Object processing method, data processing method and object processing model training method

Abstract

The invention provides an object processing method, a data processing method and an object processing model training method, and the object processing method comprises the steps: determining target skeleton data corresponding to a plurality of object parts of a target object; inputting the target skeleton data into an object processing model to obtain associated skeleton data corresponding to an associated object of the target object, the associated object being an object worn by the target object, the associated skeleton data being determined according to sub-associated skeleton data corresponding to at least one sub-associated object, the sub-associated skeleton data is determined according to fused skeleton features corresponding to an object part set, and the fused skeleton features are determined according to target skeleton data corresponding to object parts contained in the object part set; and generating the associated object according to the associated skeleton data. The object generation method can be widely applied to the fields of digital culture product manufacturing software, digital culture creative software, digital culture creative design and the like in the digital creative technology.

Description

Technical Field

[0001] This manual pertains to the fields of digital cultural and creative activities and artificial intelligence technology, with a particular focus on object processing methods, data processing methods, and object processing model training methods. Background Technology

[0002] With the development of game technology, clothing is a key element in shaping the image of 3D characters (such as game characters, film and animation characters, and virtual humans). By dressing 3D characters in various outfits, realism and visual expressiveness are achieved. Typically, to avoid clipping issues, clothing skeletons are used as motion control fulcrums to achieve coordinated movement between clothing and character actions. However, currently, creating clothing skeletons usually requires manually setting their positions, which is labor-intensive, time-consuming, heavily reliant on human experience, and has low generation efficiency. Therefore, an effective technical solution is urgently needed to address these problems. Summary of the Invention

[0003] In view of the above, embodiments of this specification provide an object processing method. This specification also relates to an object processing apparatus, a data processing method, a data processing apparatus, an object processing model training method, an object processing model training apparatus, a computing device, a computer-readable storage medium, and a computer program product, to solve the aforementioned problems existing in the prior art.

[0004] According to a first aspect of the embodiments of this specification, an object processing method is provided, including: Determine the target skeleton data corresponding to multiple object parts of the target object; The target skeleton data is input into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object. The associated object is the object worn by the target object. The associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-associated object. The sub-associated skeleton data is determined based on the fused skeleton features corresponding to the object part set. The fused skeleton features are determined based on the target skeleton data corresponding to the object parts included in the object part set. The associated object is generated based on the associated skeletal data.

[0005] According to a second aspect of the embodiments of this specification, an object processing apparatus is provided, comprising: The determination module is configured to determine the target skeleton data corresponding to multiple object parts of the target object; The input module is configured to input the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object, wherein the associated object is the object worn by the target object, the associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-associated object, the sub-associated skeleton data is determined based on the fused skeleton features corresponding to the object part set, and the fused skeleton features are determined based on the target skeleton data corresponding to the object parts included in the object part set; The generation module is configured to generate the associated object based on the associated skeletal data.

[0006] According to a third aspect of the embodiments of this specification, a data processing method is provided, comprising: Determine the target skeleton data corresponding to multiple parts of the target virtual 3D character; The target skeleton data is input into the object processing model to obtain the associated skeleton data corresponding to the clothing object of the target virtual 3D character. The associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-clothing object. The sub-associated skeleton data is determined based on the fused skeleton features corresponding to the character part set. The fused skeleton features are determined based on the target skeleton data corresponding to the character parts included in the character part set. The clothing object is generated based on the associated skeletal data.

[0007] According to a fourth aspect of the embodiments of this specification, a data processing apparatus is provided, comprising: The determination module is configured to determine the target skeleton data corresponding to multiple parts of the target virtual 3D character; The input module is configured to input the target skeletal data into the object processing model to obtain the associated skeletal data corresponding to the clothing object of the target virtual 3D character. The associated skeletal data is determined based on the sub-associated skeletal data corresponding to at least one sub-clothing object. The sub-associated skeletal data is determined based on the fused skeletal features corresponding to the character part set. The fused skeletal features are determined based on the target skeletal data corresponding to the character parts included in the character part set. The generation module is configured to generate the clothing object based on the associated skeletal data.

