Digital human binding evaluation method
By comparing the data similarity of the digital human model to be tested and the standard digital human model, evaluating the suitability of the binding method, and making intelligent adjustments, the problem of lack of a unified evaluation mechanism in the existing technology is solved, and the digital human is improved The quality and efficiency of animation.
Patent Information
- Application Number
- CN202311218488.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-06-01
- Filing Date
- 2023-09-20
- Publication Date
- 2024-12-03
AI Technical Summary
The existing technology lacks a unified and standard evaluation mechanism for digital human binding methods, and cannot objectively measure the suitability of the binding methods, which affects the improvement of digital human animation effects.
By obtaining the digital human model to be tested and the standard digital human model, using the target-driven method to drive the two, calculate the similarity of the data after the drive, to evaluate the binding effect, and adjust the binding point or method according to the evaluation results to achieve intelligence Optimize the binding effect of the integration.
It provides objective and standard evaluation of digital human binding methods to help find the most suitable binding methods, improve the fidelity and nature of digital human animations, and reduce the cost of manual adjustment.
Smart Images

Figure CN119068158A_ABST
Abstract
Description
[0001] This application claims the priority of the Chinese patent application filed with the China Patent Office on June 1, 2023, with application number 202310644328.6 and application name “A Digital Human Binding Evaluation Method”, all contents of which are incorporated by reference in this application. Technical Field
[0002] The present application relates to the field of computer vision technology, and in particular to a digital human binding evaluation method. Background Art
[0003] Digital human is a digital human image that is close to a real person and is synthesized by computer means. The three-dimensional (3D) digital human produced by combining computer graphics, computer vision and other technologies has a more realistic display effect and has therefore been widely used.
[0004] The production process of digital human can usually include the main steps of modeling, binding, driving and rendering. Among them, binding refers to the binding of the skeleton and expression base of the digital human model through manual or automatic algorithms, that is, binding the skeleton and expression base to the appropriate points of the body and face of the digital human model respectively, and then in the driving step, the body and face of the digital human model can be driven to move, presenting the corresponding digital human animation effect.
[0005] At present, there are endless ways to bind digital humans. Some of these binding methods are implemented manually, while others are implemented by automated algorithms. Different binding methods will bind different points on the digital human model, and the action effects presented will also be different. Generally, if the binding method used for the digital human model is more suitable, the final animation effect will be more realistic and natural. On the contrary, if the binding method used is not suitable, the final animation effect will be very poor.
[0006] However, there is no unified and standard evaluation mechanism for the binding effect of digital human binding methods in related technologies, and it is impossible to objectively measure whether the currently used binding method is suitable for the digital human model, which is not conducive to improving the digital human production effect. Summary of the invention
[0007] The embodiments of the present application provide a digital human binding evaluation method, apparatus, computing device, device cluster, computer storage medium, and computer program product, which can objectively evaluate whether the binding method used in the digital human production process is suitable.
[0008] In a first aspect, an embodiment of the present application provides a digital human binding evaluation method, the method comprising: obtaining a digital human model to be tested and a standard digital human model, the digital human model to be tested describing various points at which an object is bound by a target binding method, the object comprising a skeleton and / or an expression base, and the standard digital human model being used to characterize an evaluation standard for the binding effect of the object; driving the digital human model to be tested and the standard digital human model respectively by a target driving method to obtain test drive data and standard drive data, the test drive data being used to characterize an animation formed by the digital human to be tested being driven, and the standard drive data being used to characterize an animation formed by the standard digital human model being driven; performing a similarity evaluation on the test drive data and the standard drive data to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested by the evaluation data.
[0009] In this embodiment, after the user binds the skeleton and / or expression base of the digital human model to be tested by the target binding method, the binding evaluation can be performed on the server side together with the standard digital human model. Among them, the standard digital human model is a digital human model that has completed the skeleton and facial expression base binding and has been verified in large quantities, so as to characterize the binding effect evaluation standard of the digital human model. In this way, the same driving method can be used on the server to drive the digital human model to be tested and the standard digital human model respectively, so that the two can be moved to form corresponding animations, and the respective animations can be formed into the test drive data and the standard drive data, so as to evaluate the binding effect gap between the digital human model to be tested and the standard digital model by calculating the similarity between the test drive data and the standard drive data, that is, to evaluate the binding effect of the target binding method on the digital human model to be tested, and obtain the objective evaluation result of whether the binding method adopted in the digital human production process is suitable, which is conducive to finding a binding method suitable for the digital human to be tested (or a digital human with a similar image) in the digital human development process.
[0010] In some possible implementations, the digital human model to be tested includes a first digital human model to be tested and a second digital human model to be tested, the first digital human model to be tested and the second digital human model to be tested are created based on the same digital human image, and the first digital human model to be tested is bound by a first target binding method, and the second digital human model to be tested is bound by a second target binding method; after obtaining the digital human model to be tested and the standard digital human model, the method specifically includes: driving the first digital human model to be tested, the second digital human model to be tested, and the standard digital human model respectively by a target driving method to obtain first post-drive data to be tested of the first digital human model to be tested, second post-drive data to be tested of the second digital human model to be tested, and standard post-drive data; evaluating the similarity between the first post-drive data to be tested and the standard post-drive data to obtain first evaluation data, and evaluating the similarity between the second post-drive data to be tested and the standard post-drive data to obtain second evaluation data; and determining the binding method with the best binding effect on the digital human image from the first target binding method and the second target binding method according to the comparison result of the first evaluation data and the second evaluation data.
[0011] In this way, based on the same digital human image, different digital human models to be tested can be obtained by using different binding methods, and uploaded to the server together with the standard digital human model for driving in the same way. Then, by evaluating the similarity of the post-driving data between different digital human models to be tested and the standard digital human, the optimal binding method that best suits the current digital human image can be found from these different binding methods.
[0012] In some embodiments, performing similarity evaluation on the post-flooding data to be tested and the standard post-flooding data to obtain evaluation data includes: calculating the similarity between the post-flooding data to be tested and the standard post-flooding data to obtain a similarity value; and comparing the similarity value with a preset threshold to obtain evaluation data.
[0013] In this embodiment, the evaluation data can be obtained by calculating the similarity between the post-drive data, and the calculation logic is simple.
[0014] In some embodiments, a first point cloud is associated with the post-drive data to be tested, the first point cloud includes points described by the digital human model to be tested, and a second point cloud is associated with the standard post-drive data, the second point cloud includes points bound to the skeleton and / or expression base on the standard digital human model; similarity evaluation is performed on the post-drive data to be tested and the standard post-drive data to obtain evaluation data, including: calculating the similarity between the post-drive data to be tested and the standard post-drive data to obtain a similarity value; calculating the similarity between the first point cloud and the second point cloud to obtain point distribution difference information between the first point cloud and the second point cloud, the point distribution difference information includes a target point, and the target point is a point that makes the post-drive data to be tested different from the standard post-drive data; comparing the similarity value with a preset threshold to obtain a comparison result; comparing the comparison result with the point distribution difference information to generate evaluation data.
[0015] In this embodiment, evaluation data can be obtained based on the similarity value and point difference between the post-drive data to automatically determine the points to be adjusted, thereby facilitating intelligent adjustment of the binding effect of the digital human model to be tested.
[0016] In some embodiments, the point distribution difference information also includes a low-score factor, and the low-score factor is used to describe the cause of the difference between the post-flooding data to be tested and the standard post-flooding data.
[0017] In this way, by calculating the distribution difference of points based on point cloud similarity, the causes of the differences can also be determined, such as incorrect binding of target points, large or small amplitude of movement formed by target points, etc., so as to further realize the intelligent adjustment of the binding effect of the digital human model to be tested.
[0018] In some embodiments, the digital human model to be tested is obtained from a user terminal, and the method further includes: when the evaluation data does not reach a preset threshold, determining a target adjustment strategy from an adjustment strategy library; the adjustment strategy library includes one or more adjustment strategies, and the adjustment strategy is used to adjust the binding state of the expression base or skeleton bound to the digital human model to be tested, and the target adjustment strategy is one of the adjustment strategies in the adjustment strategy library; sending the target adjustment strategy to the user terminal to adjust the digital human model to be tested according to the target adjustment strategy.
[0019] In this way, when the evaluation result does not meet expectations, the corresponding target adjustment strategy can be matched to make adjustments according to the strategy to obtain a digital human model with higher binding quality.
[0020] In some embodiments, in the adjustment strategy library, each adjustment strategy is associated with a similarity value interval and a type of low-score factor, and the similarity value intervals and / or types of low-score factors associated with different adjustment strategies are different, wherein the similarity value interval is used to describe the numerical range of the similarity value, and the low-score factor is used to describe the cause of the difference between the post-drive data to be tested and the standard post-drive data; the evaluation data includes the similarity value, the target point and the low-score factor; determining the target adjustment strategy from the adjustment strategy library includes: matching the target adjustment strategy for adjusting the target point on the digital human model to be tested from the adjustment strategy library according to the similarity value interval to which the similarity value in the evaluation data belongs and the type of the low-score factor in the evaluation data.
[0021] In this embodiment, the adjustment strategy is associated with the similarity value interval and the low score factor type. When the evaluation data includes the similarity value and the low score factor, the target adjustment strategy can be matched in a targeted manner to obtain more accurate and objective adjustment guidance.
[0022] In some embodiments, the digital human model to be tested and the standard digital human model are driven respectively by a target driving method to obtain the driven data to be tested and the standard driven data, including: obtaining control information, the control information is at least used to describe the target driving method, a target driving model that executes the target driving method, and a target driving data source; according to the control information, in the target driving model that executes the target driving method, using the target driving data source to drive the digital human model to be tested and the standard digital human model to generate animation respectively; according to the animations generated by the digital human model to be tested and the standard digital human model, respectively obtain the driven data to be tested and the standard driven data.
