Target object part identification method, model training method and computing device

Abstract

The embodiment of the invention relates to the technical field of computers, in particular to a target object part recognition method, a model training method and computing device.The target object part recognition method comprises the steps that multiple to-be-processed images of a target object part are obtained, and the shooting angles of the multiple to-be-processed images are different; and inputting the plurality of to-be-processed images into an image processing model to obtain a plurality of skeleton point information of the target object part, the image processing model being obtained based on training of a reference model, the reference model identifying the skeleton point information of the target object part according to the part image of the target object part, and obtaining a plurality of skeleton point information of the target object part. The number of model parameters of the reference model is greater than the number of model parameters of the image processing model; and determining an identification result of the target object part according to the information of the plurality of skeleton points.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and in particular to a method for identifying a part of a target object, a model training method, and a computing device. Background Art

[0002] The recognition technology of human foot shape is an important part of modern biomechanics and medicine, which aims to evaluate the health of individual feet by analyzing the morphological and functional characteristics of the foot. The research and development in this field began with the understanding of the pressure and distribution pattern of the foot during human walking. With the advancement of science and technology, traditional manual measurement and visual evaluation methods have gradually been supplemented or even replaced by automated and digital technologies.

[0003] Early foot shape recognition mainly relied on the experience of clinicians and simple physical measurement tools, such as foot length and width measuring tape. However, these methods have certain limitations, such as strong subjectivity and low accuracy. With the development of computer technology and sensor technology, more advanced recognition methods have emerged, such as pressure distribution analysis systems, 3D scanning technology, and smart wearable devices.

[0004] The pressure distribution analysis system can record the pressure changes in various parts of the sole of the foot in real time during walking through the sensor pads set on the ground, thereby helping to identify the type of arch (such as normal arch, flat foot, etc.) and whether there are abnormal phenomena such as inversion and pronation. 3D scanning technology can obtain a 3D model of the foot in a short time, providing accurate data support for shoe design and customization of orthopedic devices.

[0005] In recent years, with the development of the Internet of Things and big data, smart wearable devices have also been widely used in foot health monitoring. Such devices can not only record the user's gait data for a long time, but also detect the movement state and pressure distribution of the foot through built-in sensors, and then conduct in-depth health assessments through the cloud-based big data analysis platform. In practical applications, they lack convenience, are costly, and have poor recognition accuracy. Therefore, an effective technical solution is urgently needed to solve the above problems. Summary of the invention

[0006] In view of this, an embodiment of this specification provides a method for identifying a target object part. One or more embodiments of this specification also relate to a device for identifying a target object part, a model training method, a model training device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0007] According to a first aspect of an embodiment of this specification, a method for identifying a part of a target object is provided, comprising: Acquire a plurality of to-be-processed images of a target object portion, wherein the plurality of to-be-processed images are taken at different angles; Inputting the multiple images to be processed into an image processing model to obtain multiple bone point information of the target object part, wherein the image processing model is obtained based on the training of a reference model, the reference model recognizes the bone point information of the target object part according to the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model; The recognition result of the target object part is determined according to the multiple skeleton point information.

[0008] According to a second aspect of an embodiment of this specification, there is provided a device for identifying a part of a target object, comprising: An acquisition module is configured to acquire a plurality of images to be processed of a target object portion, wherein the plurality of images to be processed are shot at different angles; An input module is configured to input the multiple images to be processed into an image processing model to obtain multiple bone point information of the target object part, wherein the image processing model is obtained based on reference model training, the reference model recognizes the bone point information of the target object part according to the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model; The determination module is configured to determine the recognition result of the target object part according to the multiple skeleton point information.

[0009] According to a third aspect of an embodiment of this specification, a model training method is provided, including: Determine a first sample image, a second sample image and label bone point information corresponding to the second sample image of the target object part; According to the first sample image, the second sample image, the label skeleton point information and the reference model, the image processing model to be trained is trained until an image processing model that meets the training stop condition is obtained; The image processing model is used to execute the target object part recognition method provided in the first aspect of the embodiment of this specification.

[0010] According to a fourth aspect of the embodiments of this specification, a model training device is provided, including: A determination module, configured to determine the first sample image, the second sample image and the label bone point information corresponding to the second sample image of the target object part; A training module is configured to train the image processing model to be trained according to the first sample image, the second sample image, the label skeleton point information and the reference model until an image processing model that meets the training stop condition is obtained; The image processing model is used to execute the target object part recognition method provided in the first aspect of the embodiment of this specification.

[0011] According to a fifth aspect of an embodiment of this specification, a computing device is provided, including: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the above method are implemented.

[0012] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, and the steps of the above method are implemented when the computer program / instruction is executed by a processor.

[0013] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0014] An embodiment of the present specification provides a method for identifying a target object part, comprising: acquiring a plurality of images to be processed of the target object part, wherein the plurality of images to be processed are shot at different angles; inputting the plurality of images to be processed into an image processing model to obtain a plurality of bone point information of the target object part, wherein the image processing model is trained based on a reference model, the reference model identifies the bone point information of the target object part based on the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model; determining the recognition result of the target object part based on the plurality of bone point information.