[0008] According to a fifth aspect of the embodiments of this specification, a method for training an object processing model is provided, comprising: Obtain sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to sample associated objects, wherein the sample associated objects are the objects worn by the sample target object; The sample target skeleton data is input into the object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object. The predicted associated skeleton data is determined based on the sample sub-associated skeleton data corresponding to at least one sample sub-associated object. The sample sub-associated skeleton data is determined based on the sample fused skeleton features corresponding to the object part set. The sample fused skeleton features are determined based on the sample target skeleton data corresponding to the object parts included in the object part set. The object processing model is trained based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stops, thus obtaining a fully trained object processing model.

[0009] According to a sixth aspect of the embodiments of this specification, an object processing model training apparatus is provided, comprising: The acquisition module is configured to acquire sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to sample associated objects, wherein the sample associated objects are objects worn by the sample target object; The input module is configured to input the sample target skeleton data into the object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object. The predicted associated skeleton data is determined based on the sample sub-associated skeleton data corresponding to at least one sample sub-associated object. The sample sub-associated skeleton data is determined based on the sample fusion skeleton features corresponding to the object part set. The sample fusion skeleton features are determined based on the sample target skeleton data corresponding to the object parts included in the object part set. The training module is configured to train the object processing model based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stops, thereby obtaining a trained object processing model.

[0010] According to a seventh aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above method.

[0011] According to an eighth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0012] According to a ninth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0013] This specification provides an object processing method, comprising: determining target skeletal data corresponding to multiple object parts of a target object; inputting the target skeletal data into an object processing model to obtain associated skeletal data corresponding to associated objects of the target object, wherein the associated objects are objects worn by the target object, the associated skeletal data are determined based on sub-associated skeletal data corresponding to at least one sub-associated object, the sub-associated skeletal data are determined based on fused skeletal features corresponding to a set of object parts, the fused skeletal features are determined based on target skeletal data corresponding to object parts included in the set of object parts; and generating the associated object based on the associated skeletal data.

[0014] In the above method, an object processing model can be used to predict the associated bone data of related objects worn by the target object based on the target bone data corresponding to multiple object parts of the target object. Related objects are then generated based on this associated bone data, achieving automatic generation of bone data for related objects and thus improving generation efficiency. Furthermore, the object processing model can group multiple object parts of the target object to obtain an object part set. Based on the fused bone features corresponding to the object part set, sub-associated bone data of the corresponding sub-associated objects can be obtained. The associated bone data of the related objects is then generated based on the sub-associated bone data of the sub-associated objects. Further, by analyzing the relationship between the split and grouped object part set and the sub-associated objects, the associated bone data of the related objects is generated, further improving generation efficiency and reducing manpower and time consumption. Attached Figure Description

[0015] Figure 1 This is a flowchart of an object processing method provided in one embodiment of this specification; Figure 2 This is a schematic diagram of the target object in an object processing method provided in one embodiment of this specification; Figure 3 This is a schematic diagram of associated objects in an object processing method provided in one embodiment of this specification; Figure 4 This is a schematic diagram of the generation process of each sub-associated object in an object processing method provided in an embodiment of this specification; Figure 5 This is a schematic diagram of object processing model training in an object processing method provided in one embodiment of this specification; Figure 6 This is a flowchart illustrating an object processing method for model training provided in one embodiment of this specification. Figure 7 This is a schematic diagram of the structure of an object processing device provided in one embodiment of this specification; Figure 8 This is a flowchart of a data processing method provided in one embodiment of this specification; Figure 9 This is a schematic diagram of the structure of a data processing device provided in one embodiment of this specification; Figure 10 This is a flowchart of an object processing model training method provided in one embodiment of this specification; Figure 11 This is a schematic diagram of the structure of an object processing model training device provided in one embodiment of this specification; Figure 12 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0016] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0017] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0018] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0019] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0020] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0021] 3D: Three-Dimensional, a representation with three spatial coordinate axes: X, Y, and Z.

[0022] Biped skeleton: A standard skeletal rigging system designed for bipedal humanoid characters, including hierarchical joints such as the pelvis, spine, head, and limbs.

[0023] BVH files: BioVision Hierarchy, are used to store 3D skeletal animation data, such as motion capture data.