[0023] In this embodiment, the digital human model can be driven by the driving mode, driving model and driving data source specified by the user to meet the personalized needs of the user.
[0024] In some embodiments, before driving the digital human model to be tested and the standard digital human model respectively through a target driving method, the method includes: obtaining control information, the control information including task information of the digital human model to be tested; generating driving recommendation items based on the task information, the driving recommendation items including at least one recommended driving method, at least one recommended driving model and at least one recommended driving data source; sending the driving recommendation items to a user terminal; obtaining driving information determined by the user terminal from the driving recommendation items, the driving information including a target driving method, a target driving model for executing the target driving method, and a target driving data source.
[0025] In this embodiment, it is also possible to automatically match appropriate driving modes, driving models and driving data sources according to the tasks of the digital human model, such as virtual anchors, game characters and other tasks, so as to improve the intelligence of the evaluation and reduce manual input.
[0026] In some embodiments, after obtaining the digital human model to be tested and the standard digital human model, the method specifically includes: in a target-driven model that executes a target-driven mode, using a first target-driven data source to drive the digital human model to be tested and the standard digital human model to obtain third post-drive data to be tested of the digital human model to be tested and first standard post-drive data of the standard digital human model; calculating the similarity between the third post-drive data to be tested and the first standard post-drive data to obtain third evaluation data; in the target-driven model, using a second target-driven data source to drive the digital human model to be tested and the standard digital human model to obtain fourth post-drive data to be tested of the digital human model to be tested and second standard post-drive data of the digital human model to be tested, the first target-driven data source and the second target-driven data source being different types of data belonging to the same driving mode; calculating the similarity between the fourth post-drive data to be tested and the second standard post-drive data to obtain fourth evaluation data; comparing the third evaluation data with the fourth evaluation data to determine the binding effect of the digital human model to be tested under the target-driven mode.
[0027] In this embodiment, various types of data sources can be used to drive the digital human model in different driving modes. The rich types of driving data sources are helpful to more accurately evaluate the binding effect of the digital human model to be tested.
[0028] In a second aspect, an embodiment of the present application provides a digital human binding evaluation device, the device comprising: an acquisition module, used to acquire a digital human model to be tested and a standard digital human model, the digital human model to be tested describes different points of skeleton and / or expression base binding through a target binding method, and the standard digital human model is used to characterize the binding effect evaluation standard; a driving module, used to drive the digital human model to be tested and the standard digital human model respectively through a target driving method, to obtain test drive data and standard drive data, the test drive data is used to characterize the animation formed by the digital human to be tested being driven, and the standard drive data is used to characterize the animation formed by the standard digital human model being driven; an evaluation module, used to perform a similarity evaluation between the test drive data and the standard drive data to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested through the evaluation data.
[0029] In some possible implementations, the digital human model to be tested includes a first digital human model to be tested and a second digital human model to be tested, the first digital human model to be tested and the second digital human model to be tested are created based on the same digital human image, and the first digital human model to be tested is bound by a first target binding method, and the second digital human model to be tested is bound by a second target binding method; the driving module can be specifically used to drive the first digital human model to be tested, the second digital human model to be tested, and the standard digital human model respectively by a target driving method to obtain first post-drive data to be tested of the first digital human model to be tested, second post-drive data to be tested of the second digital human model to be tested, and standard post-drive data; the evaluation module can be specifically used to: evaluate the similarity between the first post-drive data to be tested and the standard post-drive data to obtain first evaluation data, and evaluate the similarity between the second post-drive data to be tested and the standard post-drive data to obtain second evaluation data; the evaluation module can also be used to determine the binding method with the best binding effect on the digital human image from the first target binding method and the second target binding method according to the comparison result of the first evaluation data and the second evaluation data.
[0030] In some embodiments, the evaluation module may be specifically used to calculate the similarity between the post-drive data to be tested and the standard post-drive data to obtain a similarity value; and compare the similarity value with a preset threshold to obtain evaluation data.
[0031] In some embodiments, a first point cloud is associated with the post-drive data to be tested, the first point cloud includes points described by the digital human model to be tested, and a second point cloud is associated with the standard post-drive data, the second point cloud includes points bound to the skeleton and / or expression base on the standard digital human model; the evaluation module can be specifically used to calculate the similarity between the post-drive data to be tested and the standard post-drive data to obtain a similarity value; and calculate the similarity between the first point cloud and the second point cloud to obtain point distribution difference information between the first point cloud and the second point cloud, the point distribution difference information includes a target point, and the target point is a point that makes the post-drive data to be tested different from the standard post-drive data; the evaluation module can also be used to compare the similarity value with a preset threshold to obtain a comparison result; and compare the comparison result with the point distribution difference information to generate evaluation data.
[0032] In some embodiments, the point distribution difference information also includes a low-score factor, and the low-score factor is used to describe the cause of the difference between the post-flooding data to be tested and the standard post-flooding data.
[0033] In some embodiments, the digital human model to be tested is obtained from a user terminal, and the evaluation module can also be used to: when the evaluation data does not reach a preset threshold, determine a target adjustment strategy from an adjustment strategy library; the adjustment strategy library includes one or more adjustment strategies, and the adjustment strategy is used to adjust the binding state of the expression base or skeleton bound to the digital human model to be tested, and the target adjustment strategy is one of the adjustment strategies in the adjustment strategy library; send the target adjustment strategy to the user terminal to adjust the digital human model to be tested according to the target adjustment strategy.
[0034] In some embodiments, in the adjustment strategy library, each adjustment strategy is associated with a similarity value interval and a type of low-score factor, and the similarity value intervals and / or types of low-score factors associated with different adjustment strategies are different, wherein the similarity value interval is used to describe the numerical range of the similarity value, and the low-score factor is used to describe the cause of the difference between the post-drive data to be tested and the standard post-drive data; the evaluation data includes similarity values, target points and low-score factors; the evaluation module can be specifically used to: match the target adjustment strategy for adjusting the target point on the digital human model to be tested from the adjustment strategy library according to the similarity value interval to which the similarity value in the evaluation data belongs and the type to which the low-score factor in the evaluation data belongs.
[0035] In some embodiments, the driving module can be specifically used to: obtain control information, the control information is at least used to describe the target driving mode, the target driving model that executes the target driving mode, and the target driving data source; according to the control information, in the target driving model that executes the target driving mode, use the target driving data source to drive the digital human model to be tested and the standard digital human model to generate animations; according to the animations generated by the digital human model to be tested and the standard digital human model, respectively, obtain the post-drive data to be tested and the post-drive data to be standard.
[0036] In some embodiments, the acquisition module can also be used to: acquire control information, the control information includes task information of the digital human model to be tested; generate driving recommendation items according to the task information, the driving recommendation items include at least one recommended driving mode, at least one recommended driving model and at least one recommended driving data source; and send the driving recommendation items to the user terminal. In addition, the acquisition module can also be used to: acquire driving information determined by the user terminal from the driving recommendation items, the driving information includes a target driving mode, a target driving model for executing the target driving mode, and a target driving data source.
[0037] In some embodiments, the driving module can be specifically used for: in a target-driven model that executes a target-driven mode, using a first target-driven data source to drive a digital human model to be tested and a standard digital human model, and obtaining third post-drive data to be tested of the digital human model to be tested and first standard post-drive data of the standard digital human model; the evaluation module can be specifically used for: calculating the similarity between the third post-drive data to be tested and the first standard post-drive data, and obtaining third evaluation data; the driving module can also be specifically used for: in a target-driven model, using a second target-driven data source to drive a digital human model to be tested and a standard digital human model, and obtaining fourth post-drive data to be tested of the digital human model to be tested and second standard post-drive data of the digital human model to be tested, the first target-driven data source and the second target-driven data source being different types of data belonging to the same driving mode; the evaluation module can also be specifically used for: calculating the similarity between the fourth post-drive data to be tested and the second standard post-drive data, and obtaining fourth evaluation data; and comparing the third evaluation data with the fourth evaluation data to determine the binding effect of the digital human model to be tested in the target-driven mode.
[0038] In a third aspect, an embodiment of the present application provides a computing device, comprising: at least one memory for storing programs; and at least one processor for executing programs stored in the memory; wherein, when the program stored in the memory is executed, the processor is used to execute the method described in the first aspect or any possible implementation of the first aspect.
[0039] In a fourth aspect, an embodiment of the present application provides a computing device cluster, which includes at least one computing device, each computing device including a processor and a memory; the processor of at least one computing device is used to execute instructions stored in the memory of at least one computing device, so that the computing device cluster executes the method described in the first aspect or any possible implementation of the first aspect.
[0040] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.
[0041] In a sixth aspect, an embodiment of the present application provides a computer program product. When the computer program product runs on a processor, the processor executes the method described in the first aspect or any possible implementation of the first aspect.
[0042] It can be understood that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1It is a schematic diagram of an application scenario of performing digital human binding assessment provided by an embodiment of the present application;
[0044] Figure 2 It is a flowchart of a digital human binding evaluation method provided in an embodiment of the present application;
[0045] Figure 3 is a flowchart of a digital human binding evaluation method in a specific example of the present application;
[0046] Figure 4 It is a flowchart of a digital human binding evaluation method in a specific embodiment of the present application;
[0047] Figure 5 It is a flowchart of a digital human binding evaluation method in a specific embodiment of the present application;
[0048] Figure 6 is a schematic diagram of an adjustment strategy library provided in an embodiment of the present application;
[0049] Figure 7 It is a flowchart of a digital human binding evaluation method in a specific embodiment of the present application;
[0050] Figure 8 is a schematic diagram of an algorithm library provided in an embodiment of the present application;
[0051] Fig. 9 It is a flowchart of a digital human binding evaluation method in a specific embodiment of the present application;
[0052] Fig.10 It is a flowchart of a digital human binding evaluation method in a specific embodiment of the present application;
[0053] Fig.11 It is a structural schematic diagram of a digital human binding evaluation device provided in an embodiment of the present application;
[0054] Fig.12 This is a structural diagram of a digital human binding evaluation device in a specific embodiment of the present application.