[0015] The above method uses an image processing model to perform skeleton point recognition on multiple images to be processed of the target object part, obtains multiple skeleton point information of the target object part, and determines the recognition result of the target object part based on the multiple skeleton point information. The target object part can be recognized only by uploading the image through the device, thereby achieving the convenience of target object part recognition and reducing the recognition cost. In addition, the image processing model can be trained using a reference model with a large number of model parameters to further improve the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1This is a schematic diagram of an application scenario of a method for identifying a target object part provided by an embodiment of this specification; Figure 2 is a flow chart of a method for identifying a target object part provided by an embodiment of this specification; Figure 3 It is a system architecture diagram of a method for identifying a target object part provided by an embodiment of this specification; Figure 4 is a process flow chart of a method for identifying a target object part provided by an embodiment of the present specification; Figure 5 is a flow chart of a model training method provided by an embodiment of this specification; Figure 6 is a processing flow chart of a model training method provided by an embodiment of this specification; Figure 7 It is a structural schematic diagram of a target object part recognition device provided by an embodiment of this specification; Figure 8 It is a structural schematic diagram of a model training device provided by an embodiment of this specification; Fig. 9 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0017] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0018] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0019] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0020] In addition, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0021] In one or more embodiments of this specification, a large model refers to a deep learning model with large-scale model parameters, which usually contains hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than 10 trillion model parameters. A large model can also be called a foundation model / foundation model. The large model is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks, and the model has good generalization ability, such as a large-scale language model (LLM), a multi-modal pre-training model, etc.

[0022] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image caption (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0023] In this specification, a method for identifying a part of a target object is provided. This specification also relates to a device for identifying a part of a target object, a model training method, a model training device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0024] See also Figure 1 , Figure 1 A schematic diagram of an application scenario of a method for identifying a part of a target object provided according to an embodiment of the present specification is shown.

[0025] Figure 1 It includes a terminal side device 102 and a cloud side device 104.

[0026] During specific implementation, the target user can upload or take pictures of multiple images to be processed of the target object part through the terminal side device 102, and the terminal side device 102 can send the multiple images to be processed to the cloud side device 104. The cloud side device 104 can call the image processing model to perform skeleton point recognition on the multiple images to be processed, obtain multiple skeleton point information of the target object part, and calculate the recognition result for the target object part based on the multiple skeleton point information. The cloud side device 104 sends the recognition result to the terminal side device 102 to realize the recognition of the target object part of the target user.

[0027] The end-side device 102 may include a browser, an APP (Application), or a web application such as an H5 (Hyper Text Markup Language 5, version 5 of Hypertext Markup Language) application, or a light application (also known as a mini-program, a lightweight application) or a cloud application, etc. The end-side device may be based on a software development kit (SDK) of the corresponding service provided by the server, such as based on the real-time communication (RTC, Real Time Communication) SDK development and acquisition. The end-side device may be deployed in an electronic device and needs to rely on the device to run or some APPs in the device to run. The electronic device may have a display screen and support information browsing, such as a personal mobile terminal such as a mobile phone, a tablet computer, a personal computer, etc. Various other types of applications may also be configured in the electronic device, such as human-computer dialogue applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0028] The cloud-side device 104 can be understood as a server that provides various services, including physical servers and cloud servers, such as a server that provides communication services for multiple clients, a server for background training that supports the model used on the client, and a server that processes the data sent by the client. It should be noted that the cloud-side device 104 can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. The cloud-side device 104 can also be a server for a distributed system, or a server combined with a blockchain. The cloud-side device 104 can also be a cloud server for basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0029] It is worth noting that the image processing method provided in the embodiments of this specification can be executed by the cloud-side device 104. In other embodiments of this specification, the image processing model can be deployed in the terminal-side device 102, so that the terminal-side device 102 can also have similar functions as the cloud-side device 104, thereby executing the image processing method provided in the embodiments of this specification; in other embodiments, the image processing method provided in the embodiments of this specification can also be executed jointly by the terminal-side device 102 and the cloud-side device 104.

[0030] See also Figure 2 , Figure 2 A flow chart of a method for identifying a target object part provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0031] Step 202: Acquire a plurality of to-be-processed images of a target object portion, wherein the plurality of to-be-processed images are taken at different angles.

[0032] Specifically, the target object part recognition method provided in the embodiments of this specification can be applied to the client.

[0033] Among them, the target object part can be, for example, a part of the human body, such as a hand, a foot, and the like. For ease of understanding, the embodiments of this specification take the target object part as an example of a foot, so the image to be processed can be understood as an image of the foot. The shooting angles of the multiple images to be processed are different, for example, they can include an image of the front of the human foot, an image of the back of the heel of the human foot, an image of the side of the human foot, and the like. In actual applications, an image of the inner side of the human foot can be obtained to analyze the arch problem of the human body, and an image of the back of the heel of the human body can be obtained to analyze the heel problem of the human body.

[0034] Based on this, see Figure 3 , Figure 3 A system architecture diagram of a target object part recognition method provided according to an embodiment of the present specification is shown. Figure 3 As shown, the target object part can be identified based on a target object part identification system 300. The identification system 300 can include an image acquisition unit 302. The image acquisition unit 302 can acquire a plurality of to-be-processed images of a human foot.

[0035] In a specific implementation, the step of obtaining a plurality of to-be-processed images of the target object part includes: According to a preset shooting rule, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

[0036] The preset shooting rules may be understood as rules for guiding the target user to shoot the target object part, and the preset shooting rules may include guiding the target user's shooting position, shooting distance, shooting angle, etc.

[0037] Specifically, the image acquisition unit 302 may guide the target user to shoot the feet based on a preset shooting rule, and obtain a plurality of to-be-processed images of the target user's feet.

[0038] In actual application, the image acquisition unit 302 can guide the target user to take three images of the feet to be processed, which may include the target user's left foot image, right foot image, and heel images of both feet. In addition, the image acquisition unit 302 can also guide the target user to take foot images at other angles, such as the target user's feet images, left foot inner side image, left foot outer side image, right foot inner side image, and right foot outer side image.

[0039] In addition, the image acquisition unit 302 may also directly receive multiple foot images uploaded by the target user without shooting. The embodiment of this specification does not limit the method for acquiring the images to be processed.

