Multi-task face beauty prediction method and apparatus, device and medium
Through multi-task learning and regularization processing, the problem of label noise impact in face beauty prediction model training is solved, the prediction accuracy and model robustness are improved, and more efficient face beauty prediction is achieved.
Patent Information
- Application Number
- PCT/CN2023/095557
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-11
- Filing Date
- 2023-05-22
- Publication Date
- 2024-11-14
AI Technical Summary
The existing face beauty prediction method requires a large amount of label data during model training, and label noise affects the model accuracy, resulting in a decrease in prediction effect.
Using multi-task face beauty prediction method, by training the multi-task teacher model, using shared feature networks and independent feature networks to extract features, and combining beauty level labels and noise labels for regularization, conducting collaborative teaching to obtain target model parameters. Weak the effects of overfitting and noise labeling.
It significantly improves the robustness and generalization ability of the model, improves the accuracy of facial beauty prediction, and reduces the negative impact of noise labels on the prediction network.
Smart Images

Figure CN2023095557_14112024_PF_FP_ABST
Abstract
Description
Multi-task facial beauty prediction method, device, equipment and medium Technical Field
[0001] The embodiments of the present application relate to, but are not limited to, the field of image recognition, and in particular to a multi-task facial beauty prediction method, apparatus, device, and medium. Background Art
[0002] Current facial beauty prediction methods typically require a large amount of labeled data for model training. Furthermore, when labeling noisy samples, label quality is affected by subjective factors, tool technology, and other factors, such as manual or machine data annotation, which can also introduce label noise. This label noise can significantly impact model accuracy and reduce facial beauty prediction performance.
[0003] Summary of the Invention
[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.
[0005] The embodiments of the present application aim to solve at least one of the technical problems existing in the prior art. The embodiments of the present application provide a multi-task facial beauty prediction method, apparatus, device and medium, which can improve the accuracy of facial beauty prediction.
[0006] An embodiment of the first aspect of the present invention, a multi-task face beauty prediction method, includes:
[0007] Get face image;
[0008] Training a multi-task teacher model according to the face image to obtain a first model, wherein the multi-task model includes a shared feature network layer for extracting shared features and a plurality of independent feature network layers for extracting different specific features;
[0009] Regularizing the first model according to the beauty level label and the noise label of the face image to obtain a second model;
[0010] Performing collaborative teaching on a plurality of said second models to obtain target model parameters;
[0011] Substituting the target model parameters into the multi-task student model to obtain the target model;
[0012] The target model is used to perform facial beauty prediction processing to obtain a facial beauty prediction result.
[0013] In certain embodiments of the first aspect of the present invention, the facial image includes a plurality of facial beauty-related data; and the step of training a multi-task teacher model based on the facial image to obtain a first model comprises:
[0014] Inputting the plurality of facial beauty-related data into the shared feature network layer for feature extraction to obtain shared features;
[0015] Inputting the plurality of facial beauty-related data into the corresponding independent feature network layer for feature extraction to obtain a plurality of corresponding specific features;
[0016] The shared features and the multiple specific features are respectively input into multiple task network layers of the multi-task teacher model to obtain multiple corresponding task output results, so as to train the multi-task teacher model to obtain a first model.
[0017] In certain embodiments of the first aspect of the present invention, regularizing the first model according to the beauty level label and the noise label of the facial image to obtain the second model includes:
[0018] The first model is trained based on face images with beauty level labels to obtain a base model;
[0019] The first model is trained according to the face image with the noise label to obtain a noise model;
[0020] Inversely updating the parameters of the basic model according to the loss function of the noise model;
[0021] The second model is obtained by regularizing the basic model after updating the parameters.
[0022] In certain embodiments of the first aspect of the present invention, performing regularization processing on the basic model after parameter update to obtain the second model includes:
[0023] Add a softmax layer to the base model after updating the parameters;
[0024] Set layer weights and bias values for the output of the base model with the softmax layer added;
[0025] Perform dropout regularization on the output of the base model with set layer weights and bias values to obtain the target output;
[0026] Obtaining a target loss function of a base model according to the target output and the first loss function;
[0027] The parameters of the basic model are optimized according to the objective loss function to obtain a second model.
[0028] In certain embodiments of the first aspect of the present invention, the first loss function is a cross entropy loss function.