[0055] Fig.13 is a schematic diagram of the structure of a computing device provided in an embodiment of the present application;
[0056] Fig.14 It is a structural diagram of a device cluster provided in an embodiment of the present application;
[0057] Fig.15 It is a structural diagram of a device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION
[0058] The term "and / or" in this article is a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The symbol " / " in this article indicates that the associated objects are in an or relationship, for example, A / B means A or B.
[0059] The terms "first" and "second" in the specification and claims herein are used to distinguish different objects rather than to describe a specific order of the objects. For example, a first response message and a second response message are used to distinguish different response messages rather than to describe a specific order of the response messages.
[0060] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.
[0061] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc.; multiple elements refer to two or more elements, etc.
[0062] To facilitate understanding of the technical solution of the present application, the terms involved in this document are explained below.
[0063] Digital human is a technology that uses computer technology to synthesize digital human beings, creating a digital human image that is close to real people. Digital human can be divided into two types: two-dimensional digital human and three-dimensional digital human.
[0064] Three-dimensional (3D) digital human, 3D digital human is an important form of digital human, which is mainly achieved through the combination of different technologies such as computer graphics and computer vision. The production process of 3D digital human includes several processes such as modeling, binding, driving, and rendering. Generally, two-dimensional (2D) digital human is saved in video format, while 3D digital human data is saved in animation format. The head of 3D digital human can be in the form of expression base or mesh vertex, etc. The data of 3D digital human is mostly modeled through such abstract data.
[0065] Rigging: Mapping key points to the digital human model through manual or automated algorithms. The number and position of key point bindings affect the movement effect of the digital human. Rigging is a key step in making the digital human move. A good rigging algorithm not only makes the digital human move more realistically and accurately, but also reduces the pressure of subsequent digital human processes. The rigging algorithm is very important for the final digital human drive. To put it simply, a puppet is a system that is manipulated like a digital human. The rigging step is to determine the position of the puppet string, and the subsequent drive is to control these positions to make the digital human move.
[0066] In order to make the digital human perform animation, it is necessary to set up a skeleton system on the digital human model and bind the skeleton to the key points of the digital human model, so that the posture and movement of the digital human can be controlled by operating the skeleton. The quality of skeleton binding will affect the smoothness of the digital human's movement.
[0067] Blend shapes binding: Blend shapes is an animation method used for facial expressions of animated characters. It can deform the "base mesh" into the "target shape" through vertex interpolation. Blend shapes binding is a very important binding method for facial binding (controlling facial movements) in 3D digital humans. A good binding method can correctly assign blend shapes to appropriate facial points based on a real person image. If the quality of blend shapes binding is not good, the facial points of the modeled 3D digital human will deviate greatly from the real person image, which will not conform to the actual facial movements of the real person image.
[0068] Drive: The bound digital human is animated through voice or visual signals. The driven digital human can move like a real person, so the animation effect of the driven digital human can be evaluated through subjective and objective algorithms.
[0069] In the related technologies, 3D digital humans of different styles such as realistic, cartoon, and aesthetic can be produced. Different styles of digital humans can have multiple binding methods, which can be implemented manually or by automated algorithms. However, in the related technologies, there is no objective and standard mechanism to evaluate whether the binding method used in the production process of digital humans is appropriate, or to evaluate which binding method has better binding effect, which is not conducive to improving the production effect of digital humans.
[0070] In order to achieve an objective evaluation of the digital human binding method, a digital human binding evaluation method is provided in the embodiment of the present application. The method mainly drives the digital human model to be tested, which is bound by the target binding method, and the digital human model bound by the standardization using the same driving algorithm, and obtains the driving data of the two digital human models (which can be animation, video, image, etc.). Then the driving data is similarly calculated to determine the gap between the digital human model to be tested and the digital human model bound by the standardization. In this way, based on the gap, it can be evaluated whether the target binding method makes the animation effect presented by the digital human model to be tested close to the digital human model bound by the standardization, and then determine whether the target binding method is suitable for the digital human model to be tested, so as to achieve an objective, standard and unified binding effect evaluation.
[0071] To facilitate understanding of the technical solution of the embodiment of the present application, an application scenario of the embodiment of the present application is introduced below.
[0072] For example, Figure 1 The diagram shows an application scenario of an embodiment of the present application. Figure 1 As shown, the binding engineer can bind the skeleton or facial expression base of the 3D digital human model 100 on the user terminal 10. It can be understood that the facial expression of the 3D digital human model 100 can be presented by the changes of eyebrows, eyes, nose, mouth, chin, cheeks and other parts, so the points of different parts of the face of the 3D digital human model 100 can be bound to the corresponding expression bases to form a facial expression. In addition, the points of different parts of the body of the 3D digital human model 100 (such as shoulders, arms, waist, legs, etc.) can be bound to the corresponding positions of the skeleton to form the movements of the body parts, such as bending of the elbows and twisting of the waist joints.
[0073] In this example, at least one binding method (i.e., target binding method) is used on the user terminal 10 to bind the skeleton or facial expression base of the 3D digital human model 100. Different target binding methods can bind the skeleton and / or facial expression base to different points on the 3D digital human model 100. After binding is completed using each target binding method, a digital human model 101 to be tested can be obtained.
[0074] Next, the binder can obtain a standard digital human model 102 on the user terminal 10 side. The standard digital human model 102 can be a digital human model that has completed skeleton and facial expression binding and has been extensively verified. Therefore, the standard digital human model 102 can represent the evaluation standard of the digital human model binding effect.
[0075] Then, the binder can upload the digital human model 101 to be tested and the standard digital human model 102 to the server 20. The server 20 uses the same driving model to drive the two digital human models 101 and 102 respectively, and obtains the data of the two digital human models 101 and 102 after being driven (also referred to as "driven data" in this article). The driven data represents the animation presented after the face, body, etc. of the corresponding digital human model are moved. The driven data can be in image format, animation format or video format, but is not limited to this. Then, the server 20 performs similarity calculation on the two driven data to obtain a similarity value. The larger the similarity value, the closer the binding effect of the digital human model 101 to be tested is to the standard digital human model 102, and the higher the suitability of the target binding method for the digital human model 101 to be tested. Conversely, the smaller the similarity value, the lower the suitability.
[0076] Thus, the server 20 can realize objective, unified and standardized evaluation of digital human binding methods, which is conducive to finding a better binding method for digital human model binding, thereby facilitating improving the binding effect in the digital human production process.
[0077] It is understood that the user terminal 10 may be a smart phone, a personal computer, etc., but is not limited thereto. The server 20 may be a hardware server, a cloud server, or a virtualized device (such as a virtual machine, a container, etc.).
[0078] It can be understood that the digital human in this article is sometimes also called virtual human, virtualized digital human, etc. Digital human can specifically refer to a digital character image created using digital technology that is close to the human image (or anthropomorphic image).
[0079] In addition, in some digital animal production processes, the evaluation method of this example can also be used to evaluate the binding method used when performing bone binding on digital animal models.
[0080] Next, based on the above description, a digital human binding evaluation method provided by an embodiment of the present application is introduced. It can be understood that the method is proposed based on the above description, and part or all of the contents of the method can be referred to the above description.
[0081] See also Figure 2 , Figure 2 : is a flowchart of a digital human binding evaluation method provided in an embodiment of the present application. It can be understood that the method can be executed by any device, equipment, platform, or device cluster with computing and processing capabilities. Figure 2 As shown, the method may include:
[0082] S210, obtaining a digital human model to be tested and a standard digital human model.
[0083] In this embodiment, the digital human model to be tested obtained by the server can be a digital human model that has completed skeleton binding, or expression base binding, or both skeleton and expression base binding through the target binding method. The standard digital human model obtained by the server is a digital human model that has completed skeleton and facial expression base binding and has been extensively verified, so as to characterize the binding effect evaluation standard of the digital human model. As an example, the server can be a hardware server, a cloud server, or other virtualized devices, such as a virtual machine, a container, etc. As a specific example, the server can be Figure 1 The server 20 shown in FIG. 2 is also taken as an example below.
[0084] For example, the digital human model to be tested and the standard digital human model can be the same, similar, or both digital human models have some specific image features. Figure 3 As shown in (3a), the face of the digital human model 101 to be tested is bound to multiple expression bases 1 through a target binding method. A standard digital human model 102 with a similar image and also bound to these expression bases 1 can be used to evaluate the facial binding effect of the digital human model 101 to be tested.
[0085] In this example, different target binding methods can bind the skeleton and / or expression base to different points on the same digital human model, so the digital human model to be tested 101 is described with various points bound to the skeleton and / or various expression bases. In this example, the target binding method can be manually implemented or an automatically executed binding algorithm (such as the skeleton automatic binding plug-in of meta Human Creator (an application), the facial binding plug-in of maya (an application), but not limited thereto), which is not limited here.
[0086] Exemplarily, the digital human model 101 to be tested and the standard digital human model 102 may both be uploaded from the user terminal 10 to the server 20. In other examples, the digital human model 101 to be tested may be uploaded from the user terminal 10 to the server 20, and the standard digital human model 102 may be stored on the server 20 for invocation. But not limited thereto.
[0087] S220, driving the digital human model to be tested and the standard digital human model respectively in a target driving manner to obtain the driven data to be tested and the standard driven data.
