the cross entropy loss for view prediction, the triplet loss, the center loss, and the orthogonal loss to ensure that the features learned by the network are both discriminative for vehicle identity and invariant to view changes. By leveraging LiDAR annotations and multi-task learning with orthogonal constraints, the VMM framework effectively addresses the view change challenge in V-ReID. This innovative approach allows the network to decouple identity and view features, thereby improving the performance of vehicle re-identification tasks.

[0076] like Figure 6Figure 2 shows some retrieval results of ReID using VMM on the VeRI dataset. The leftmost image represents the query image, and the rightmost image represents the top ten matching results sorted by confidence. The image in the red box represents a mismatch.

[0077] Embodiment 2: In the second aspect, as Figure 5 As shown, in order to achieve the above-mentioned purpose, the present invention discloses a perspective-aware vehicle re-identification system based on multi-task learning, comprising:

[0078] The model training module 11 is used to receive a 3D object detection data set, input the LiDAR annotations in the 3D object detection data set into a pre-established viewpoint predictor, and obtain a trained viewpoint predictor;

[0079] The feature extraction module 12 is used to generate the view labels of the ReID data set using the trained view predictor, input the view labels of the ReID data set into a pre-established multi-task feature extractor, and output the view feature vector and the corresponding predicted view and the identity feature vector and the corresponding identity ID;

[0080] The comprehensive loss module 13 is used to calculate the comprehensive loss function using the view feature vector and the identity feature vector to obtain the total loss function, thereby improving the performance of vehicle re-identification.

[0081] In combination with the second aspect, in some implementations of the second aspect, the system further includes: the 3D target detection data set of the model training module 11 crops the target image according to the 2D bounding box, filters out images smaller than a preset pixel, and then performs super-resolution processing using the ESRGAN method;

[0082] Alternatively, the pre-established viewpoint predictor of the model training module 11 uses ResNet-101 as a backbone network, and the original classification layer is replaced by a viewpoint prediction layer;

[0083] Or the loss function L of the pre-established view predictor of the model training module 11 predictor as follows:

[0084]

[0085] Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value;

[0086] The pre-built multi-task feature extractor of the feature extraction module 12 uses ResNet50-IBN-a as the backbone network and combines the SE channel attention module to enhance the feature representation capability;

[0087] The comprehensive loss function of the comprehensive loss module 13 includes the cross entropy loss of identity classification, the cross entropy loss of view prediction, the triplet loss, the center loss and the orthogonal loss;

[0088] The calculation formula of the comprehensive loss function L of the comprehensive loss module 13 is:

[0089] L=L id +λ1L viewpoint +λ2L triplet +λ3L center +λ4L orth ,

[0090] Among them, λ1, λ2, λ3 and λ4 are hyperparameters that control the relative importance of each loss term, L id is the cross entropy loss for identity classification, L viewpoint is the cross entropy loss of view prediction, L triplet is the triplet loss, L center is the center loss, L orth is the orthogonal loss;

[0091] Orthogonal loss of the comprehensive loss module:

[0092]

[0093] where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation of |B| is the number of samples in the batch.

[0094] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0095] It needs to be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and the computer program is executed by a processor to execute the above method. The storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0096] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0097] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, the present disclosure may have various changes and improvements, and these changes and improvements fall within the scope of the present disclosure to be protected.

Claims

1. A perspective-aware vehicle re-identification method based on multi-task learning, characterized in that: The method comprises the following steps: receiving a 3D object detection dataset, and inputting LiDAR annotations in the 3D object detection dataset into a pre-established viewpoint predictor to obtain a trained viewpoint predictor; The trained viewpoint predictor is used to generate the viewpoint labels of the ReID dataset, and the viewpoint labels of the ReID dataset are input into the pre-established multi-task feature extractor, and the viewpoint feature vector and the corresponding predicted viewpoint and identity feature vector and the corresponding identity ID are output; The comprehensive loss function is calculated using the view feature vector and the identity feature vector to obtain the total loss function, which can improve the performance of vehicle re-identification.