[0029] In certain embodiments of the first aspect of the present invention, there are two second models; and the collaborative teaching of the plurality of second models to obtain target model parameters comprises:
[0030] determining a first sample according to a second network, wherein the first sample is a sample having a clean probability greater than a preset probability when the second network performs feedforward selection;
[0031] Determine a second sample according to another second network, where the second sample is a sample whose clean probability is greater than a preset probability when the other second network performs feedforward selection;
[0032] selecting a third sample from the first sample according to a small loss criterion, wherein the ratio of the number of the third sample to the number of the first sample is a preset ratio;
[0033] selecting a fourth sample from the second sample according to a small loss criterion, wherein the ratio of the number of the fourth sample to the number of the second sample is a preset ratio;
[0034] Training one of the second networks with the fourth sample to obtain first model parameters, and training another of the second networks with the third sample to obtain second model parameters;
[0035] Target model parameters are obtained according to the first model parameters and the second model parameters.
[0036] In certain embodiments of the first aspect of the present invention, the preset ratio is related to a forgetting rate of the second model.
[0037] According to an embodiment of the second aspect of the present invention, a facial beauty prediction device includes:
[0038] An input unit, the input unit being used to obtain a face image;
[0039] A model training unit, wherein the model training unit is used to train a multi-task teacher model according to the face image to obtain a first model, wherein the multi-task model includes a shared feature network layer for extracting shared features and a plurality of independent feature network layers for extracting different specific features;
[0040] A regularization unit, configured to perform regularization processing on the first model according to the beauty level label and the noise label of the face image to obtain a second model;
[0041] a collaborative teaching unit, configured to collaboratively teach the plurality of second models to obtain target model parameters, and substitute the target model parameters into the multi-task student model to obtain a target model;
[0042] A prediction unit is used to use the target model to perform facial beauty prediction processing to obtain a facial beauty prediction result.
[0043] An embodiment of the third aspect of the present invention is an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the multi-task facial beauty prediction method as described above is implemented.
[0044] An embodiment of the fourth aspect of the present invention is a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the multi-task face beauty prediction method as described above.
[0045] The above scheme has at least the following beneficial effects: feature compression is achieved through regularization, which can optimize the training process and thus weaken the influence of bias terms related to overfitting; combined with collaborative teaching, it further prevents the network from overfitting noisy labels and reduces the negative impact of noisy labels on the face beauty prediction network; a model framework is established based on multi-task training, and the supervision information of multiple related tasks is utilized to share model parameters, thereby effectively improving the learning ability of the model; the robustness of the model to noise and the generalization ability of the model are significantly improved, thereby improving the accuracy of face beauty prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] The accompanying drawings are used to provide a further understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.
[0047] FIG1 is a step diagram of a multi-task face beauty prediction method provided by an embodiment of the present application;
[0048] FIG2 is a sub-step diagram of step S300;
[0049] FIG3 is a sub-step diagram of step S340;
[0050] FIG4 is a sub-step diagram of step S400;
[0051] FIG5 is a structural diagram of a facial beauty prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.
[0053] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and the like in the specification, claims, or accompanying drawings are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.
[0054] The embodiments of the present application are further described below with reference to the accompanying drawings.
[0055] The embodiments of the present application provide a multi-task method for predicting facial beauty.
[0056] Referring to FIG1 , the multi-task face beauty prediction method includes but is not limited to the following steps:
[0057] Step S100, obtaining a face image;
[0058] Step S200, training a multi-task teacher model based on the face image to obtain a first model;
[0059] Step S300, regularizing the first model according to the beauty level label and noise label of the face image to obtain a second model;
[0060] Step S400, performing collaborative teaching on multiple second models to obtain target model parameters;
[0061] Step S500, substituting the target model parameters into the multi-task student model to obtain the target model;
[0062] Step S600: performing facial beauty prediction processing using the target model to obtain a facial beauty prediction result.
[0063] In this embodiment, feature compression is achieved through regularization, which can optimize the training process and thus weaken the influence of bias terms related to overfitting; and combined with collaborative teaching, it further prevents the network from overfitting noisy labels and reduces the negative impact of noisy labels on the face beauty prediction network; a model framework is established based on multi-task training, and the supervision information of multiple related tasks is utilized to share model parameters, thereby effectively improving the model's learning ability and prediction accuracy; significantly improving the model's robustness to noise and the model's generalization ability, thereby improving the accuracy of face beauty prediction.
[0064] For step S100, facial images can be obtained from an external database, such as the Large Scale Asia Facial Beauty Database (LSAFBD) and the SCUT-FBP5500 database, etc. The facial images are used to train the network model.
[0065] In addition, face images need to be preprocessed, such as image transformation and correction, image enhancement, etc.