[0088] In this embodiment, combined with Figure 3As shown in (3b), the server 20 can drive the digital human model 101 to be tested and the standard digital human model 102 using the same target drive mode X1, so that the digital human model 101 to be tested and the standard digital human model 102 are animated, and the corresponding animations are presented and recorded, and the recorded data is the post-drive data. For the convenience of explanation, the post-drive data of the digital human model 101 to be tested is referred to as the post-drive data V1 to be tested, and the post-drive data of the standard digital human model 102 is referred to as the standard post-drive data V2. Among them, the post-drive data V1 to be tested represents the animation formed by the digital human model 101 to be tested being driven, and the standard post-drive data V2 is used to represent the animation formed by the standard digital human model 102 being driven.
[0089] Exemplarily, the post-drive data V1 and V2 may be image sequences, animations or videos, but are not limited thereto. By way of example and not limitation, image sequences may include images in formats such as bmp, jpg, png, and tif, animations may include formats such as GIF, FLIC, and SWF, and videos may include formats such as AVI, MOV, FLV, RMVB, and WMV.
[0090] Exemplarily, the driving mode of any digital human model may include visual driving, voice driving, and text driving, but is not limited thereto. Among them, the principle of video driving may include: capturing the actions of real people through motion capture devices such as cameras, or shooting a video recording real people's actions, or a preset set of action sequences, etc., and migrating the logic of these actions to the digital human model, so that the digital human model presents corresponding actions. The principle of voice driving may include: driving the face of the digital human model to present corresponding lip shape or facial expression through audio data. The principle of text driving may include: driving the face of the digital human model to present corresponding lip shape or facial expression according to text data.
[0091] S230 , performing similarity evaluation on the post-drive data to be tested and the post-drive data to be standard to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested through the evaluation data.
[0092] In this embodiment, combined with Figure 3 As shown in (3c), the server 20 can determine the similarity of the animation formed between the digital human model 101 to be tested and the standard digital human model 102 by calculating the similarity between the post-processing data V1 to be tested and the standard post-processing data V2. For example, if the post-processing data V1 to be tested and the standard post-processing data V2 are image sequences, the similarity between the two image sequences is calculated; if the post-processing data V1 to be tested and the standard post-processing data V2 are in video format, the similarity between the two videos is calculated.
[0093] Exemplarily, when calculating the similarity between the tested post-drive data V1 and the standard post-drive data V2, the calculation is performed according to the time series attribute of the data.
[0094] For example, the post-drive data V1 to be tested is a sequence including n images, and the standard post-drive data V2 is a sequence including n images, and each image in each sequence records a frame of animation formed by the corresponding digital human model. Then, the similarity can be calculated between the first image in the post-drive data V1 to be tested and the first image in the standard post-drive data V2, and the similarity can be calculated between the second image in the post-drive data V1 to be tested and the second image in the standard post-drive data V2, ... In this way, n similarity values can be obtained, and the average of the n similarity values is used as the evaluation data V3 to evaluate the animation similarity between the digital human model 101 to be tested and the standard digital human model 102, that is, to evaluate the gap between the binding effect presented by the target binding method on the digital human model 101 to be tested and the binding effect evaluation standard, so as to determine whether the target binding method is suitable for the digital human model 101 to be tested.
[0095] For example, if the evaluation data V3 reaches the preset threshold, it can be determined that the binding effect of the digital human model 101 to be tested after being bound by the target binding method reaches the expected result, and the evaluation is passed. The target binding method can be used as the default binding method for the digital human model 101 to be tested (or a digital human model with a similar image), and the digital human model 101 to be tested can enter the subsequent rendering and other production processes. On the contrary, if the evaluation fails, the user (who can be a binder or an evaluator) can adjust the binding points of the digital human model 101 to be tested or rebind it with other binding methods.
[0096] In this way, by evaluating the digital human binding effect through an objective, unified and standard evaluation method, it is helpful to determine whether the current binding method is suitable for the digital model 101 to be tested, which is not only conducive to making adjustments when the current binding method is not suitable, but also conducive to efficiently determining the most suitable binding method to complete the binding when binding digital human models with the same or similar images, so as to improve the efficiency and quality of digital human production.
[0097] The following describes a digital human binding evaluation method in a specific embodiment of the present application in conjunction with the accompanying drawings.
[0098] See also Figure 4 , Figure 4 FIG. 1 is a flowchart of a digital human binding evaluation method provided by a specific embodiment of the present application. Figure 4 As shown, the method may include:
[0099] S410: The server obtains the digital human model to be tested and the standard digital human model.
[0100] In this embodiment, the digital human model to be tested 101 and the standard digital human model 102 acquired by the server 20 may be uploaded by the user terminal 10 .
[0101] Exemplarily, different binding methods bind different points on the 3D digital human model. For example, when performing expression base binding on the mouth of a 3D digital human model of a film and television character SS, target binding method A can bind the expression base to 52 points out of 100 points at the mouth, and target binding method B can bind the expression base to 55 points out of 100 points at the mouth. In this way, the two binding methods A and B will eventually make the 3D digital human model present different mouth animation effects. In this embodiment, after any target binding method is used to perform skeleton and / or expression base binding operations on the 3D digital human model, the obtained digital human model 101 to be tested is described with information on different points bound to the skeleton and / or each expression base.
[0102] Exemplarily, the standard digital human model 102 can be a digital human model of a universal character image with good generalization, or it can be a pre-made digital human model of a specific image. For example, the standard digital human model 102 is a digital human model of an Asian image, a digital human model of a certain animation or film character, or a digital human model close to a real person image, etc.
[0103] In some specific examples, a model library can be pre-built, and the model library includes several standard digital human models 102. The images of the standard digital human models 102 in the model library can be the same or different, and any standard digital human model 102 can be a model that completes skeleton binding, a model that completes expression base binding, or a model that completes skeleton and expression base binding. In this example, the model library can be deployed on a server (such as server 20, but not limited to this), and then the user terminal 10 can request to obtain a corresponding standard digital human model 102 from the server as the evaluation standard of the digital human model 101 to be tested. Alternatively, the model library can be deployed on the user terminal 10 for the user terminal 10 to call.
[0104] In this embodiment, the digital human model 101 to be tested has similar appearance features to the standard digital human model 102. After the user creates a digital human model 102 to be tested on the user terminal 10, a corresponding standard digital human model 102 can be selected and sent to the server 20. For example, the user obtains a digital human model 101 to be tested with an Asian image through the target binding method A on the user terminal 10, and a standard digital human model 102 with an Asian image can be selected as the binding effect evaluation standard of the digital human model 101 to be tested.
[0105] In this embodiment, the server 20 may also obtain a corresponding evaluation request from the user terminal 10 , and in response to the evaluation request, evaluate the binding effect of the digital human model 101 to be tested according to the standard digital human model 102 .
[0106] After obtaining the digital human model 101 to be tested and the standard digital human model 102 in S410, the server further executes:
[0107] S420, driving the digital human model to be tested and the standard digital human model respectively through the target driving mode to obtain the driven data to be tested and the driven data of the standard.
[0108] In this step, S420 may specifically include:
[0109] S421, obtain control information.
[0110] In this embodiment, the server 20 can interact with the user terminal 10 to obtain control information input by the user to perform drive control. As an example, the control information is used to describe one or more of the following information:
[0111] The task information, target driving mode, target driving model, target driving data source, evaluation location, etc. of the digital human model to be tested, but not limited thereto.
[0112] Exemplarily, the task information describes the purpose of the digital human model to be tested, such as being used as a virtual anchor, virtual customer service, virtual character, etc.
[0113] Exemplarily, the target drive mode can be but not limited to voice drive, video drive or text drive. The target drive model is a model for executing the target drive mode. In this example, each drive mode can be executed by multiple drive models, and a target drive model is determined from these drive models. The target drive data source is audio data, video data, or text data input into the target drive model to execute the target drive mode to drive the corresponding digital human model.
[0114] Exemplarily, the evaluation part may include the face or parts of the face (such as the mouth, eyes, cheeks, etc.), and may also include the body or parts of the body (such as hands, arms, legs, feet, etc.). The user may specify the binding effect evaluation for a certain evaluation part.
[0115] In some possible implementations, when the control information includes a target driving mode, a target driving model, and a target driving data source, the server 20 may further execute:
[0116] S422, calling a target drive model for executing a target drive mode according to the control information;
[0117] S423, in the target-driven model, using the target-driven data source to respectively drive the digital human model to be tested and the standard digital human model to generate corresponding animations.
[0118] In this example, the server 20 can call a specified target drive model according to the control information to drive the digital human model 101 to be tested and the standard digital human model 102, and use the target drive data source to animate the two digital human models 101 and 102 to generate corresponding animations. For example, the server 20 calls the voice drive mode to associate a voice drive model M1, and uses the specified 100 audio data as the target drive data source, and voice drives the two digital human models 101 and 102 respectively through the voice drive model M1, so that the faces of the two digital human models 101 and 102 present the lip shape (or overall facial expression) animation corresponding to the 100 audio data.
[0119] It is understandable that the target binding method can bind the face or body of the digital human, and the adopted target driving model also drives the corresponding parts that have completed the binding.
[0120] S424, obtaining the post-drive data to be tested and the standard post-drive data according to the animations generated by the digital human model to be tested and the standard digital human model respectively.
[0121] In this step, the target driving model drives the face or body of the two digital human models 101, 102, and can output animation format or video format files of the digital human models 101, 102 respectively. These files can be used as post-driving data, and can also be further formed into image sequences as post-driving data.