2. The perspective-aware vehicle re-identification method based on multi-task learning according to claim 1, characterized in that: The 3D object detection dataset crops the target image according to the 2D bounding box, filters out images smaller than a preset pixel, and then uses the ESRGAN method for super-resolution processing.

3. The perspective-aware vehicle re-identification method based on multi-task learning according to claim 1, characterized in that: The pre-established viewpoint predictor uses ResNet-101 as the backbone network, and the original classification layer is replaced by the viewpoint prediction layer.

4. The view-aware vehicle re-identification method based on multi-task learning according to claim 3, characterized in that: The loss function L of the pre-established viewpoint predictor is predictor as follows: Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value.

5. The view-aware vehicle re-identification method based on multi-task learning according to claim 1, characterized in that: The pre-built multi-task feature extractor adopts ResNet50-IBN-a as the backbone network and combines the SE channel attention module to enhance the feature representation capability.

6. The view-aware vehicle re-identification method based on multi-task learning according to claim 1, characterized in that: The comprehensive loss function includes cross entropy loss for identity classification, cross entropy loss for view prediction, triplet loss, center loss and orthogonal loss.

7. The view-aware vehicle re-identification method based on multi-task learning according to claim 6, characterized in that: The calculation formula of the comprehensive loss function L is: L=L id +λ1L viewpoint +λ2L triplet +λ3L center +λ4L orth , Among them, λ1, λ2, λ3 and λ4 are hyperparameters that control the relative importance of each loss term, and L id is the cross entropy loss for identity classification, L viewpoint is the cross entropy loss of view prediction, L triplet is the triplet loss, L center is the center loss, L orth is the orthogonal loss.

8. The view-aware vehicle re-identification method based on multi-task learning according to claim 7, characterized in that: The orthogonal loss: where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation of |B| is the number of samples in the batch.

9. A perspective-aware vehicle re-identification system based on multi-task learning, characterized in that: include: A model training module is used to receive a 3D object detection dataset, input the LiDAR annotations in the 3D object detection dataset into a pre-established viewpoint predictor, and obtain a trained viewpoint predictor; A feature extraction module is used to generate the view labels of the ReID dataset using the trained view predictor, input the view labels of the ReID dataset into a pre-established multi-task feature extractor, and output the view feature vector and the corresponding predicted view and identity feature vector and the corresponding identity ID; The comprehensive loss module is used to calculate the comprehensive loss function using the view feature vector and the identity feature vector to obtain the total loss function, thereby improving the performance of vehicle re-identification.

10. The view-aware vehicle re-identification system based on multi-task learning according to claim 9, characterized in that: The 3D target detection dataset of the model training module crops the target image according to the 2D bounding box, filters out images smaller than a preset pixel, and then uses the ESRGAN method for super-resolution processing; Or the pre-built viewpoint predictor of the model training module uses ResNet-101 as the backbone network, and the original classification layer is replaced by the viewpoint prediction layer; Or the loss function L of the pre-established viewpoint predictor of the model training module predictor as follows: Where N is the number of training samples, is the true viewing angle value of the i-th sample, α i is the predicted viewing angle value; The pre-built multi-task feature extractor of the feature extraction module uses ResNet50-IBN-a as the backbone network and combines the SE channel attention module to enhance the feature representation capability; The comprehensive loss function of the comprehensive loss module includes cross entropy loss for identity classification, cross entropy loss for view prediction, triplet loss, center loss, and orthogonal loss; The calculation formula of the comprehensive loss function L of the comprehensive loss module is: L=L id +λ1L viewpoint +λ2L triplet +λ3L center +λ4L orth , Among them, λ1, λ2, λ3 and λ4 are hyperparameters that control the relative importance of each loss term, and L id is the cross entropy loss for identity classification, L viewpoint is the cross entropy loss of view prediction, L triplet is the triplet loss, L center is the center loss, L orth is the orthogonal loss; Orthogonal loss of the comprehensive loss module: where r id (x i ) represents the i-th sample x i The identity feature representation, r viewpoint (x i ) represents the i-th sample x i The visual feature representation of |B| is the number of samples in the batch.

Images