[0066] Image preprocessing can normalize and standardize facial images, which is beneficial for subsequent model training.
[0067] Facial images include a variety of facial beauty-related data, such as gender data, facial beauty data, etc.
[0068] 2 , in step S200 , training a multi-task teacher model based on a face image to obtain a first model includes but is not limited to the following steps:
[0069] Step S210: Inputting a plurality of facial beauty-related data into a shared feature network layer for feature extraction to obtain shared features;
[0070] Step S220: Inputting the plurality of facial beauty-related data into the corresponding independent feature network layer for feature extraction to obtain a plurality of corresponding specific features;
[0071] In step S230 , the shared features and the multiple specific features are respectively input into multiple task network layers of the multi-task teacher model to obtain multiple corresponding task output results, so as to train the multi-task teacher model to obtain a first model.
[0072] The multi-task model includes a shared feature network layer for extracting shared features, multiple independent feature network layers for extracting different specific features, and multiple task network layers. The multiple task network layers are connected to the multiple independent feature network layers in a one-to-one correspondence.
[0073] In tasks related to face beauty prediction, such as beauty prediction based on face beauty data and gender prediction based on gender data, there are similar task data features. A multi-task learning network can be used, which enables multiple task features to promote learning from each other and perform predictions on multiple tasks simultaneously.
[0074] The parameters in the shared feature network layer are shared uniformly between different tasks; the parameters in the independent feature network layer are independent of each other, which effectively reduces the risk of model overfitting.
[0075] 3 , in step S300 , regularizing the first model according to the beauty level label and the noise label of the face image to obtain the second model includes but is not limited to the following steps:
[0076] Step S310, training a first model based on the face image with the beauty level label to obtain a base model;
[0077] Step S320, training a first model based on the face image with the noise label to obtain a noise model;
[0078] Step S330, reversely updating the parameters of the basic model according to the loss function of the noise model;
[0079] Step S340 , performing regularization processing on the basic model after parameter update to obtain a second model.
[0080] For n face images, the face images are x i Denotes, i∈n,; the beauty level label of the face image is y i ∈{1,K,C}, then the training set of face images with beautiful grade labels can be expressed as D={(x1,y1),(x2,y2),K,(x n ,y n )}. The noise label is y′ i The training set of face images with noise labels can be expressed as D′={(x1,y′1),(x2,y′2),K,(x n ,y′ n )}.
[0081] For step S310, the beautiful level label y i The first model is trained on the face image of the human face to obtain the base model, so that the base model includes a vector θ. The vector θ includes the layer weight value and the bias value.
[0082] For step S320, the noise label y′ i The first model is trained on the face image to obtain the noise model Z.
[0083] For step S330, according to the loss function L of the noise model z Update the parameters of the base model in reverse order.
[0084] In step S340, regularization is performed on the basic model after parameter update to obtain a second model, including but not limited to the following steps:
[0085] Step S341, adding a softmax layer to the basic model after parameter update;
[0086] Step S342, setting layer weight values and bias values for the output of the base model with the softmax layer added;
[0087] Step S343, performing dropout regularization processing on the output of the basic model with the set layer weight values and bias values to obtain the target output;
[0088] Step S344, obtaining a target loss function of the basic model according to the target output and the first loss function;
[0089] Step S345: Optimize the parameters of the basic model according to the target loss function to obtain a second model.
[0090] Specifically, a softmax layer is added to the output of the basic model after the parameters are updated, and the layer weight value W and bias value h are set for the output of the basic model with the softmax layer added, and the output of the softmax layer is subjected to dropout regularization to obtain the target output. Use the cross entropy loss function to output the target The target loss function of the model is obtained, and the parameters of the basic model are optimized according to the target loss function to obtain the second model.
[0091] The above process can be expressed by the following formula: g=σ(Wh) a:Bern(q)
[0092] Among them, z i is the output value of the i-th output node of the softmax layer of the basic model, N is the number of output nodes, Bern(q) is the Bernoulli distribution, a is a random variable, and e is the Hadamard product.
[0093] In step S400, two second models are typically collaboratively taught. One second network is h1, and the other second network is h2. Collaborative teaching involves, for two networks with identical architectures, one network selecting its own mini-batch of samples and giving it to its peer network. Collaborative teaching is performed using a small loss criterion to update parameters.