[0040] In summary, by guiding the target user to take pictures of the target object, the image to be processed is made more in line with the recognition requirements, the accuracy of the recognition result is further improved, and the shooting is simple and the cost is low.

[0041] Furthermore, photographing the target object part according to a preset photographing rule to obtain a plurality of to-be-processed images of the target object part includes: In response to the image shooting instruction, displaying the shooting area on the display interface; Detecting skeleton points of the target object part in the shooting area, and calculating shooting index data of the target object part according to the skeleton points; When it is determined according to the shooting index data that the target object part meets the preset shooting condition, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

[0042] Among them, the image shooting instruction can be understood as the instruction for the target user to click on the shooting control of the display interface, the shooting area can be understood as the area used to shoot the target object part, and the shooting area can display the current shooting picture, and the shooting index data can be understood as the index data used to detect whether the target object part of the target user enters the shooting area. The shooting index data may include the number of foot bone points, the number of left foot bone points, the number of right foot bone points, the difference between the horizontal coordinate of the left ankle and the horizontal coordinate of the right ankle, the ratio of the calf length to the height of the shooting area, etc. The preset shooting condition can be understood as the shooting index data meeting the preset index threshold.

[0043] Based on this, in response to the target user's instruction to click on the shooting control of the display interface, the shooting area can be displayed on the display interface, the target user can adjust the position of the target object part so that the target object part enters the shooting area, detect the skeleton points of the target object part in the shooting area, calculate the shooting index data of the target object part according to the skeleton points, and when it is determined according to the shooting index data that the shooting index data meets the preset index threshold, it can be determined that the target object part meets the preset shooting condition, and the target object part is photographed to obtain multiple images of the target object part to be processed.

[0044] In summary, by judging whether the target object part meets the preset shooting conditions, the target object part is directly photographed if the preset shooting conditions are met, so as to obtain the unprocessed image of the target object part of the target user, which is convenient for subsequent identification.

[0045] In addition, after calculating the shooting index data of the target object part, the method further includes: If it is determined according to the shooting index data that the target object part does not meet the preset shooting condition, generating prompt information, and displaying the prompt information on the display interface, so that the target user can adjust the target object part according to the prompt information; The step of detecting the skeleton points of the target object part in the shooting area and calculating the shooting index data of the target object part according to the skeleton points is continued.

[0046] The prompt information may be used to prompt the target user to adjust the target object part. For example, the prompt information may prompt the target user that the target user is currently close to the camera, or may prompt the target user that the target object part has not entered the shooting area.

[0047] Specifically, when it can be determined that the shooting index data does not meet the preset index threshold, it can be determined that the target object part does not meet the preset shooting conditions, and a prompt message can be generated and displayed on the display interface, so that the target user can adjust the position, direction and angle of the target object according to the prompt message, and continue to execute the above-mentioned steps of detecting the skeleton points of the target object part in the shooting area, and calculating the shooting index data of the target object part according to the skeleton points, until it is determined that the target object part meets the preset shooting conditions, and the target object part is photographed.

[0048] In summary, when it is determined that the target object part does not meet the preset shooting conditions, prompt information can be generated to prompt the target user to make adjustments, thereby achieving the acquisition of the to-be-processed image that meets the shooting requirements.

[0049] In practical applications, before photographing the target object part to obtain a plurality of to-be-processed images of the target object part, the method further includes: According to the preset shooting rules, shooting prompt information corresponding to different shooting angles is generated, and the shooting prompt information is displayed on the display interface, so that the target user can determine the shooting angle of the target object part according to the shooting prompt information.

[0050] Among them, the preset shooting rule can be understood as a rule for shooting the target object part, for example, the preset shooting rule can be a rule for shooting three foot images of the target user's feet from the front, side and back. Then, different shooting angles can include front angles, side angles and back angles. The shooting prompt information can be used to prompt the target user different shooting angles, for example, the shooting prompt information can prompt the target user to turn his back to the camera to shoot a back foot image, or prompt the target user to face the camera to shoot a front foot image.

[0051] Specifically, shooting prompt information corresponding to different shooting angles can be generated according to preset shooting rules, and the shooting prompt information can be displayed to the target user through a display interface, so that the target user can determine and adjust the shooting angle of the target object according to the shooting prompt information.

[0052] In actual applications, the target user's foot can be judged in real time based on the calculated shooting index data, and a foot image can be taken when it is determined that the target user's foot meets the conditions for entering the picture (i.e., the preset shooting conditions). For example, when the number of foot bone points in the shooting area is less than 1, it means that the conditions for entering the picture are not met, and a prompt message "please enter the picture of the foot" is generated; when the ratio of the calf length in the shooting area to the height of the shooting area is greater than 1, a prompt message "the distance is close" is generated; when the ratio of the calf length to the height of the shooting area is less than 2, a prompt message "the distance is far" is generated. When the difference between the horizontal coordinates of the left ankle and the horizontal coordinates of the right ankle in the shooting area is greater than 3, a shooting prompt message "please turn your back to the camera" is generated. When shooting the left foot image, if the number of left foot bone points in the shooting area is less than 1, a prompt message "please enter the picture of the left foot" is generated; when shooting the right foot image, if the number of right foot bone points in the shooting area is less than 1, a prompt message "please enter the picture of the right foot" is generated. When photographing the left foot or the right foot, if the number of foot bone points in the photographing area is greater than or equal to 1, and 1<the ratio of the calf length to the height of the photographing area<2, and the number of left foot bone points is greater than 1 or the number of right foot bone points is greater than 1, it is determined that the foot meets the picture-in condition and the left foot image or the right foot image is photographed. When photographing the heels of both feet, if the number of foot bone points in the photographing area is greater than or equal to 1, and 1<the ratio of the calf length to the height of the photographing area<2, and the number of left foot bone points is greater than 1, and the number of right foot bone points is greater than 1, and the ratio of the calf length to the height of the photographing area is less than 3, it is determined that the foot meets the picture-in condition and the back heel images of both feet are photographed.