[0122] In this example, the post-drive data V1 to be tested and the standard post-drive data V2 are data of the same format, which is convenient for subsequent comparative calculation. In some other possible examples, the post-drive data V1 to be tested and the standard post-drive data V2 can also be data of different formats, for example, the post-drive data V1 to be tested is an image, and the standard post-drive data V2 is a video, and the standard post-drive data V2 can be decomposed into several frames of images for similarity calculation based on image time sequence attributes.
[0123] S430, performing similarity evaluation on the post-drive data to be tested and the post-drive data of the standard to obtain evaluation data.
[0124] In this example, S430 may specifically include:
[0125] S431, calculating the similarity between the post-drive data to be tested and the post-drive data of the standard to obtain a similarity value.
[0126] In this step, the post-driving data to be tested and the standard post-driving data can be similarly calculated according to the time series attribute. For example, the post-driving data to be tested V1 is a sequence of n images, and the standard post-driving data V2 is a sequence of n images. The first image of the post-driving data to be tested V1 is similar to the first image in the standard post-driving data V, and the second image of the post-driving data to be tested V1 is similar to the second image in the standard post-driving data V, and so on.
[0127] Exemplarily, there may be multiple similarity values calculated between the post-drive data V1 to be tested and the standard post-drive data V2, and the value range of each similarity value is [0, 1]. When there are multiple similarity values, the similarity between the post-drive data V1 to be tested and the standard post-drive data V2 may be represented by the average of all similarity values. It can be understood that the value range of the average of the similarity values is also [0, 1].
[0128] As an example, the similarity calculation may be performed by calculating cosine distance, Euclidean distance, etc., but is not limited thereto.
[0129] In some other possible examples, similarity can also be evaluated in a subjective manner. For example, an evaluation dimension is predefined, and multiple people score two post-drive data based on the evaluation dimension. The difference in the scores of the two post-drive data can be regarded as a similarity value.
[0130] S432, comparing the similarity value with a preset threshold to obtain evaluation data.
[0131] In this step, the mean value of the similarity value outputted in step S431 (or the similarity value obtained by scoring) can be compared with the preset similarity threshold, and the difference obtained by comparison can be used as the evaluation data V3. If the similarity value represented by the evaluation data V3 is lower than the similarity threshold, it indicates that the target binding method makes the animation effect formed by the digital human model 101 to be tested and the animation effect of the standard digital human model 102 have a large difference, that is, the target binding method has a poor binding effect on the digital human model 101 to be tested. On the contrary, if the similarity value represented by the evaluation data V3 is higher than the similarity threshold, it indicates that the animation effect formed by the digital human model 101 to be tested using the target binding method has a small difference with the animation effect of the standard digital human model 102, and the binding effect of the digital human model 101 to be tested by the target binding method meets the standard and meets expectations. If the average similarity value is 1, it means that the bound digital human model 101 to be tested is perfectly consistent with the standard digital human model 102. If the average similarity is 0, it means that the digital human model 101 to be tested is completely inconsistent with the standard digital human model 102, and the binding effect of the target binding method on the digital human model 101 to be tested is very poor. The target binding method is not suitable for the digital human model 101 to be tested, and the evaluation fails.
[0132] In this way, it is evaluated whether the binding method used by the digital human model 101 to be tested meets the evaluation standard and whether the binding effect meets the expectation, so as to determine whether the binding method is suitable for the digital human model 101 to be tested. If the binding effect of the digital human model 101 to be tested meets the expectation, the digital human model 101 to be tested can continue the subsequent rendering and other production processes. If the binding method is not suitable and the binding effect does not meet the expectation, the digital human model 101 to be tested can be adjusted in time until a better binding method with a satisfactory binding effect is found.
[0133] Moreover, in this embodiment, this objective and unified evaluation method is applicable to the binding effects of different parts of the digital human model, including the evaluation of the binding methods of the head and body. In some examples, the evaluation results can be obtained through automatic calculation or manual scoring, which has high generalization.
[0134] Exemplarily, when it is determined through S430 that the evaluation data of the digital human model 101 to be tested does not reach a preset threshold (ie, a similarity threshold), the server 20 may further execute:
[0135] S440, determining a target adjustment strategy from the adjustment strategy library, and returning the target adjustment strategy to the user terminal.
[0136] In this example, the server 20 may deploy an adjustment strategy library, in which a number of adjustment strategies Y1, Y2, Y3, ... may be configured, each of which is used to adjust the binding state of the expression base or skeleton bound to the digital human model to be tested. The binding state may include the range of motion formed by the bound points, but is not limited thereto.
[0137] In some possible implementations, when the evaluation data V3 does not reach the preset threshold, the operation and maintenance personnel can manually analyze the points that need to be adjusted or the movement amplitude of the points that need to be adjusted on the digital human model 101 to be tested, and input instructions to the server 20, so that the server 20 calls a target adjustment strategy from the adjustment strategy library and returns it to the user terminal 10, and then adjusts the digital human model 101 to be tested on the user terminal 10 according to the target adjustment strategy.
[0138] S450, repeating the above-mentioned processing from S423 to S440 on the adjusted digital human model to be tested and the standard digital human model until the binding effect of the digital human model to be tested reaches the expected level.
[0139] In this embodiment, the server 20 repeats the above-mentioned loop processing from S410 to S440 on the adjusted digital human model to be tested and the standard digital human model, i.e., loop processing of a series of operations such as driving, evaluating, and adjusting when the evaluation does not meet expectations, until the loop reaches the upper limit of the number of times or the evaluation data V3 of the adjusted digital human model to be tested 101 meets expectations. Then, the adjusted digital human model to be tested that passes the evaluation is input into the subsequent production process such as rendering.
[0140] Next, a digital human binding evaluation method in a specific embodiment of the present application is described.
[0141] See also Figure 5 , Figure 5 FIG. 1 is a flowchart of a digital human binding evaluation method provided by a specific embodiment of the present application. Figure 5 As shown, the method may include:
[0142] S510: The server obtains a digital human model to be tested and a standard digital human model.
[0143] S520, driving the digital human model to be tested and the standard digital human model respectively in a target driving manner to obtain the driven data to be tested and the standard driven data.
[0144] In this embodiment, the execution principle of steps S510 to S520 is similar to that of steps S410 to 420 in the above embodiment. For details, please refer to the above description of steps S410 to 420, which will not be repeated here.
[0145] S530, performing similarity evaluation on the post-drive data to be tested and the post-drive data of the standard to obtain evaluation data.
[0146] In this embodiment, the post-processing data V1 to be tested can be associated with a corresponding point cloud (referred to herein as the first point cloud S1), the first point cloud S1 includes the points described by the digital human model to be tested, and the standard post-processing data V2 is also associated with a corresponding point cloud (referred to herein as the second point cloud S2), the second point cloud S2 includes the points on the standard digital human model that are bound to the skeleton and / or expression base. For example, the post-processing data V1 to be tested and the standard post-processing data V2 are both videos recording the facial animation of the digital human, the post-processing data V1 to be tested is associated with 1000 points on the face of the digital human model 101 to be tested that are bound to the expression base 1, and the standard post-processing data V2 is associated with 1050 points on the face of the standard digital human model 102 that are bound to the expression base 1.
[0147] Then, step S530 may specifically include:
[0148] S531, calculating the similarity between the post-drive data to be tested and the post-drive data of the standard to obtain a similarity value.
[0149] In this step, the execution principle is similar to that of step S430 in the above embodiment and will not be described in detail.
[0150] S532, calculating the similarity between the first point cloud and the second point cloud, and obtaining point distribution difference information between the first point cloud and the second point cloud, wherein the point distribution difference information includes a target point and a low-score factor, wherein the target point is a point that causes a difference between the post-drive data to be tested and the standard post-drive data, and the low-score factor is used to describe a cause of a difference between the post-drive data to be tested and the standard post-drive data.
[0151] In this step, since there may be differences in facial contour, distribution of facial features, body shape, etc. between the digital human model 101 to be tested and the standard digital human model 102, the difference in animation effect between the digital human model 101 to be tested and the standard digital human model 102 can be regarded as the difference in different movement amplitudes produced by the same image (or the same shape). Therefore, the target point where the test post-drive data V1 and the standard post-drive data V2 differ, and the cause of the difference between the test post-drive data V1 and the standard post-drive data V2 can be determined by calculating the distribution difference of the binding points related to the test post-drive data V1 and the standard post-drive data V2. As an example, the cause may include incorrect point binding, small movement amplitude formed by the point, large movement amplitude formed by the point, etc., but is not limited to this.
[0152] For examples, please continue to refer to Figure 5 . When the digital human model 101 to be tested is driven to obtain a 10-second speaking video, 52 points of the mouth of the digital human model 101 to be tested are associated (or marked), forming a point cloud S1, while when the standard digital human model 102 is driven to obtain a 10-second speaking video, 55 points of the mouth of the standard digital human model 102 are associated (or marked), forming a point cloud S2. The similarity between the point cloud S1 and the point cloud S2 is calculated (for example, the Euclidean distance between the point clouds S1 and S2 is calculated, but not limited thereto), and the distribution difference of the points associated with the two speaking videos is obtained, and it is determined which points of the mouth of the digital human model 101 to be tested are incorrectly bound, or which points form a larger or smaller movement amplitude than the binding effect of the mouth of the standard digital human model 102, thereby obtaining the points to be adjusted (i.e., the target points) and the corresponding low-score factors. It can be understood that the similarity between point clouds can be calculated by any suitable mature algorithm, and this example is not limited to it.
[0153] S533, comparing the similarity value with a preset threshold to obtain a comparison result.
[0154] In this step, the similarity value between the post-driving data V1 to be tested and the post-driving data V2 to be standard is compared with a preset similarity threshold to obtain a comparison result.