[0094] 4 , performing collaborative teaching on multiple second models to obtain target model parameters includes but is not limited to the following steps:
[0095] Step S410, determining a first sample according to a second network;
[0096] Step S420, determining a second sample according to another second network;
[0097] Step S430, selecting a third sample from the first sample according to a small loss criterion;
[0098] Step S440, selecting a fourth sample from the second sample according to a small loss criterion;
[0099] Step S450: training one of the second networks using the fourth sample to obtain first model parameters, and training another of the second networks using the third sample to obtain second model parameters;
[0100] Step S460: Obtain target model parameters according to the first model parameters and the second model parameters.
[0101] For step S410, the network h1 selects its small batch sample D1 as the first sample. The first sample is a sample whose clean probability is greater than the preset probability when the second network performs feedforward selection. The first sample may be a clean sample or an unclean sample.
[0102] For step S420, network h2 selects its small batch sample D2 as the second sample. The second sample is a sample whose clean probability is greater than a preset probability when another second network performs feedforward selection. The second sample may be a clean sample or an unclean sample.
[0103] For step S430, the third sample D1 is obtained by selecting samples in the ratio of R(T) according to the small loss criterion. The ratio of the number of the third samples to the number of the first samples is a preset ratio R(T). Step S430 can be expressed by the following formula:
[0104] For step S440, the fourth sample D2 is obtained by selecting the sample according to the small loss criterion and the ratio of R(T). The ratio of the number of the fourth samples to the number of the second samples is a preset ratio R(T). Step S440 can be expressed by the following formula:
[0105] For step S450, the third sample Send it to network h2 for training, and update the parameter ω2 of network h2, and The training is sent to the network h1, and the parameter ω1 of the network h1 is updated. Step S450 can be expressed by the following formula: Where η is the learning rate.
[0106] Specifically, the preset ratio is related to the forgetting rate of the second model, and the preset ratio can be expressed by the following formula: R(T)=1-λ forget Among them, λ forget is the forgetting rate of the second model.
[0107] In step S500, the target model parameters are substituted into the multi-task student model to obtain the target model, wherein the multi-task student model has the same architecture as the multi-task teacher model.
[0108] In step S600, facial beauty prediction processing is performed using the target model to obtain a facial beauty prediction result.
[0109] An embodiment of the present invention provides a facial beauty prediction device.
[0110] 5 , the facial beauty prediction apparatus includes an input unit 10 , a model training unit 20 , a regularization unit 30 , a collaborative teaching unit 40 , and a prediction unit 50 .
[0111] Among them, the input unit 10 is used to obtain a facial image; the model training unit 20 is used to train a multi-task teacher model based on the facial image to obtain a first model, and the multi-task model includes a shared feature network layer for extracting shared features and multiple independent feature network layers for extracting different specific features; the regularization unit 30 is used to regularize the first model according to the beauty level label and noise label of the facial image to obtain a second model; the collaborative teaching unit 40 is used to collaboratively teach multiple second models to obtain target model parameters, and substitute the target model parameters into the multi-task student model to obtain a target model; the prediction unit 50 is used to use the target model to perform facial beauty prediction processing to obtain a facial beauty prediction result.
[0112] It can be understood that the face beauty prediction device in this embodiment adopts the multi-task face beauty prediction method as above. Each unit of the face beauty prediction device in this embodiment corresponds to the steps of the above-mentioned multi-task face beauty prediction method, solves the same technical problems as the above-mentioned multi-task face beauty prediction method, and has the same beneficial effects as the above-mentioned multi-task face beauty prediction method.
[0113] An embodiment of the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the multi-task facial beauty prediction method as described above is implemented.
[0114] The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.
[0115] In general, for the hardware structure of the electronic device, the processor can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.
[0116] The memory can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory can store an operating system and other application programs. 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 is called by the processor to execute the methods of the embodiments of this application.
[0117] The input / output interface is used to realize information input and output.
[0118] The communication interface is used to realize the communication interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0119] The bus transmits information between the various components of the device (such as the processor, memory, input / output interface, and communication interface). The processor, memory, input / output interface, and communication interface communicate with each other within the device through the bus.
[0120] An embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions for executing the multi-task face beauty prediction method as described above.
[0121] It should be appreciated that the method steps in the embodiments of the present invention can be implemented or executed by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. In addition, for this purpose, the program can be run on a programmed application-specific integrated circuit.
[0122] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that can be executed by one or more processors.
[0123] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a smartphone, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Various aspects of the present invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, RAM, ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or portions thereof, can be transmitted over a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques of the present invention, the present invention also includes the computer itself.
[0124] The computer program can be applied to input data to perform the functions described herein, thereby converting the input data to generate output data that is stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on the display.