[0053] In summary, by displaying the shooting prompt information, it is used to achieve the acquisition of the to-be-processed images at multiple different shooting angles, which facilitates the subsequent part recognition based on the to-be-processed images at multiple different shooting angles, and further ensures the accuracy of the recognition results.

[0054] Step 204: Input the multiple images to be processed into an image processing model to obtain multiple bone point information of the target object part, wherein the image processing model is trained based on a reference model, the reference model recognizes the bone point information of the target object part according to the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model.

[0055] The reference model may be a large model. When the target object is a foot, the plurality of bone point information may include a left arch point, a left ankle point, a left heel, a right arch point, a right ankle point, a right heel, and the like.

[0056] In practical applications, combined with the above Figure 3 The recognition system 300 may further include an image recognition unit 304, which may call an image processing model to recognize foot bone point information in the image to be processed.

[0057] In a specific implementation, the step of inputting the plurality of images to be processed into an image processing model to obtain information of a plurality of skeleton points of the target object comprises: Inputting the plurality of images to be processed into an image processing model in sequence to obtain target image features of the plurality of images to be processed, wherein the target image features are used to describe the bone point information of the target object part; The target image features are decoded to obtain information of multiple skeleton points of the target object.

[0058] Specifically, multiple images to be processed can be input into the image processing model in sequence to obtain target image features of each image to be processed, and the target image features of each image to be processed can be decoded to obtain multiple bone point information of the target object part.

[0059] In practical applications, the information of multiple skeleton points of the target object can be realized by marking them in the image to be processed. Then, the outputted information of multiple skeleton points can be understood as an image marked with multiple skeleton points.

[0060] For example, after obtaining the foot image, the foot image can be input into the trained image processing model to obtain left arch point 1, left arch point 2, left arch point 3, left ankle point 1, left ankle point 2, left heel, right arch point 1, right arch point 2, right arch point 3, right ankle point 1, right ankle point 2, and right heel.

[0061] In summary, by predicting skeleton points through an image processing model with a small number of model parameters, computing resources are reduced and recognition speed is improved.

[0062] In addition, the training steps of the image processing model include: Determine a first sample image, a second sample image and label bone point information corresponding to the second sample image of the target object part; The image processing model to be trained is trained according to the first sample image, the second sample image, the label skeleton point information and the reference model, until an image processing model that meets the training stop condition is obtained.

[0063] Specifically, the training process of the image processing model is similar to the model training method provided in an embodiment of the present specification described below, and will not be repeated in this embodiment of the present specification.

[0064] Step 206: Determine the recognition result of the target object part according to the plurality of skeleton point information.

[0065] Among them, the recognition result of the target object part can be understood as the recognition result of the foot shape, and the recognition result can be, for example, flat feet, valgus feet, inversion feet, high arch feet, etc.

[0066] In a specific implementation, determining the recognition result of the target object part according to the plurality of skeleton point information includes: Calculating the part index data of the target object part according to the plurality of skeleton point information; The recognition result of the target object part is determined according to the part index data.

[0067] Among them, the part index data may include the arch angle of the arch point of the foot, the ankle angle of the ankle point, etc.

[0068] In practical applications, the part index data can be calculated through multiple skeletal point information to determine whether the human foot in the foot image has flat feet, high arches, inversion, valgus and other problems. Specifically, the arch angle formed by left arch point 1, left arch point 2 and left arch point 3 can be calculated to determine whether the left foot is flat foot or high arch foot. When the arch angle is greater than 4, the recognition result of the left foot is flat foot, and when the arch angle is less than 5, the recognition result of the left foot is high arch foot. Correspondingly, it is also possible to determine whether the right foot is flat foot or high arch foot based on the three arch points of the right foot.

[0069] Alternatively, the ankle angle formed by the left heel, left ankle point 1 and left ankle point 2 may be calculated, and the left foot may be judged whether it is inversion or valgus according to the ankle angle. When the ankle angle is greater than 6, the identification result of the left foot is valgus, and when the ankle angle is less than 7, the identification result of the left foot is inversion. Correspondingly, it is also possible to judge whether the right foot is valgus or inversion according to the right heel and right ankle point.

[0070] To sum up, the above method uses an image processing model to perform skeleton point recognition on multiple images to be processed of the target object part, obtains multiple skeleton point information of the target object part, and determines the recognition result of the target object part based on the multiple skeleton point information. The target object part can be recognized only by uploading the image through the device, thereby achieving the convenience of target object part recognition and reducing the recognition cost. In addition, the image processing model can be trained using a reference model with a large number of model parameters to further improve the recognition accuracy.

[0071] The following combination Figure 4 Taking the application of the target object part recognition method provided in this specification to the human foot as an example, the target object part recognition method is further described. Figure 4A flowchart of a processing process of a method for identifying a part of a target object provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0072] Step 402 : Determine whether the target user has uploaded a foot image, if so, execute step 430 , if not, execute step 404 .

[0073] Step 404: When the target user takes images of the left foot, the rear heel, and the right foot, determine whether the target user's feet are in the picture. If so, execute steps 406, 408, and 410; if not, execute step 416.

[0074] Step 406: When the target user takes a picture of his left foot, determine whether the left foot is in the picture. If so, execute step 412; if not, execute step 418.

