Face scoring model training method, face scoring method, and related apparatus

Abstract

The present application provides a face scoring model training method, a face scoring method, and a related apparatus. The face scoring model training method comprises: generating pseudo labels on the basis of a first image annotated with a scoring label, and allocating the pseudo labels to second images not annotated with scoring labels to obtain a third image; obtaining a first scoring result on the basis of the first image and the third image; performing feature extraction on the basis of the first scoring result and a first feature to obtain a second feature, and performing feature fusion on the basis of the second feature to obtain a fused feature; obtaining a second scoring result on the basis of the fused feature; extracting a multi-scale feature of a face image to be trained, and obtaining a third scoring result on the basis of a third feature, the second scoring result, and the multi-scale feature; and obtaining a trained face scoring model on the basis of the first scoring result, the second scoring result, and the third scoring result. Weighted fusion is performed by integrating a plurality of scoring results, and the importance and reliability of different features are combined, so that the accuracy and reliability of target scoring results are ensured.

Description

Training methods for face scoring models, face scoring methods and related devices Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a training method for a face scoring model, a face scoring method, and related apparatus. Background Technology

[0002] Face scoring technology combines image recognition and deep learning to rate the attractiveness of facial images. Face scoring models typically require massive amounts of data for training to achieve good results, but in practical applications, there is a problem of insufficient labeled data in the training image database. Furthermore, complex image features and long data sequences can significantly reduce the accuracy and reliability of face scoring models.

[0003] Summary of the Invention

[0004] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0005] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a training method for a face scoring model, a face scoring method and related devices, which can improve the accuracy and reliability of prediction scoring.

[0006] An embodiment of the first aspect of this application provides a method for training a face scoring model, comprising:

[0007] Obtain a face image to be trained and input the face image to be trained into a face scoring model. The face image to be trained includes a first image labeled with a scoring label and a second image without a scoring label.

[0008] A pseudo-label is generated based on the first image, and the pseudo-label is assigned to the second image to obtain a third image labeled with the pseudo-label;

[0009] A first feature is obtained by extracting features from the first image and the third image, and a first score result is obtained based on the first feature.

[0010] Based on the first scoring result and the first feature, feature extraction is performed to obtain the second feature, and feature fusion is performed based on the second feature to obtain the fused feature;

[0011] A third feature is obtained by extracting features based on the fusion features, and a second scoring result is obtained based on the third feature.

[0012] Extract multi-scale features from the training face images, and obtain a third score result based on the third feature, the second score result, and the multi-scale features;

[0013] The parameters of the face scoring model are adjusted based on the first, second, and third scoring results to obtain a trained face scoring model.

[0014] According to certain embodiments of the first aspect of this application, generating pseudo-tags based on the first image includes:

[0015] Weakly supervised learning is performed on the classifier of the face scoring model based on the first image to generate pseudo-labels.

[0016] According to certain embodiments of the first aspect of this application, assigning the pseudo-label to the second image to obtain a third image labeled with the pseudo-label includes:

[0017] The pseudo-labels are filtered based on their confidence levels to obtain the filtered pseudo-labels.

[0018] The filtered pseudo-labels are assigned to the second image to obtain a third image labeled with pseudo-labels.

[0019] According to certain embodiments of the first aspect of this application, the step of extracting a first feature from the first image and the third image, and obtaining a first scoring result based on the first feature, includes:

[0020] The classifier of the face scoring model is semi-supervised learning based on the first image and the third image to extract the first feature;

[0021] Based on the first feature, a classification prediction is performed to obtain the first score result.

[0022] According to certain embodiments of the first aspect of this application, the step of extracting a second feature based on the first scoring result and the first feature, and fusing features based on the second feature to obtain a fused feature, includes:

[0023] The second feature is extracted by performing a convolution based on the first scoring result and the first feature.

[0024] Transform the second feature to a feature space of the same dimension;

[0025] The transformed second feature is then fused to obtain the fused feature.

[0026] According to certain embodiments of the first aspect of this application, the step of extracting a third feature based on the fused features and obtaining a second scoring result based on the third feature includes:

[0027] The classifier of the face scoring model is trained based on the fused features, and the fused features are mapped to the class activation map to obtain the third feature;

[0028] Global max pooling is performed based on the third feature to obtain the second score result.