[0155] S534, compare the results with the point distribution difference information to generate evaluation data
[0156] In this step, the comparison result is combined with the point distribution difference information to generate evaluation data of the digital human model 101 to be tested.
[0157] S540: Determine a target adjustment strategy from the adjustment strategy library and return it to the user terminal.
[0158] In this example, combined with Figure 6 As shown, in the adjustment strategy library 2 deployed on the server 20, each adjustment strategy Y1, Y2, Y3, ... can be associated with a similarity value interval and a type of low score factor. Figure 6 As shown, the types of low-score factors may include but are not limited to: incorrect point binding (first cause), small amplitude of point-formed movement (second cause), large amplitude of point-formed movement (third cause), etc. In addition, multiple similarity value intervals (such as Figure 6 The first interval, the second interval, ...) shown in the figure are used to describe different similarity value ranges, and each similarity value interval is lower than the above similarity threshold. Each similarity value interval is respectively combined with each type of low-scoring factor and then associated with an adjustment strategy Y1, Y2, Y3, ..., to establish a similar Figure 5 The mapping relationship is shown. Moreover, the adjustment point action amplitudes are different between different adjustment strategies involving the adjustment action amplitude.
[0159] In this way, the server 20 matches the target adjustment strategy for adjusting the target point on the digital human model 101 to be tested according to the similarity value interval to which the similarity value in the evaluation data V3 belongs and the low score factor, and returns it to the user terminal 10, and adjusts the binding status of the target point (including whether it is bound, the amplitude of the action generated after binding (which can be achieved by adjusting the weight of the target point) automatically or manually, or by changing the binding method, to form a new digital human model 101 to be tested.
[0160] S550, repeating the above-mentioned processing from S520 to S540 on the adjusted digital human model to be tested and the standard digital human model until the binding effect of the digital human model to be tested reaches the expected level.
[0161] In this step, the execution principle is similar to that of step S450 in the above embodiment and will not be described in detail.
[0162] In some specific embodiments, the server 20 can also automatically match a suitable driving mode in combination with the application scenario of the digital human model 101 to be tested, so as to achieve a more accurate and objective binding effect evaluation. Figure 7A schematic diagram of the process of a digital human binding evaluation method provided in a specific embodiment of the present application is shown. Figure 7 As shown, the digital human binding evaluation method may include, when executed on the server 20:
[0163] S610: Obtain a digital human model to be tested and a standard digital human model.
[0164] The implementation principle of this step can refer to the description of S410 in the above embodiment, which will not be repeated here.
[0165] 621, obtaining control information, the control information including task information;
[0166] S622: Generate a driving recommendation item according to the control information.
[0167] In this embodiment, combined with Figure 8 As shown, the algorithm library 3 can be pre-deployed on the server 20, and the algorithm library 3 includes a variety of driving models, driving data sources, etc. for driving the face or body of a digital human. The server 20 can match suitable driving recommendations from the algorithm library 33 in combination with the task information in the control information.
[0168] For example, Figure 8 As shown, the algorithm library 3 can be divided into body drive and face drive. Among them, body drive can include video drive mode X01 and voice drive mode X02, and face drive can include video drive mode X11, voice drive mode X12 and text drive mode X13, and each drive mode is associated with multiple drive models and multiple groups of drive data sources. For example, the drive data source of the video drive mode X01 can include multiple groups of video data X01a, X01b, ..., and the drive data source of the video drive mode X11 can include multiple groups of video data X11a, X11b, ..., and these video data can be different types of action sequences / videos such as real-life dance movements and martial arts movements. The drive data source of the voice drive mode X02 includes multiple groups of voice data X02a, X02b, ..., and the drive data source of the voice drive mode X12 includes multiple groups of voice data X12a, X12b, ..., and these voice data can be different types of voice data such as broadcast voice, singing voice, Chinese audio, English audio, etc. The text data X13a X13b, ..., etc. can include data with different text contents.
[0169] It can be understood that there can be multiple driving models capable of video driving, voice driving, and text driving, and all of them can be mature models in the field. For example, voice driving of the face can be performed through a voice operated character animation (VOCA) driving model, or through a Mesh talk voice driving model, or through an Emo talk voice driving model, etc. Or, for example, video driving of the body can be performed through a motion capture model based on carrying location information in full frames into Human Pose and Shape Estimation (CLIFF), and text driving of the face can be performed through a text driving model TTS (Text To Speech & Animation).
[0170] In this example, the possible test range of each digital human model to be tested can be covered by the driving mode, driving model and driving data source included in the algorithm library 3.
[0171] In this embodiment, the server 20 can match the driving recommendation items from the algorithm library 3 according to the task information and binding parts (or designated evaluation parts) of the digital human model 101 to be tested, including recommended driving methods, recommended driving models, recommended driving data sources, etc. For example, the digital human model 101 to be tested describes the binding points of the expression base, and will perform singing as a virtual anchor. Then, the server 20 can match the voice driving method X12 of the face drive as the recommended driving method, and use the broadcasting voice X12a and singing voice X12b under the driving method X12 as the recommended driving data source, and use the multiple driving models X121, X122, ... under the driving method X12 as the recommended driving models. As a specific example, each driving recommendation item can also be divided into the best recommendation item, the second best recommendation item, etc. according to the attributes such as the driving model performance.
[0172] S623, sending the driver recommendation item to the user terminal;
[0173] S624, obtaining a target driving mode, a target driving model, and a target driving data source determined by the user terminal according to the driving recommendation item.
[0174] In this embodiment, the server 20 sends the driving recommendation items to the user terminal, and the user selects a target driving mode, a target driving model, and a target driving data source from the driving recommendation items and returns them to the server 20 .
[0175] When the digital human model 101 to be tested needs to make richer and more complex voices, expressions, and actions during application, the target drive data source may include multiple groups. For ease of explanation, the first target drive data source and the second target drive data source may be taken as an example. The first target drive data source and the second target drive data source belong to different types of data under the same drive mode. Then, the digital human binding evaluation method may also include:
[0176] S625, in the target-driven model executing the target-driven mode, the digital human model to be tested and the standard digital human model are driven by the first target-driven data source to obtain the third tested driven data of the digital human model to be tested and the first standard driven data of the standard digital human model.
[0177] In this embodiment, first, in the target drive model, the first target drive data source is used to drive the digital human model 101 to be tested and the standard digital human model 102 respectively, to obtain the post-drive data of the digital human model 101 to be tested, i.e., the third post-drive data to be tested, and the post-drive data of the standard digital human model 102, i.e., the first standard post-drive data. Fig. 9 As shown in (9a) and (9b), 100 broadcast voices X12a are used to drive two digital human models 101 and 102 in the VOCA driving model respectively, generating audio and video (i.e., multimedia files containing sound and image) V1a and V2a of the two digital human models 101 and 102 speaking.
[0178] S626, calculating the similarity between the third post-driving data to be tested and the first standard post-driving data to obtain third evaluation data.
[0179] In this step, based on the principle similar to step S430 in the above embodiment, the similarity between the third post-driving data to be tested and the first standard post-driving data can be calculated to obtain corresponding evaluation data, ie, third evaluation data V3a.
[0180] S627, in the target driving model, using the second target driving data source to drive the digital human model to be tested and the standard digital human model, to obtain fourth driven data to be tested of the digital human model to be tested and second standard driven data of the digital human model to be tested.
[0181] In this example, in the target drive model, the second target drive data source is used to drive the digital human model 101 to be tested and the standard digital human model 102 to obtain the post-drive data of the digital human model 101 to be tested, i.e., the fourth post-drive data to be tested, and the post-drive data of the standard digital human model 102, i.e., the second standard post-drive data. Fig. 9As shown in (9a) and (9b), after obtaining the audio and video V1a and V2a, 100 singing voices X12b are used to drive the two digital human models 101 and 102 respectively in the VOCA driving model to generate the audio and video V1b and V2b of the two digital human models 101 and 102 singing.
[0182] S628, calculating the similarity between the fourth tested post-driving data and the second standard post-driving data to obtain fourth evaluation data.
[0183] In this step, based on the principle similar to step S430 in the above embodiment, the similarity between the fourth tested post-driving data and the second standard post-driving data can be calculated to obtain corresponding evaluation data, ie, fourth evaluation data V3b.
[0184] S630: Compare the third evaluation data with the fourth evaluation data to determine the binding effect of the digital human model to be tested in the target driving mode.
[0185] In this step, continue to refer to Fig. 9 As shown in (9c), after obtaining the third evaluation data V3a and the fourth evaluation data V3b, the animation effect presented by the digital human model 101 under the drive of different types of driving data sources can be evaluated accordingly, so as to determine the standard degree of the digital human model 101 under the test when presenting rich and complex facial or body animations, and determine the binding effect of the digital human model under the target driving mode.
[0186] In this way, by controlling variables in this way, the digital human model 101 to be tested is evaluated for multiple types of driving data sources under the same driving mode, so as to obtain a more accurate and reliable evaluation result.
[0187] S640: When the third evaluation data or the fourth evaluation data does not reach the preset threshold, a corresponding target adjustment strategy is determined from the adjustment strategy library and returned to the user terminal.
[0188] In this step, when any evaluation data does not reach a preset threshold (which can be set by the user according to task information), the target adjustment strategy corresponding to the evaluation data is matched from the algorithm library 3 to adjust the binding state of the expression base or skeleton bound to the target point of the digital human model 101 to be tested.
[0189] Since the digital human model 101 to be tested needs to make richer and more complex voices, expressions, and movements, the various points on the digital human model 101 to be tested that are bound to the expression base or skeleton may present different degrees of binding effects under the drive of different data sources. Therefore, in this example, the evaluation data V3 obtained from different data sources can be adjusted by controlling variables to improve the production quality of the digital human model 101 to be tested.