[0075] Step 408: When the target user takes a picture of his right foot, determine whether the right foot is in the picture. If so, execute step 412; if not, execute step 420.

[0076] Step 410: Determine whether the toes of the target user are facing backwards. If so, execute step 412; if not, execute step 422.

[0077] Step 412: Determine whether the target user is at an appropriate distance from the camera. If so, execute step 428; if not, execute step 414.

[0078] Step 414: Determine whether the target user is too far from the camera. If so, execute step 424; if not, execute step 426.

[0079] Step 416: Generate a prompt message “Please enter the picture with your feet”.

[0080] Step 418: Generate a prompt message “Please enter the picture with your left foot”.

[0081] Step 420: Generate a prompt message “Please enter the picture with your right foot”.

[0082] Step 422: Generate a prompt message "Please turn your back to the camera" Step 424: Generate prompt message "Please come closer" Step 426: Generate prompt message "Please stay away" Step 428: The foot is successfully captured.

[0083] Step 430: Perform problem recognition on the foot image to obtain a recognition result of the foot.

[0084] To summarize, the above method obtains multiple foot images by photographing the human foot, performs bone point recognition on the multiple foot images, obtains multiple bone point information of the human foot, and determines the recognition result of the human foot based on the multiple bone point information, thereby achieving convenience in foot recognition and reducing recognition costs.

[0085] See also Figure 5 , Figure 5 A flowchart of a model training method provided according to an embodiment of the present specification is shown, which specifically includes the following steps.

[0086] Step 502: Determine the first sample image, the second sample image and the label skeleton point information corresponding to the second sample image of the target object part; Step 504: training the image processing model to be trained according to the first sample image, the second sample image, the label skeleton point information and the reference model until an image processing model that meets the training stop condition is obtained.

[0087] The image processing model is used to execute a method for identifying a target object part provided in an embodiment of this specification.

[0088] Wherein, the target object part can be a human foot, the first sample image can be understood as an unlabeled training sample, and the second sample image and the labeled bone point information corresponding to the second sample image can be understood as a training sample and a training label. It can be understood that a large number of first sample images (i.e., unlabeled training data) and / or a large number of second sample images and the labeled bone point information corresponding to the second sample image (i.e., labeled training data) can be used for model training. The training stop condition can be understood as the number of training times meeting a preset number threshold and / or the model loss value meeting a preset loss value threshold.

[0089] Specifically, there is currently a lack of image processing models that can identify the arch point of the human foot. Based on this, the embodiments of this specification can train image processing models. Since the data set for model training is usually manually collected and labeled, the amount of labeled data is relatively small due to labor cost constraints. In addition, online real-time recognition is usually required for foot recognition, and the recognition speed is required to be high. Therefore, it is necessary to train a model with a small number of model parameters to improve the recognition speed. However, the ability of a model with a small number of model parameters to extract features is weaker than that of a model with a large number of model parameters, so it is necessary to use less labeled data to train an image processing model with a small number of model parameters.

[0090] Based on this, since the model with a large number of model parameters has strong expressive ability, it can quickly learn the characteristics of the labeled data and extract more robust features, thereby expressing a stronger ability to recognize the arch point of the foot. However, the model with a large number of model parameters is slow. Therefore, the model with a large number of model parameters can be used as a reference model to assist in training the image processing model, thereby adding unlabeled data pages to the training process. That is to say, the training data of the image processing model to be trained includes foot images with manually labeled arch points and foot images without labeled arch points.

[0091] In a specific implementation, the image processing model to be trained is trained according to the first sample image, the second sample image, the label skeleton point information and the reference model until an image processing model that meets the training stop condition is obtained, including: Determining a first loss value according to the first sample image and the reference model; Determining a target loss value according to the second sample image, the label skeleton point information and the reference model; The image processing model to be trained is trained according to the first loss value and the target loss value until an image processing model that meets the training stop condition is obtained.

[0092] The label skeleton point information may be understood as skeleton point annotation for the second sample image.

[0093] Specifically, a first loss value can be determined based on the first sample image and the reference model, and a target loss value can be determined based on the second sample image, the label bone point information and the reference model. The image processing model to be trained can be trained according to the first loss value and the target loss value, respectively, until an image processing model that meets the training stop conditions is obtained.

[0094] In another embodiment of the present specification, when the training data only includes an unlabeled first sample image, the image processing model to be trained can be trained only based on the first loss value until an image processing model that meets the training stop condition is obtained.

[0095] In a specific implementation, determining the first loss value according to the first sample image and the reference model includes: Inputting the first sample image into the reference model to obtain a first reference feature; Inputting the first sample image into the image processing model to be trained to obtain a first prediction feature; A first loss value is calculated based on the first reference feature and the first predicted feature.

[0096] Specifically, the first sample image can be input into the reference model to obtain a first reference feature, the first sample image can be input into the image processing model to be trained to obtain a first prediction feature, and the first loss value can be calculated based on the first reference feature and the first prediction feature.

[0097] In summary, the first reference features with strong robustness extracted by the reference model are used to promote the feature extraction capability of the image processing model to be trained, which is equivalent to increasing the training data and ensuring the processing performance of the image processing model.

[0098] In a specific implementation, calculating the first loss value according to the first reference feature and the first prediction feature includes: Calculating an initial loss value according to the first reference feature and the first prediction feature; A first loss value is calculated based on the initial loss value and the preset weight parameter.

[0099] In practical applications, the formula for calculating the initial loss value is as follows.

[0100]

[0101] in, is the initial loss value, N is the total number of first sample images, and i can be understood as the i-th of the N first sample images. is the first reference feature of the i-th first sample image, is the first prediction feature of the i-th first sample image.