[0125] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and intent of the present application, and that the scope of the present application is defined by the claims and their equivalents.
[0126] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the present embodiment.
Claims
1. A multi-task face beauty prediction method, characterized in that: include: Get face image; Training a multi-task teacher model according to the face image to obtain a first model, wherein the multi-task model includes a shared feature network layer for extracting shared features and a plurality of independent feature network layers for extracting different specific features; Regularizing the first model according to the beauty level label and the noise label of the face image to obtain a second model; Performing collaborative teaching on a plurality of said second models to obtain target model parameters; Substituting the target model parameters into the multi-task student model to obtain the target model; The target model is used to perform face beauty prediction processing to obtain a face beauty prediction result.
2. The multi-task face beauty prediction method according to claim 1, characterized in that: The face image includes a plurality of face beauty-related data; the step of training a multi-task teacher model according to the face image to obtain a first model includes: Inputting a plurality of the facial beauty-related data into the shared feature network layer for feature extraction to obtain shared features; Inputting the plurality of facial beauty-related data into the corresponding independent feature network layers for feature extraction to obtain a plurality of corresponding specific features; The shared features and the multiple specific features are respectively input into multiple task network layers of the multi-task teacher model to obtain multiple corresponding task output results, so as to train the multi-task teacher model to obtain a first model.
3. The multi-task face beauty prediction method according to claim 1, characterized in that: The step of regularizing the first model according to the beauty level label and the noise label of the face image to obtain the second model includes: The first model is trained according to the face images with beauty level labels to obtain a base model; Training the first model according to the face image with noise labels to obtain a noise model; Inversely updating the parameters of the basic model according to the loss function of the noise model; The second model is obtained by regularizing the basic model after updating the parameters.
4. The multi-task face beauty prediction method according to claim 3, characterized in that: The step of performing regularization processing on the basic model after updating the parameters to obtain the second model includes: Add a softmax layer to the base model after updating parameters; Set layer weights and bias values for the output of the base model with the softmax layer added; Perform dropout regularization on the output of the basic model that sets the layer weights and bias values to obtain the target output; Obtaining a target loss function of a basic model according to the target output and the first loss function; The parameters of the basic model are optimized according to the objective loss function to obtain a second model.
5. The multi-task face beauty prediction method according to claim 4, characterized in that: The first loss function is a cross entropy loss function.
6. The multi-task face beauty prediction method according to claim 1, characterized in that: There are two second models; and the method of performing collaborative teaching on the plurality of second models to obtain target model parameters includes: Determine a first sample according to one of the second networks, wherein the first sample is a sample whose clean probability is greater than a preset probability when one of the second networks performs feedforward selection; Determine a second sample according to another second network, wherein the second sample is a sample whose clean probability is greater than a preset probability when another second network performs feedforward selection; Selecting a third sample from the first sample according to a small loss criterion, wherein the ratio of the number of the third sample to the number of the first sample is a preset ratio; Selecting a fourth sample from the second sample according to a small loss criterion, wherein the ratio of the number of the fourth sample to the number of the second sample is a preset ratio; Training one of the second networks with the fourth sample to obtain first model parameters, and training another second network with the third sample to obtain second model parameters; A target model parameter is obtained according to the first model parameter and the second model parameter.
7. The multi-task face beauty prediction method according to claim 6, characterized in that: The preset ratio is related to the forgetting rate of the second model.
8. A facial beauty prediction device, characterized in that: include: An input unit, the input unit is used to obtain a face image; A model training unit, the model training unit is used to train a multi-task teacher model according to the face image to obtain a first model, the multi-task model includes a shared feature network layer for extracting shared features and a plurality of independent feature network layers for extracting different specific features; A regularization unit, configured to perform regularization processing on the first model according to the beauty level label and the noise label of the face image to obtain a second model; A collaborative teaching unit, the collaborative teaching unit is used to collaboratively teach the plurality of the second models to obtain target model parameters, and substitute the target model parameters into the multi-task student model to obtain the target model; A prediction unit is used to use the target model to perform face beauty prediction processing to obtain a face beauty prediction result.
9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the multi-task human face beauty prediction method as described in any one of claims 1 to 7 is implemented.
10. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to execute the multi-task face beauty prediction method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Multi-dimensional task face beauty prediction method and system and storage medium
CN111695602A
Human face key point detection method and system based on knowledge distillation and multi-task learning
CN115393934A
Face beauty prediction method and device, electronic equipment and storage medium
CN115953817A
Facial beauty prediction method and device based on multi-task migration
US20220092297A1