[0029] According to certain embodiments of the first aspect of this application, adjusting the parameters of the face scoring model based on the first scoring result, the second scoring result, and the third scoring result to obtain the trained face scoring model includes:

[0030] The target score is obtained based on the first preset weight parameter, the second preset weight parameter, the third preset weight parameter, the first score result, the second score result, and the third score result;

[0031] The loss function value is obtained based on the target scoring result;

[0032] The parameters of the face scoring model are adjusted based on the loss function value to obtain the trained face scoring model.

[0033] A second aspect of this application provides a face scoring method, comprising:

[0034] Obtain the face image of the person to be scored;

[0035] The face image to be scored is input into a trained face scoring model for scoring, and the scoring result is obtained;

[0036] The trained face scoring model is obtained by training according to the face scoring model training method as described in any one of claims 1 to 7.

[0037] According to a third aspect of this application, an electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the training method for the face scoring model described in the first aspect of this application and the face scoring method described in the second aspect of this application.

[0038] According to a fourth aspect of this application, a computer-readable storage medium stores a computer program that, when executed by a processor, implements a training method for a face scoring model according to an embodiment of a first aspect of this application and a face scoring method according to an embodiment of a second aspect of this application.

[0039] The above scheme has at least the following beneficial effects: It generates pseudo-labels based on a first image labeled with rating tags, assigns these pseudo-labels to a second image without rating tags, resulting in a third image labeled with pseudo-labels; it extracts features from the first and third images to obtain a first feature, and obtains a first rating result based on the first feature; it extracts features from the first rating result and the first feature to obtain a second feature, and fuses these features to obtain a fused feature; it extracts features from the fused feature to obtain a third feature, and obtains a second rating result based on the third feature; it extracts multi-scale features from the training face image, and obtains a third rating result based on the third feature, the second rating result, and the multi-scale features; it adjusts the parameters of the face rating model based on the first, second, and third rating results to obtain a trained face rating model; and by combining multiple rating results with weighted fusion, considering the importance and reliability of different features, it ensures the accuracy and reliability of the target rating result. Attached Figure Description

[0040] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0041] Figure 1 is a flowchart illustrating the steps of the face scoring method provided in an embodiment of this application;

[0042] Figure 2 is a flowchart illustrating the steps of the training method for the face scoring model provided in the embodiments of this application;

[0043] Figure 3 is a diagram of the sub-steps for generating pseudo-tags from a first image according to an embodiment of this application;

[0044] Figure 4 is a diagram of the sub-steps of assigning pseudo-labels to the second image to obtain a third image labeled with pseudo-labels, according to an embodiment of this application.

[0045] Figure 5 is a sub-step diagram of step S400 provided in the embodiments of this application;

[0046] Figure 6 is a sub-step diagram of step S500 provided in the embodiments of this application;

[0047] Figure 7 is a sub-step diagram of step S600 provided in an embodiment of this application;

[0048] Figure 8 is a sub-step diagram of step S800 provided in an embodiment of this application. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0051] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0052] An embodiment of this application provides a face scoring method.

[0053] Referring to Figure 1, the face scoring method includes the following steps:

[0054] Step S110: Obtain the face image of the person to be scored;

[0055] Step S120: Input the face image to be scored into the trained face scoring model for scoring, and obtain the scoring result.

[0056] The trained face scoring model was obtained by training according to the following face scoring model training method.

[0057] Referring to Figure 2, the training method for the face scoring model includes the following steps:

[0058] Step S200: Obtain the face image to be trained and input the face image to be trained into the face scoring model;

[0059] Step S300: Generate pseudo-labels based on the first image, assign the pseudo-labels to the second image, and obtain a third image labeled with pseudo-labels;

[0060] Step S400: Extract features from the first image and the third image to obtain the first feature, and obtain the first score result based on the first feature;

[0061] Step S500: Extract features based on the first score result and the first feature to obtain the second feature, and fuse features based on the second feature to obtain the fused feature;

[0062] Step S600: Extract the third feature based on the fused features, and obtain the second score result based on the third feature;

[0063] Step S700: Extract multi-scale features from the training face image, and obtain the third score result based on the third feature, the second score result, and the multi-scale features;

[0064] Step S800: Adjust the parameters of the face scoring model based on the first, second, and third scoring results to obtain the trained face scoring model.