[0102] The preset weight parameter can be set to increase as shown in the following formula.

[0103]

[0104] in, is the preset weight parameter, is the total number of rounds of model training, The current training round.

[0105] The formula for calculating the first loss value is as follows.

[0106]

[0107] in, is the first loss value.

[0108] In summary, since the ability of the reference model to identify the arch point gradually increases with the training process, that is, as the training time increases, the first reference feature extracted by the reference model is more robust and more beneficial to promoting the learning of the image processing model to be trained. Therefore, by setting an increasing preset weight parameter, as the training progresses, the larger the value of the preset weight parameter, the better the learning ability of the image processing model to be trained.

[0109] In a specific implementation, determining the target loss value according to the second sample image, the label skeleton point information and the reference model includes: Inputting the second sample image into the reference model to obtain a second reference feature; Inputting the second sample image into the image processing model to be trained, obtaining a second prediction feature and determining second prediction bone point information corresponding to the second sample image according to the second prediction feature; Calculating a second loss value according to the second reference feature and the second predicted feature; Calculating a third loss value according to the second predicted skeleton point information and the label skeleton point information; The target loss value is calculated according to the second loss value, the third loss value and a preset weight parameter.

[0110] It can be understood that the process of calculating the second loss value based on the second reference feature and the second prediction feature is similar to the process of calculating the first loss value mentioned above, and will not be repeated here.

[0111] In practical applications, the formula for calculating the third loss value is as follows.

[0112]

[0113] in, is the third loss value, is the second predicted skeleton point information corresponding to the i-th second sample image, is the label bone point information corresponding to the i-th second sample image.

[0114] The formula for calculating the target loss value is as follows.

[0115]

[0116] in, is the target loss value.

[0117] In addition, before training the image processing model to be trained based on the first sample image, the second sample image, the label skeleton point information and the reference model, the method further includes: Inputting the second sample image into the reference model to obtain a second reference feature, and determining second reference bone point information corresponding to the second sample image according to the second reference feature; Calculating a reference loss value according to the second reference skeleton point information and the label skeleton point information; The reference model is optimized according to the reference loss value.

[0118] In practical applications, the formula for calculating the reference loss value is as follows.

[0119]

[0120] in, is the reference loss value, is the second reference skeleton point information corresponding to the i-th second sample image.

[0121] The following combination Figure 6 , taking the application of the model training method provided in this specification in an image processing model for foot recognition as an example, the model training method is further described. Figure 6 A processing flow chart of a model training method provided by an embodiment of the present specification is shown, which specifically includes the following steps.

[0122] Step 602: Input the first sample image into the reference model to obtain the first reference feature F1.

[0123] Step 604: Input the first sample image into the image processing model to be trained to obtain the first prediction feature F2.

[0124] Step 606: Calculate a first loss value according to the first reference feature and the first predicted feature.

[0125] Specifically, the process described in step 602 to step 606 is a process of calculating the first loss value based on the unlabeled data. The process described in step 608 to step 612 is a process of calculating the second loss value based on the labeled data. It can be understood that the first loss value and the second loss value are both feature loss values, and the calculation process is similar.

[0126] Step 608: Input the second sample image into the reference model to obtain a second reference feature F1, and determine the second reference bone point information P1 corresponding to the second sample image according to the second reference feature.

[0127] Step 610: Input the second sample image into the image processing model to be trained, obtain the second prediction feature F2, and determine the second prediction skeleton point information P2 corresponding to the second sample image based on the second prediction feature.

[0128] Step 612: Calculate a second loss value based on the second reference feature and the second predicted feature.

[0129] Step 614: Calculate a third loss value based on the second predicted skeleton point information and the label skeleton point information.

[0130] Step 616: Calculate a reference loss value based on the second reference skeleton point information and the label skeleton point information.

[0131] Specifically, the reference model can be optimized according to the reference loss value. The target loss value is calculated according to the second loss value, the third loss value and the preset weight parameter. The image processing model to be trained is trained according to the first loss value and the target loss value.

[0132] In summary, through the above training process, the image processing model to be trained is trained based on the reference model to ensure the processing performance of the image processing model.

[0133] Corresponding to the above method embodiment, this specification also provides an embodiment of a device for identifying a part of a target object. Figure 7 FIG. 1 is a schematic diagram showing a structure of a target object part recognition device provided by an embodiment of the present specification. Figure 7 As shown, the device comprises: An acquisition module 702 is configured to acquire a plurality of to-be-processed images of a target object portion, wherein the plurality of to-be-processed images are taken at different angles; An input module 704 is configured to input the multiple images to be processed into an image processing model to obtain multiple bone point information of the target object part, wherein the image processing model is obtained based on reference model training, the reference model recognizes the bone point information of the target object part according to the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model; The determination module 706 is configured to determine the recognition result of the target object part according to the plurality of skeleton point information.

[0134] In an optional embodiment, the acquisition module 702 is further configured to: According to a preset shooting rule, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

[0135] In an optional embodiment, the acquisition module 702 is further configured to: In response to the image shooting instruction, displaying the shooting area on the display interface; Detecting skeleton points of the target object part in the shooting area, and calculating shooting index data of the target object part according to the skeleton points; When it is determined according to the shooting index data that the target object part meets the preset shooting condition, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

[0136] In an optional embodiment, the acquisition module 702 is further configured to: If it is determined according to the shooting index data that the target object part does not meet the preset shooting condition, generating prompt information, and displaying the prompt information on the display interface, so that the target user can adjust the target object part according to the prompt information; The step of detecting the skeleton points of the target object part in the shooting area and calculating the shooting index data of the target object part according to the skeleton points is continued.