[0024] 6D 3D Rotation Representation: A continuous, singularity-free 3D rotation representation that encodes a rotation using 6 numbers (i.e., a six-dimensional vector).

[0025] MLP: Multilayer Perceptron, a type of fully connected feedforward neural network.

[0026] Adam: Adaptive Moment Estimation, an optimizer that adapts to the learning rate for training neural networks.

[0027] SmoothL1loss: A loss function that combines the advantages of L1 and L2 loss. It uses L2 loss for small errors, making it smooth and differentiable, and uses L1 loss for large errors, making it robust to outliers.

[0028] SMPL: Skinned Multi-Person Linear Model, a parametric 3D human body model that can generate realistic human body meshes with a small number of parameters.

[0029] In practical applications, with the development of game technology, clothing is a key element in shaping the character's personality in the creation of 3D characters (such as game characters, film and animation characters, and virtual humans). The skirt, as the most dynamic part of the clothing, directly affects the realism and visual expressiveness of the character's movements. In practice, creating skirts using static meshes can result in stiff, unresponsive skirts or clipping issues when the character walks or jumps, severely compromising immersion. To meet the requirements of skirts swaying naturally with character movements, without clipping between skirt layers, and conforming to physical laws, the technical approach of controlling skirt dynamics using skeletons has emerged. A skirt skeleton can be understood as dividing the skirt into independently movable areas, using bones as motion control fulcrums, and combining weight allocation and physical rules to achieve coordinated movement between the skirt and the character. Currently, the skirt position can usually be manually set on keyframes using skeleton creation tools, which consumes significant human and time resources, heavily relies on experience in skeleton generation, and has low generation efficiency. Therefore, an effective technical solution is urgently needed to address these problems.

[0030] This specification provides an object processing method, and also relates to an object processing apparatus, a data processing method, a data processing apparatus, an object processing model training method, an object processing model training apparatus, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0031] The object processing method and data processing method provided in this specification can be used to create digital cultural products. By determining the target skeletal data corresponding to multiple object parts of the target object, the object processing model calculates the associated skeletal data of the associated objects worn by the target object based on these target skeletal data, and then generates associated objects based on the associated skeletal data, so that the associated objects can move with the movement of the target object, thereby realizing the generation of digital cultural products and further providing users with realistic and immersive digital cultural content services.

[0032] Figure 1 A flowchart of an object processing method according to an embodiment of this specification is shown, which specifically includes the following steps: Step 102: Determine the target skeleton data corresponding to multiple object parts of the target object.

[0033] Specifically, the object processing method provided in the embodiments of this specification can be applied to the server side, and the object processing method can be used for the skeleton generation of associated objects worn by the target object.

[0034] The target object can be understood as a virtual object, including but not limited to game characters, film and animation characters, and virtual humans. Furthermore, the target object can be an object that needs to wear clothing. The multiple body parts of the target object can be understood as multiple body parts of the target object, including but not limited to the pelvis, spine, thoracic vertebrae, lumbar vertebrae, neck, head, hip joints, legs, shoulders, and arms. These body parts constitute the joints of the target object's skeleton and can be used to define the target object's posture. Furthermore, different postures defined by these body parts will also affect the posture of the clothing worn by the target object.

[0035] Based on this, target skeletal data corresponding to multiple body parts of the target object can be extracted.

[0036] In practical applications, the object processing method provided in the embodiments of this specification can be used to generate skirts for target objects. For ease of understanding, the embodiments of this specification are used as an example to illustrate the generation of skirts worn by a 3D virtual character, and do not affect the application of the object processing method provided in the embodiments of this specification to the generation of other clothing.

[0037] For example, the target object could be a 3D character model without a skirt, and the associated object could be the skirt worn by the 3D character model. See also Figure 2 and Figure 3 , Figure 2 A schematic diagram of a target object is shown in an object processing method according to an embodiment of this specification. Figure 3 A schematic diagram of associated objects in an object processing method provided according to an embodiment of this specification is shown.

[0038] In specific implementation, determining the target skeletal data corresponding to multiple object parts of the target object includes: Determine the target action data corresponding to the target object; Based on the target motion data, the bone rotation data corresponding to multiple object parts of the target object are extracted as the target bone data.