[0065] In step S200 of some embodiments, the face image to be trained is acquired. On the one hand, a face image can be captured by a photographic device, such as a camera or webcam; on the other hand, face images can be acquired from a large online database.

[0066] A subset of facial images are manually scored for beauty, and these scores are then labeled to obtain score tags for each image. This results in a first image labeled with a score tag and a second image without a score tag. These two images together form a set of facial images to be used for training.

[0067] Preprocessing steps such as alignment, data augmentation, and normalization are performed on the face images to be trained. For example, data augmentation techniques such as rotation, scaling, translation, flipping, and color transformation are used to preprocess the face images to be trained to increase the diversity and robustness of the data.

[0068] The preprocessed face images to be trained are input into the face scoring model.

[0069] In step S300 of some embodiments, a pseudo-label is generated based on the first image, and the pseudo-label is assigned to the second image to obtain a third image labeled with the pseudo-label.

[0070] For example, a pre-trained convolutional neural network, such as ResNet-50, is used to extract multi-scale feature maps of the face image to be trained. The multi-scale feature maps include feature maps of Conv3, Conv4 and Conv5 layers.

[0071] Referring to Figure 3, generating pseudo-labels based on the first image includes the following steps: Step S310, performing weakly supervised learning on the classifier of the face scoring model based on the first image to generate pseudo-labels. Specifically, the classifier of the face scoring model is trained using the first image labeled with scoring labels. The trained classifier is then used to predict and generate pseudo-labels for the second image without scoring labels. Weakly supervised learning can reduce image annotation costs and quickly obtain a large number of annotated images.

[0072] Referring to Figure 4, pseudo-labels are assigned to the second image to obtain a third image labeled with pseudo-labels, including the following steps:

[0073] Step S320: Filter the pseudo-labels according to the confidence level corresponding to the pseudo-labels to obtain the filtered pseudo-labels;

[0074] Step S330: Assign the filtered pseudo-labels to the second image to obtain the third image labeled with pseudo-labels.

[0075] Specifically, a classifier for a face rating model is trained using a first image labeled with rating tags. The trained classifier is then used to predict the ratings of a second image without tags, generating multiple pseudo-labels. During pseudo-label generation, a confidence probability for each pseudo-label is also generated. Pseudo-labels are then selected based on their confidence levels, typically choosing the pseudo-label with the highest confidence. Each second image without a rating tag is assigned the class with the highest probability as its pseudo-label. This process can be enhanced using entropy minimization, encouraging the model to make low-entropy predictions on unlabeled data; lower entropy results in more accurate predictions.

[0076] Referring to Figure 5, in step S400 of some embodiments, a first feature is obtained by feature extraction based on the first image and the third image, and a first score result is obtained based on the first feature, including the following steps:

[0077] Step S410: Perform semi-supervised learning on the classifier of the face scoring model based on the first image and the third image to extract the first feature;

[0078] Step S420: Perform classification prediction based on the first feature to obtain the first score result.

[0079] The generated pseudo-labels are used together with the real labels to further train the model. In each iteration, pseudo-labeled data with high confidence are added to the training set, predictions with low confidence are removed, and the model is updated. The performance of the model trained with pseudo-labels is evaluated. In some methods, the weights of the pseudo-label part can be adjusted through a deterministic simulation process to avoid getting trapped in local optima. The above process is iterated until the model performance no longer improves or a predetermined number of iterations is reached. In each iteration, the model may need to be fine-tuned to ensure the quality of the pseudo-labels and the model's generalization ability.

[0080] The trained classifier extracts features from the first and third images to obtain the first feature, and performs classification prediction based on the first feature to obtain the first score result.

[0081] Referring to Figure 6, in step S500 of some embodiments, feature extraction is performed based on the first scoring result and the first feature to obtain the second feature, and feature fusion is performed based on the second feature to obtain the fused feature, including the following steps:

[0082] Step S510: Perform convolution based on the first scoring result and the first feature to extract the second feature;

[0083] Step S520: Transform the second feature into a feature space of the same dimension;

[0084] Step S530: Perform feature fusion on the transformed second feature to obtain the fused feature.