[0137] In an optional embodiment, the acquisition module 702 is further configured to: According to the preset shooting rules, shooting prompt information corresponding to different shooting angles is generated, and the shooting prompt information is displayed on the display interface, so that the target user can determine the shooting angle of the target object part according to the shooting prompt information.

[0138] In an optional embodiment, the input module 704 is further configured to: Inputting the plurality of images to be processed into an image processing model in sequence to obtain target image features of the plurality of images to be processed, wherein the target image features are used to describe the bone point information of the target object part; The target image features are decoded to obtain information of multiple skeleton points of the target object.

[0139] In an optional embodiment, the determining module 706 is further configured to: Calculating the part index data of the target object part according to the plurality of skeleton point information; The recognition result of the target object part is determined according to the part index data.

[0140] In an optional embodiment, the device further includes a training module configured to: Determine a first sample image, a second sample image and label bone point information corresponding to the second sample image of the target object part; The image processing model to be trained is trained according to the first sample image, the second sample image, the label skeleton point information and the reference model, until an image processing model that meets the training stop condition is obtained.

[0141] To summarize, the above-mentioned device uses an image processing model to perform skeleton point recognition on multiple images to be processed of the target object part, obtains multiple skeleton point information of the target object part, and determines the recognition result of the target object part based on the multiple skeleton point information. The target object part can be recognized only by uploading the image through the device, thereby realizing the convenience of target object part recognition and reducing recognition costs. In addition, the image processing model can be trained using a reference model with a large number of model parameters to further improve recognition accuracy.

[0142] The above is a schematic scheme of a target object part identification device of this embodiment. It should be noted that the technical scheme of the target object part identification device and the technical scheme of the target object part identification method described above belong to the same concept, and the details not described in detail in the technical scheme of the target object part identification device can be referred to the description of the technical scheme of the target object part identification method described above.

[0143] Corresponding to the above method embodiment, this specification also provides a model training device embodiment, Figure 8 FIG. 1 shows a schematic diagram of a model training device provided by an embodiment of the present specification. Figure 8 As shown, the device comprises: A determination module 802 is configured to determine the first sample image, the second sample image and the label skeleton point information corresponding to the second sample image of the target object part; The training module 804 is configured to train the image processing model to be trained according to the first sample image, the second sample image, the label skeleton point information and the reference model until an image processing model that meets the training stop condition is obtained; The image processing model is used to execute the target object part recognition method provided in the first aspect of the embodiment of this specification.

[0144] In an optional embodiment, the training module 804 is further configured to: Determining a first loss value according to the first sample image and the reference model; Determining a target loss value according to the second sample image, the label skeleton point information and the reference model; The image processing model to be trained is trained according to the first loss value and the target loss value until an image processing model that meets the training stop condition is obtained.

[0145] In an optional embodiment, the training module 804 is further configured to: Inputting the first sample image into the reference model to obtain a first reference feature; Inputting the first sample image into the image processing model to be trained to obtain a first prediction feature; A first loss value is calculated based on the first reference feature and the first predicted feature.

[0146] In an optional embodiment, the training module 804 is further configured to: Calculating an initial loss value according to the first reference feature and the first prediction feature; A first loss value is calculated based on the initial loss value and the preset weight parameter.

[0147] In an optional embodiment, the training module 804 is further configured to: Inputting the second sample image into the reference model to obtain a second reference feature; Inputting the second sample image into the image processing model to be trained, obtaining a second prediction feature and determining second prediction bone point information corresponding to the second sample image according to the second prediction feature; Calculating a second loss value according to the second reference feature and the second predicted feature; Calculating a third loss value according to the second predicted skeleton point information and the label skeleton point information; The target loss value is calculated according to the second loss value, the third loss value and a preset weight parameter.

[0148] In an optional embodiment, the training module 804 is further configured to: Inputting the second sample image into the reference model to obtain a second reference feature, and determining second reference bone point information corresponding to the second sample image according to the second reference feature; Calculating a reference loss value according to the second reference skeleton point information and the label skeleton point information; The reference model is optimized according to the reference loss value.

[0149] In summary, through the above training process, the image processing model to be trained is trained based on the reference model to ensure the processing performance of the image processing model.

[0150] The above is a schematic scheme of a model training device of this embodiment. It should be noted that the technical scheme of the model training device and the technical scheme of the above-mentioned model training method belong to the same concept. For details not described in detail in the technical scheme of the model training device, please refer to the description of the technical scheme of the above-mentioned model training method. Fig. 9The structure block diagram of a computing device 900 provided according to an embodiment of the present specification is shown. The components of the computing device 900 include but are not limited to a memory 910 and a processor 920. The processor 920 is connected to the memory 910 via a bus 930, and a database 950 is used to store data.

[0151] The computing device 900 also includes an access device 940 that enables the computing device 900 to communicate via one or more networks 960. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 940 may include one or more of any type of network interface (e.g., a network interface card (NIC)) of wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a world-wide interoperability for microwave access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, a near field communication (NFC) interface, and the like.

[0152] In one embodiment of the present application, the above components of the computing device 900 and Fig. 9 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Fig. 9 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of the present application. Those skilled in the art may add or replace other components as needed.

[0153] The computing device 900 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 900 may also be a mobile or stationary server.

[0154] The processor 920 is used to execute the following computer program / instructions, which implement the steps of the above method when executed by the processor.

[0155] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computing device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0156] An embodiment of the present specification further provides a computer-readable storage medium storing a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0157] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the computer-readable storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0158] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0159] The above is a schematic solution of a computer program product of this embodiment. It should be noted that the technical solution of the computer program product and the technical solution of the above method belong to the same concept, and the details not described in detail in the technical solution of the computer program product can be referred to the description of the technical solution of the above method.