[0039] The target action data corresponding to the target object can be understood as the data of the target object when performing the target action. For example, it could be an image of the target object performing the target action. It's understandable that since the target object's posture differs depending on the action, the clothing worn by the target object will also change accordingly. Based on this, multiple target actions can be extracted from the target object, and subsequent prediction of associated skeletal data can be performed. Then, different associated objects (i.e., skirts) can be generated for different target actions. For example, when the target object is performing a starting action, the skeletal rotation data corresponding to multiple body parts of the target object can be extracted from the image of the target object performing the starting action as target skeletal data. The associated skeletal data predicted based on this target skeletal data is the associated skeletal data of the associated object worn by the target object during the starting action. The generated associated object corresponds to this starting action. Skeletal rotation data can be understood as information describing the orientation and rotation state of each joint (i.e., body part) in the target object's character skeleton in three-dimensional space.

[0040] Specifically, an image of the target object performing the target action can be acquired, and the skeletal rotation data corresponding to multiple body parts of the target object during the target action can be extracted from this image. In practice, based on the cascading relationship of the biped skeletons, the skeletal rotation data of key nodes of the target object can be extracted from the target action data corresponding to the target object; these key nodes represent the body parts of the target object. The target action data corresponding to the target object can be motion files created by the virtual character designer.

[0041] In practical applications, the target motion data corresponding to the target object can be the motion capture data of the target object (such as a BVH file), from which skeletal rotation data corresponding to multiple body parts can be extracted. Alternatively, it can be based on a deep learning model, where the target motion data is input into the deep learning model to obtain the skeletal rotation data output by the deep learning model. Skeletal rotation data can be represented by quaternions, and the skirt skeleton (i.e., the skeleton of the associated object) can include multiple chains, connected in a multi-level sequence. In addition, skeletal rotation data can also be represented by rotation matrices, Euler angles, axis angles, etc.

[0042] Taking the generation of the skirt worn by the target object as an example, the extracted target bone data corresponding to multiple object parts can include target bone data corresponding to the pelvis, target bone data corresponding to the spine, target bone data corresponding to different levels of the upper spine (such as thoracic vertebrae and lumbar vertebrae), target bone data corresponding to different positions of the neck, target bone data corresponding to the head, target bone data corresponding to the hip joint, target bone data corresponding to the lower leg, target bone data corresponding to the clavicle, target bone data corresponding to the upper arm, target bone data corresponding to the forearm, etc.

[0043] Furthermore, the skeletal data corresponding to the key points of the associated object to be generated can be extracted as target skeletal data. These key points can be understood as control points for the clothing or hair worn by the target object, and can be used to simulate the dynamic effects of the fabric or hairstyle. Taking a skirt as an example, the skeletal data corresponding to the key points of the lower and upper parts of the clothing, such as the hem, neckline, and cuffs, can also be extracted to drive fabric deformation. Additionally, the skeletal data corresponding to the key points of the back of the target object's head and the hair on its forehead can be extracted to simulate the effect of the target object's hair flowing with movement.

[0044] In practical applications, by extracting the skeletal rotation data of multiple parts of the target object, we can obtain... Here, N represents the number of bones, f represents the number of frames in the animation corresponding to the target object, and X represents the object part. X mainly consists of the target bone data corresponding to the multiple object parts and the bone data corresponding to the key points of the associated objects. In essence, target motion data can be understood as a frame of the animation where the target object performs the target action.

[0045] In summary, by determining the target action data corresponding to the target object, we can extract the skeletal rotation data of multiple body parts of the target object under the target action from the target action data. This facilitates the subsequent generation of the skeletal data of the associated objects worn by the target object under the target action, and further provides a data foundation for generating associated objects of the target object.

[0046] Furthermore, the step of extracting skeletal rotation data corresponding to multiple object parts of the target object as the target skeletal data based on the target motion data further includes: Based on the target motion data, extract the bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted bone rotation data as the target bone data.