[0085] Specifically, a convolutional neural network is used to extract the second feature by convolving the first score result and the first feature. Multiple second features exist in different dimensions; to unify the dimensions of the second features, an embedding layer is used to transform the multiple second features in different dimensions to a feature space of the same dimension. A graph neural network (GNN) is then used to fuse the transformed second features to obtain the fused feature.

[0086] It is understandable that graph neural networks refer to neural networks that learn graph-structured data and extract and discover features and patterns in graph-structured data.

[0087] Referring to Figure 7, in step S600 of some embodiments, a third feature is obtained by feature extraction based on the fused features, and a second scoring result is obtained based on the third feature, including the following steps:

[0088] Step S610: Train a classifier for the face scoring model based on the fused features, and map the fused features to the class activation map to obtain the third feature;

[0089] Step S620: Perform global max pooling based on the third feature to obtain the second score result.

[0090] Specifically, a classifier for the face scoring model is trained based on the fused features. The multi-scale fused features are then mapped to class activation maps using convolutional layers with 1×1 kernels to obtain the third feature. Global max pooling is then performed on the third feature to extract the feature score for each class, which is the second scoring result.

[0091] The use of class activation maps and global max pooling enhances the interpretability of the model, clearly showing the model's focus on different regions of the image, thus improving the credibility of the score and the visualization effect.

[0092] Class Activation Mapping (CAM) is a technique in deep learning used to visualize the decision-making process of convolutional neural networks. It identifies the regions in an image that are most critical to classification decisions, helping to understand whether the model makes classification decisions based on visual or time-series features. The generation of CAM is based on the observation that while global average pooling simplifies the data, it preserves spatial or temporal information. This allows for the inverse inference of the regions that contribute most to the classification result using the feature maps of the convolutional layers and the weights of the corresponding classes.

[0093] Specifically, the fused features are reduced to feature vectors through global average pooling; the feature vectors are multiplied by the weights of the fully connected layer and converted into probabilities through the softmax function; for a specific class, the class activation map is calculated by multiplying the weights of the corresponding class from the global average pooling layer to the softmax layer with the activation feature map of the last convolutional layer.

[0094] By fusing multi-scale feature maps and class activation maps in steps S500 and S600, the prediction accuracy is improved, significantly enhancing the prediction accuracy and the model's generalization ability.

[0095] In step S700 of some embodiments, multi-scale features of the training face image are extracted, and a third score result is obtained based on the third feature, the second score result, and the multi-scale features.

[0096] Specifically, multi-scale features are obtained from training facial images at different scales and levels, capturing various detailed information including low-level edge features and high-level semantic features, ensuring the model can comprehensively understand the image content. The third feature, the second scoring result, and the multi-scale features are fused. By combining feature information and scoring results, the complementarity between features is utilized to improve scoring accuracy and make more precise judgments. After fusing multiple third features, the second scoring result, and the multi-scale features, a third scoring result is generated through further calculation and analysis. This score integrates all feature information and optimization results, providing a comprehensive and reliable beauty evaluation.

[0097] Referring to Figure 8, in step S800 of some embodiments, the parameters of the face scoring model are adjusted according to the first scoring result, the second scoring result, and the third scoring result to obtain a trained face scoring model, including the following steps:

[0098] Step S810: Obtain the target score result based on the first preset weight parameter, the second preset weight parameter, the third preset weight parameter, the first score result, the second score result, and the third score result;

[0099] Step S820: Obtain the loss function value based on the target score result;

[0100] Step S830: Adjust the parameters of the face scoring model according to the loss function value to obtain the trained face scoring model.

[0101] Specifically, through weight allocation, different weights are assigned to each feature based on its importance and reliability to ensure that each feature contributes reasonably to the final score. A first preset weight parameter, a second preset weight parameter, and a third preset weight parameter are configured for the first, second, and third score results, respectively. The first preset weight parameter is denoted as ω1, the second preset weight parameter as ω2, and the third preset weight parameter as ω3. Furthermore, the first, second, and third preset weight parameters satisfy the following relationship: w1 + w2 + w3 = 1. The target score result can be expressed as: S = ω1*S1 + ω2*S2 + ω3*S3, where S1 is the first score result, S2 is the second score result, and S3 is the third score result.

[0102] By combining multiple scoring results with weighted fusion, and considering the importance and reliability of different features, the accuracy and reliability of the target scoring results are ensured.