[0160] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0161] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0162] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0163] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0164] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. A method for identifying a target object part, characterized in that: include: Acquire a plurality of to-be-processed images of a target object portion, wherein the plurality of to-be-processed images are taken at different angles; Inputting the multiple images to be processed into an image processing model to obtain multiple bone point information of the target object part, wherein the image processing model is obtained based on the training of a reference model, the reference model recognizes the bone point information of the target object part according to the part image of the target object part, and the number of model parameters of the reference model is greater than the number of model parameters of the image processing model; The recognition result of the target object part is determined according to the multiple skeleton point information.

2. The method according to claim 1, characterized in that: The step of obtaining a plurality of to-be-processed images of the target object part includes: According to a preset shooting rule, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

3. The method according to claim 2, characterized in that The step of photographing the target object part according to a preset photographing rule to obtain a plurality of to-be-processed images of the target object part includes: In response to the image shooting instruction, displaying the shooting area on the display interface; Detecting skeleton points of the target object part in the shooting area, and calculating shooting index data of the target object part according to the skeleton points; When it is determined according to the shooting index data that the target object part meets the preset shooting condition, the target object part is photographed to obtain a plurality of to-be-processed images of the target object part.

4. The method according to claim 3, characterized in that After calculating the shooting index data of the target object part, the method further includes: If it is determined according to the shooting index data that the target object part does not meet the preset shooting condition, generating prompt information, and displaying the prompt information on the display interface, so that the target user can adjust the target object part according to the prompt information; The step of detecting the skeleton points of the target object part in the shooting area and calculating the shooting index data of the target object part according to the skeleton points is continued.

5. The method according to claim 3, characterized in that: Before photographing the target object part to obtain a plurality of to-be-processed images of the target object part, the method further includes: According to the preset shooting rules, shooting prompt information corresponding to different shooting angles is generated, and the shooting prompt information is displayed on the display interface, so that the target user can determine the shooting angle of the target object part according to the shooting prompt information.

6. The method according to claim 1, characterized in that The step of inputting the plurality of images to be processed into an image processing model to obtain information of a plurality of bone points of the target object comprises: Inputting the plurality of images to be processed into an image processing model in sequence to obtain target image features of the plurality of images to be processed, wherein the target image features are used to describe the bone point information of the target object part; The target image features are decoded to obtain information of multiple skeleton points of the target object.

7. The method according to claim 1, characterized in that Determining the recognition result of the target object part according to the plurality of skeleton point information includes: Calculating the part index data of the target object part according to the plurality of skeleton point information; The recognition result of the target object part is determined according to the part index data.

8. The method according to claim 1, characterized in that The training steps of the image processing model include: Determine a first sample image, a second sample image and label bone point information corresponding to the second sample image of the target object part; The image processing model to be trained is trained according to the first sample image, the second sample image, the label skeleton point information and the reference model, until an image processing model that meets the training stop condition is obtained.

9. A model training method, characterized in that: include: Determine a first sample image, a second sample image and label bone point information corresponding to the second sample image of the target object part; According to the first sample image, the second sample image, the label skeleton point information and the reference model, the image processing model to be trained is trained until an image processing model that meets the training stop condition is obtained; Wherein, the image processing model is used to execute the target object part recognition method described in claims 1-8.

10. The method according to claim 9, characterized in that The step of training the image processing model to be trained according to the first sample image, the second sample image, the label skeleton point information and the reference model until an image processing model that meets the training stop condition is obtained comprises: Determining a first loss value according to the first sample image and the reference model; Determining a target loss value according to the second sample image, the label skeleton point information and the reference model; The image processing model to be trained is trained according to the first loss value and the target loss value until an image processing model that meets the training stop condition is obtained.

11. The method according to claim 10, characterized in that The determining a first loss value according to the first sample image and the reference model includes: Inputting the first sample image into the reference model to obtain a first reference feature; Inputting the first sample image into the image processing model to be trained to obtain a first prediction feature; A first loss value is calculated based on the first reference feature and the first predicted feature.

12. The method according to claim 11, characterized in that The calculating a first loss value according to the first reference feature and the first prediction feature includes: Calculating an initial loss value according to the first reference feature and the first prediction feature; A first loss value is calculated based on the initial loss value and the preset weight parameter.

13. The method according to claim 10, characterized in that The step of determining a target loss value according to the second sample image, the label skeleton point information and the reference model comprises: Inputting the second sample image into the reference model to obtain a second reference feature; Inputting the second sample image into the image processing model to be trained, obtaining a second prediction feature and determining second prediction bone point information corresponding to the second sample image according to the second prediction feature; Calculating a second loss value according to the second reference feature and the second predicted feature; Calculating a third loss value according to the second predicted skeleton point information and the label skeleton point information; The target loss value is calculated according to the second loss value, the third loss value and a preset weight parameter.

14. The method according to claim 13, characterized in that Before training the image processing model to be trained according to the first sample image, the second sample image, the label skeleton point information and the reference model, the method further includes: Inputting the second sample image into the reference model to obtain a second reference feature, and determining second reference bone point information corresponding to the second sample image according to the second reference feature; Calculating a reference loss value according to the second reference skeleton point information and the label skeleton point information; The reference model is optimized according to the reference loss value.

15. A computing device, characterized in that: include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer program / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 14 are implemented.

16. A computer-readable storage medium, characterized in that: It stores a computer program / instruction, which implements the steps of the method described in any one of claims 1 to 14 when executed by a processor.

17. A computer program product, characterized in that The method comprises a computer program / instruction which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 14.

Images