[0190] For example, the above 100 broadcast voices and 100 singing voices are still used as data sources. After evaluation, the post-drive data obtained based on the audio and video V1a and V2a are calculated by similarity, and the score data V3a of the digital human model 101 to be tested is determined to be 0.87, and the evaluation is qualified. The post-drive data obtained based on the audio and video V1b and V2b are calculated by similarity, and the score data V3b of the digital human model 101 to be tested is determined to be 0.45, and the evaluation is unqualified. In addition, after analyzing the audio and video V1b and V2b through mature machine learning technology, it is determined that the lip shape of the phonemes "zh" and "z" in the singing animation generated by the digital human model 101 to be tested is incorrect, and the low score reason is that the target points P1 and P2 of the mouth are not correctly bound. Then, the server 20 can first match the target adjustment strategy for one or all target points according to the score and the low score factor, and guide the user terminal 10 to adjust the corresponding target points.
[0191] S650, repeating the above-mentioned processing from S625 to S640 on the adjusted digital human model to be tested and the standard digital human model until the binding effect of the digital human model to be tested reaches the expected level.
[0192] In this way, the binding effect of the digital human model 101 to be tested can be comprehensively evaluated through various types of driving data sources, thereby obtaining a unified, objective, accurate and reliable evaluation result. Adaptive adjustments can be made based on the evaluation results to achieve a higher quality binding effect. Then, the adjusted digital human model to be tested that has passed the evaluation is input into subsequent production processes such as rendering.
[0193] In some specific embodiments, the server 20 may also bind a digital human model 101 to be tested by using multiple target binding methods, and evaluate these digital human models 101 to be tested to efficiently determine the target binding method that is most suitable for the digital human model 101 to be tested. Specifically, Fig.10 As shown, the digital human binding evaluation method can execute:
[0194] S1010, obtaining a first digital human model to be tested, a second digital human model to be tested, and a standard digital human model.
[0195] In this step, the server 20 can obtain multiple digital human models 101 to be tested. For ease of description, the first digital human model 101a to be tested and the second digital human model 101b to be tested are taken as examples. The first digital human model 101a to be tested and the second digital human model 101b to be tested are created based on the same digital human image. The first digital human model 101a to be tested uses the expression base and / or skeleton binding completed by the first target binding method, and the second digital human model 101b to be tested uses the expression base and / or skeleton binding completed by the second target binding method. It can be understood that the expression base and skeleton bound to the first digital human model 101a to be tested and the second digital human model 101b to be tested are the same, but the binding points are different.
[0196] S1020, driving each digital human model to be tested and the standard digital human model respectively through a target driving mode, obtaining first driven data of the first digital human model to be tested, second driven data of the second digital human model to be tested, and standard driven data.
[0197] In this step, it can be executed based on the principle similar to step S420 in the above embodiment to obtain the first driven data V11 to be tested of the first digital human model 101a to be tested, the second driven data V12 to be tested of the second digital human model 101b to be tested, and the standard driven data V2 of the standard digital human model 102. The specific driving process will not be repeated here.
[0198] S1030, evaluating the similarity between the standard post-flooding data and the first post-flooding data to be tested to obtain first evaluation data, and evaluating the similarity between the standard post-flooding data and the second post-flooding data to be tested to obtain second evaluation data.
[0199] In this embodiment, the first post-drive data V11 to be tested of the first digital human model 101a to be tested is calculated similarly with the standard post-drive data V2 to obtain the first evaluation data V31 of the first digital human model 101a to be tested, so as to evaluate the binding effect of the first target binding method on the first digital human model 101a to be tested. In addition, the second post-drive data V12 to be tested of the second digital human model 101b to be tested is calculated similarly with the standard post-drive data V2 to obtain the second evaluation data V32 of the second digital human model 101b to be tested, so as to evaluate the binding effect of the second target binding method on the second digital human model 101b to be tested. The specific principles of similarity calculation and evaluation can be referred to the description of step S430 in the above embodiment.
[0200] S1040: Determine, based on the comparison result of the first evaluation data and the second evaluation data, a binding method with the best binding effect on the digital human image from the first target binding method and the second target binding method.
[0201] In this embodiment, the server 20 can determine the evaluation data that reaches the preset threshold and has the largest value from the evaluation data V11 and V12 of the first digital human model 101a and the second digital human model 101b, that is, the highest evaluation data. Then, the target binding method used by the digital human model to be tested corresponding to the highest evaluation data is used as the optimal binding method for the current digital human image.
[0202] For example, the first digital human model 101a to be tested and the second digital human model 101b to be tested are both created based on the same digital human image F. If the evaluation data V11 is higher than the evaluation data V12, and the evaluation data V11 reaches the preset threshold, then the first target binding method is a binding method that is more suitable for the digital human image F. The first digital human model 101a to be tested can enter the subsequent production process, and when the digital human image F or a digital human with a similar image is produced again, the first target binding method can be used for binding. In this way, it is helpful to assist users in selecting a binding method suitable for certain digital human images during the digital human development process, thereby achieving more efficient and high-quality binding production.
[0203] The present application also provides a digital human binding evaluation device, such as Fig.11 As shown, the digital human binding evaluation device 1100 may include:
[0204] The acquisition module 1101 is used to acquire a digital human model to be tested and a standard digital human model. The digital human model to be tested describes different points of skeleton and / or expression base binding by a target binding method, and the standard digital human model is used to characterize a binding effect evaluation standard.
[0205] The driving module 1102 is used to drive the digital human model to be tested and the standard digital human model respectively through the target driving mode to obtain the test driven data and the standard driven data. The test driven data is used to represent the animation formed by the digital human to be tested being driven, and the standard driven data is used to represent the animation formed by the standard digital human model being driven.
[0206] The evaluation module 1103 is used to perform similarity evaluation on the post-drive data to be tested and the standard post-drive data to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested through the evaluation data.
[0207] Exemplary, combined Fig.12As shown, the acquisition module 1101 of the digital human binding evaluation device 1100 can obtain the digital human model 101 to be tested and the standard digital human model 102 from the user terminal 10, and input them into the driving module 1102 to drive them in the same target driving mode, so that they move to form an animation, and obtain the corresponding test drive data V1 and standard drive data V2. Then the evaluation module 1103 performs similarity calculation, and the obtained similarity value is compared with the preset similarity threshold value to obtain the final evaluation data. If the evaluation data meets expectations, the target binding method used by the digital human 101 to be tested is set as the default binding method of the digital human 101 to be tested or a digital human with a similar image to the digital human 101 to be tested, and the digital human 101 to be tested is input into the subsequent rendering and other production processes. On the contrary, if the evaluation fails, the corresponding information is returned to the user terminal 10 to make it bind the digital human 101 to be tested again until the binding evaluation passes.
[0208] Among them, the acquisition module 1101, the driving module 1102 and the evaluation module 1103 can all be implemented by software, or can be implemented by hardware. Exemplarily, the following takes the acquisition module 1101 as an example to introduce the implementation of the acquisition module 1101. Similarly, the implementation of the driving module 1102 and the evaluation module 1103 can refer to the implementation of the acquisition module 1101.
[0209] As an example of a software functional unit, the acquisition module 1101 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the acquisition module 1101 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including a data center or multiple data centers with similar geographical locations. Among them, usually a region may include multiple AZs.
[0210] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.
[0211] As an example of a hardware functional unit, the acquisition module 1101 may include at least one computing device, such as a server, etc. Alternatively, the acquisition module 1101 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.
[0212] The multiple computing devices included in the acquisition module 1101 can be distributed in the same region or in different regions. The multiple computing devices included in the acquisition module 1101 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the acquisition module 1101 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.
[0213] It should be noted that, in other embodiments, the acquisition module 1101 can be used to execute any step in the digital human binding evaluation method, the driving module 1102 can be used to execute any step in the digital human binding evaluation method, and the evaluation module 1103 can be used to execute any step in the digital human binding evaluation method. The steps that the acquisition module 1101, the driving module 1102, and the evaluation module 1103 are responsible for implementing can be specified as needed. The acquisition module 1101, the driving module 1102, and the evaluation module 1103 respectively implement different steps in the digital human binding evaluation method to realize all the functions of the digital human binding evaluation device.
[0214] The present application also provides a computing device 100. Fig.13 As shown, the computing device 100 includes: a bus 102, a processor 104, a memory 106, and a communication interface 108. The processor 104, the memory 106, and the communication interface 108 communicate with each other through the bus 102. The computing device 100 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 100.
[0215] The bus 102 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.13 The bus 104 is represented by only one line, but does not mean that there is only one bus or one type of bus. The bus 104 may include a path for transmitting information between various components of the computing device 100 (eg, the memory 106, the processor 104, and the communication interface 108).
[0216] The processor 104 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).
[0217] The memory 106 may include a volatile memory, such as a random access memory (RAM). The processor 104 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).
[0218] The memory 106 stores executable program codes, and the processor 104 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module 1101, the driving module 1102, and the evaluation module 1103, thereby implementing the digital human binding evaluation method. That is, the memory 106 stores instructions for executing the digital human binding evaluation method.
[0219] The communication interface 103 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 100 and other devices or a communication network.
[0220] The embodiment of the present application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smart phone.
[0221] like Fig.14 As shown, the computing device cluster includes at least one computing device 100. The memory 106 in one or more computing devices 100 in the computing device cluster may store the same instructions for executing the digital human binding evaluation method.
[0222] In some possible implementations, the memory 106 of one or more computing devices 100 in the computing device cluster may also store partial instructions for executing the digital human binding evaluation method. In other words, the combination of one or more computing devices 100 may jointly execute instructions for executing the digital human binding evaluation method.