[0103] An embodiment of this application provides an electronic device. The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the training method and face scoring method of the face scoring model as described above.

[0104] This electronic device can be any smart terminal, including computers.

[0105] In general, for the hardware structure of electronic devices, the processor can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs in order to achieve the technical solutions provided in the embodiments of this application.

[0106] The memory can be implemented in the form of read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory and called by the processor to execute the methods of the embodiments of this application.

[0107] Input / output interfaces are used to implement information input and output.

[0108] The communication interface is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0109] The bus transmits information between various components of a device, such as the processor, memory, input / output interfaces, and communication interfaces. The processor, memory, input / output interfaces, and communication interfaces communicate with each other within the device via the bus.

[0110] An embodiment of this application provides a computer storage medium. The computer storage medium stores computer-executable instructions for executing the training method and face scoring method of the face scoring model described above.

[0111] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0112] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0113] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0114] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0115] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms. Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0117] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A training method for a face scoring model, characterized in that, include: Obtain a face image to be trained and input the face image to be trained into a face scoring model. The face image to be trained includes a first image labeled with a scoring label and a second image without a scoring label. A pseudo-label is generated based on the first image, and the pseudo-label is assigned to the second image to obtain a third image labeled with the pseudo-label; A first feature is obtained by extracting features from the first image and the third image, and a first score result is obtained based on the first feature. Based on the first scoring result and the first feature, feature extraction is performed to obtain the second feature, and feature fusion is performed based on the second feature to obtain the fused feature; A third feature is obtained by extracting features based on the fusion features, and a second scoring result is obtained based on the third feature. Extract multi-scale features from the training face images, and obtain a third score result based on the third feature, the second score result, and the multi-scale features; The parameters of the face scoring model are adjusted based on the first, second, and third scoring results to obtain a trained face scoring model.

2. The training method for the face scoring model according to claim 1, characterized in that, The step of generating pseudo-labels based on the first image includes: Weakly supervised learning is performed on the classifier of the face scoring model based on the first image to generate pseudo-labels.

3. The training method for the face scoring model according to claim 1, characterized in that, Assigning the pseudo-label to the second image to obtain a third image labeled with the pseudo-label includes: The pseudo-labels are filtered based on their confidence levels to obtain the filtered pseudo-labels. The filtered pseudo-labels are assigned to the second image to obtain a third image labeled with pseudo-labels.

4. The training method for the face scoring model according to claim 1, characterized in that, The step of extracting a first feature based on the first image and the third image, and obtaining a first score result based on the first feature, includes: The classifier of the face scoring model is semi-supervised learning based on the first image and the third image to extract the first feature; Based on the first feature, a classification prediction is performed to obtain the first score result.

5. The training method for the face scoring model according to claim 1, characterized in that, The step of extracting a second feature based on the first scoring result and the first feature, and then fusing features based on the second feature to obtain a fused feature, includes: The second feature is extracted by performing a convolution based on the first scoring result and the first feature. Transform the second feature to a feature space of the same dimension; The transformed second feature is then fused to obtain the fused feature.

6. The training method for the face scoring model according to claim 1, characterized in that, The step of extracting a third feature based on the fused features and obtaining a second scoring result based on the third feature includes: The classifier of the face scoring model is trained based on the fused features, and the fused features are mapped to the class activation map to obtain the third feature; Global max pooling is performed based on the third feature to obtain the second score result.

7. The training method for the face scoring model according to claim 1, characterized in that, The step of adjusting the parameters of the face scoring model based on the first scoring result, the second scoring result, and the third scoring result to obtain the trained face scoring model includes: The target score is obtained based on the first preset weight parameter, the second preset weight parameter, the third preset weight parameter, the first score result, the second score result, and the third score result; The loss function value is obtained based on the target scoring result; The parameters of the face scoring model are adjusted based on the loss function value to obtain the trained face scoring model.

8. A face scoring method, characterized in that, include: Obtain the face image of the person to be scored; The face image to be scored is input into a trained face scoring model for scoring, and the scoring result is obtained; The trained face scoring model is obtained by training according to the face scoring model training method as described in any one of claims 1 to 7.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the training method of the face scoring model according to any one of claims 1 to 7 and the face scoring method according to claim 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method of the face scoring model according to any one of claims 1 to 7 and the face scoring method according to claim 8.

Images