[0223] It should be noted that the memory 106 in different computing devices 100 in the computing device cluster can store different instructions, which are respectively used to execute part of the functions of the digital human binding evaluation device. That is, the instructions stored in the memory 106 in different computing devices 100 can realize the functions of one or more modules among the acquisition module 1101, the driving module 1102 and the evaluation module 1103.
[0224] In some possible implementations, one or more computing devices in the computing device cluster may be connected via a network, which may be a wide area network or a local area network. Fig.15 A possible implementation is shown. Fig.15 As shown, two computing devices 100A and 100B are connected via a network. Specifically, the network is connected via a communication interface in each computing device. In this type of possible implementation, the memory 106 in the computing device 100A stores instructions for executing the functions of the acquisition module 1101. At the same time, the memory 106 in the computing device 100B stores instructions for executing the functions of the driving module 1102 and the evaluation module 1103.
[0225] Fig.15 The connection method between the computing device clusters shown may be that considering that the digital human binding evaluation method provided in this application requires a large amount of data storage and calculation, it is considered that the functions implemented by the driving module 1102 and the evaluation module 1103 are handed over to the computing device 100B for execution.
[0226] It should be understood that Fig.15The functions of the computing device 100A shown in FIG. 1 may also be completed by multiple computing devices 100. Similarly, the functions of the computing device 100B may also be completed by multiple computing devices 100.
[0227] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similar to that of Fig.14 and Fig.15 The connection mode of the computing device cluster is different in that the memory 106 in one or more computing devices 100 in the computing device cluster may store the same instructions for executing the digital human binding evaluation method.
[0228] The embodiment of the present application also provides a computer program product including instructions. The computer program product may be a software or program product including instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes the digital human binding evaluation method.
[0229] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute the digital human binding evaluation method.
[0230] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.
Claims
1. A digital human binding evaluation method, characterized in that: The method comprises: Acquire a digital human model to be tested and a standard digital human model, wherein the digital human model to be tested describes various points at which an object is bound by a target binding method, wherein the object includes a skeleton and / or an expression base, and the standard digital human model is used to characterize an evaluation standard for a binding effect on the object; By means of target driving, the digital human model to be tested and the standard digital human model are driven respectively to obtain the test driven data and the standard driven data, wherein the test driven data is used to characterize the animation formed by the digital human to be tested being driven, and the standard driven data is used to characterize the animation formed by the standard digital human model being driven; A similarity evaluation is performed on the post-drive data to be tested and the standard post-drive data to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested through the evaluation data.
2. The method according to claim 1, characterized in that The digital human model to be tested includes a first digital human model to be tested and a second digital human model to be tested, wherein the first digital human model to be tested and the second digital human model to be tested are created based on the same digital human image, and the first digital human model to be tested is bound by a first target binding method, and the second digital human model to be tested is bound by a second target binding method; After obtaining the digital human model to be tested and the standard digital human model, the method specifically includes: By means of the target driving method, the first digital human model to be tested, the second digital human model to be tested, and the standard digital human model are driven respectively to obtain first driven data to be tested of the first digital human model to be tested, second driven data to be tested of the second digital human model to be tested, and the standard driven data; Evaluate the similarity between the first post-flood data to be tested and the standard post-flood data to obtain first evaluation data, and evaluate the similarity between the second post-flood data to be tested and the standard post-flood data to obtain second evaluation data; According to the comparison result of the first evaluation data and the second evaluation data, a binding method with the best binding effect on the digital human image is determined from the first target binding method and the second target binding method.
3. The method according to claim 1 or 2, characterized in that: The performing similarity evaluation on the post-drive data to be tested and the post-drive data to be tested to obtain evaluation data includes: Calculating the similarity between the post-flooding data to be tested and the post-flooding data of the standard to obtain a similarity value; The similarity value is compared with a preset threshold to obtain the evaluation data.
4. The method according to claim 1 or 2, characterized in that: The test post-drive data is associated with a first point cloud, the first point cloud includes points described by the test digital human model, and the standard post-drive data is associated with a second point cloud, the second point cloud includes points on the standard digital human model that are bound to bones and / or expression bases; The performing similarity evaluation on the post-drive data to be tested and the post-drive data to be tested to obtain evaluation data includes: Calculating the similarity between the post-flooding data to be tested and the post-flooding data of the standard to obtain a similarity value; Calculating the similarity between the first point cloud and the second point cloud to obtain point distribution difference information between the first point cloud and the second point cloud, wherein the point distribution difference information includes a target point, and the target point is a point that causes a difference between the post-drive data to be tested and the standard post-drive data; Comparing the similarity value with a preset threshold to obtain a comparison result; The comparison result and the point distribution difference information are combined to generate the evaluation data.
5. The method according to claim 4, characterized in that The point distribution difference information also includes a low-score factor, and the low-score factor is used to describe the cause of the difference between the post-flooding data to be tested and the post-flooding data of the standard.
6. The method according to any one of claims 3 to 5, characterized in that: The digital human model to be tested is obtained from a user terminal, and the method further includes: In the case that the evaluation data does not reach the preset threshold, a target adjustment strategy is determined from an adjustment strategy library; the adjustment strategy library includes one or more adjustment strategies, and the adjustment strategy is used to adjust the binding state of the expression base or the skeleton bound to the digital human model to be tested, and the target adjustment strategy is one of the adjustment strategies in the adjustment strategy library; The target adjustment strategy is sent to the user terminal to adjust the digital human model to be tested according to the target adjustment strategy.
7. The method according to claim 6, characterized in that In the adjustment strategy library, each adjustment strategy is associated with a similarity value interval and a type of low-score factor, and the similarity value intervals and / or types of low-score factors associated with different adjustment strategies are different, wherein the similarity value interval is used to describe the numerical range of the similarity value, and the low-score factor is used to describe the cause of the difference between the post-flooding data to be tested and the standard post-flooding data; The evaluation data includes similarity value, target point and low score factor; Determining the target adjustment strategy from the adjustment strategy library includes: According to the similarity value interval to which the similarity value in the evaluation data belongs and the type to which the low-scoring factor in the evaluation data belongs, a target adjustment strategy for adjusting the target point on the digital human model to be tested is matched from the adjustment strategy library.
8. The method according to any one of claims 1 to 7, characterized in that: The method of driving the digital human model to be tested and the standard digital human model respectively by target driving to obtain the driven data to be tested and the standard driven data includes: Acquire control information, where the control information is used to describe at least the target-driven mode, a target-driven model for executing the target-driven mode, and a target-driven data source; According to the control information, in the target-driven model executing the target-driven mode, the target-driven data source is used to drive the digital human model to be tested and the standard digital human model to generate animations respectively; The to-be-tested post-drive data and the standard post-drive data are obtained respectively according to the animations generated by the to-be-tested digital human model and the standard digital human model.
9. The method according to any one of claims 1 to 7, characterized in that: Before driving the digital human model to be tested and the standard digital human model respectively by target driving, the method includes: Acquiring control information, wherein the control information includes task information of the digital human model to be tested; Generate a driving recommendation item according to the task information, wherein the driving recommendation item includes at least one recommended driving mode, at least one recommended driving model, and at least one recommended driving data source; Sending the driving recommendation item to a user terminal; The driving information determined by the user terminal from the driving recommendation items is acquired, where the driving information includes the target driving mode, a target driving model for executing the target driving mode, and a target driving data source.
10. The method according to any one of claims 1 to 7, characterized in that: After obtaining the digital human model to be tested and the standard digital human model, the method specifically includes: In the target-driven model executing the target-driven mode, the digital human model to be tested and the standard digital human model are driven by the first target-driven data source to obtain the third tested driven data of the digital human model to be tested and the first standard driven data of the standard digital human model; Calculating the similarity between the third test post-flooding data and the first standard post-flooding data to obtain third evaluation data; In the target drive model, the second target drive data source is used to drive the digital human model to be tested and the standard digital human model to obtain fourth driven data of the digital human model to be tested and second standard driven data of the digital human model to be tested, wherein the first target drive data source and the second target drive data source are different types of data belonging to the same drive mode; Calculating the similarity between the fourth test post-drive data and the second standard post-drive data to obtain fourth evaluation data; The third evaluation data and the fourth evaluation data are compared to determine the binding effect of the digital human model to be tested under the target driving mode.
11. A digital human binding evaluation device, characterized in that: The device comprises: An acquisition module, used to acquire a digital human model to be tested and a standard digital human model, wherein the digital human model to be tested describes different points of skeleton and / or expression base binding by a target binding method, and the standard digital human model is used to characterize a binding effect evaluation standard; A driving module, used to drive the digital human model to be tested and the standard digital human model respectively through a target driving mode, and obtain the test driven data and the standard driven data, wherein the test driven data is used to characterize the animation formed by the digital human to be tested being driven, and the standard driven data is used to characterize the animation formed by the standard digital human model being driven; The evaluation module is used to perform similarity evaluation on the post-drive data to be tested and the standard post-drive data to obtain evaluation data, so as to characterize the binding effect of the target binding method on the digital human model to be tested through the evaluation data.
12. A computing device cluster, characterized in that: comprising at least one computing device, each computing device comprising a processor and a memory; The processor of the at least one computing device is used to execute instructions stored in the memory of the at least one computing device, so that the computing device cluster executes the method according to any one of claims 1-10.
13. A computer program product comprising instructions, characterized in that When the instructions are executed by a computing device cluster, the computing device cluster executes the method according to any one of claims 1 to 10.
14. A computer-readable storage medium, characterized in that: The method comprises computer program instructions. When the computer program instructions are executed by a computing device cluster, the computing device cluster executes the method according to any one of claims 1 to 10.
Citation Information
Cited By
Digital human quality assessment method and device
CN120726537A
Digital Human Quality Assessment Method and Device
CN120726537B