[0047] In this context, a six-dimensional vector can be understood as six real numbers. Encoding the skeletal rotation data corresponding to multiple object parts using a six-dimensional vector can be understood as converting quaternion-represented skeletal rotation data into a 6D three-dimensional rotation representation based on the six-dimensional vector. Since quaternions are not continuous rotation representations, quaternion-represented skeletal rotation data can be converted into a 6D three-dimensional rotation representation. In other words, the converted skeletal rotation data is a 6D three-dimensional rotation representation. A 6D three-dimensional rotation representation is a parameterization method used to represent rigid body rotation in three-dimensional space. It encodes a rotation using a six-dimensional vector, but essentially still only represents a rotation with 3 degrees of freedom.

[0048] Specifically, the bone rotation data corresponding to multiple object parts of the target object can be extracted from the target motion data. Based on the six-dimensional vector, the bone rotation data corresponding to each object part is rotated and encoded, so that the bone rotation data represented by quaternions is converted into a 6D three-dimensional rotation representation, and the converted bone rotation data is used as the target bone data.

[0049] For example, based on the target action data, we extract the skeletal rotation data A1 corresponding to object part A, B1 corresponding to object part B, and C1 corresponding to object part C. Then, based on the six-dimensional vector, we can perform rotation encoding on the skeletal rotation data A1 corresponding to object part A to obtain the converted skeletal rotation data A2; based on the six-dimensional vector, we can perform rotation encoding on the skeletal rotation data B1 corresponding to object part B to obtain the converted skeletal rotation data B2; and based on the six-dimensional vector, we can perform rotation encoding on the skeletal rotation data C1 corresponding to object part C to obtain the converted skeletal rotation data C2.

[0050] In summary, by converting the skeletal rotation data represented by quaternions into a 6D three-dimensional rotation representation without the need for normalization constraints, the subsequent object processing model can directly output any 6D vector (i.e., a six-dimensional vector), which is easy to extract from the model.

[0051] Step 104: Input the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object, wherein the associated object is the object worn by the target object, the associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-associated object, the sub-associated skeleton data is determined based on the fused skeleton features corresponding to the object part set, and the fused skeleton features are determined based on the target skeleton data corresponding to the object parts contained in the object part set.

[0052] Among them, the associated object can be understood as the clothing object worn by the target object. For example, the associated object can be the skirt of the clothing worn by the target character. Then, the associated skeletal data can be the skirt skeletal data. The object processing model can be used to predict the associated skeletal data of the associated object based on the target skeletal data of the target object.

[0053] Specifically, after obtaining the target skeleton data corresponding to multiple object parts of the target object, the target skeleton data can be input into the object processing model. Based on the object processing model, feature extraction, grouping, and fusion are performed on the target skeleton data corresponding to multiple object parts to obtain the associated skeleton data corresponding to the associated objects.

[0054] Taking the generation of the skirt worn by the target object as an example, the target skeleton data corresponding to multiple object parts of the target object can be input into the object processing model to obtain the skirt skeleton data output by the object processing model.

[0055] In specific implementation, the step of inputting the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object includes: The target bone data is input into an object processing model, and feature encoding is performed on the target bone data in the object processing model to obtain the target bone features corresponding to the multiple object parts. The multiple object parts are grouped to obtain at least one set of object parts; Based on the target bone features corresponding to the object parts contained in each object part set, determine the sub-association bone data corresponding to at least one sub-association object, wherein the sub-association object and the object part set correspond to each other and have a positional association relationship; Based on the at least one sub-associated skeleton data, determine the associated skeleton data corresponding to the associated object.

[0056] The object processing model can be a trained neural network model. This model may include an encoder to extract feature vectors from the target skeleton data corresponding to each object part.

[0057] Sub-associated objects can be understood as sub-associated objects obtained by classifying and splitting associated objects. Taking a skirt as an example, since the skirt of the target object is distributed in different areas of the target object's body, the skirt skeleton can be classified according to the prior of the upper and lower associated skeletons. The skirt can be divided into the lower skirt, upper skirt, arm skirt, and head skirt, that is, divided into 4 sub-associated objects. The skeleton of the lower skirt depends on the range of motion of the pelvis, spine, thigh and calf in the target object's body parts. The skeleton of the upper skirt depends on the range of motion of the spine and sternum in the target object's body parts. The skeleton of the arm skirt depends on the range of motion of the arms in the target object's body parts. The head skirt can be divided into three dimensions: front, back and side, depending on the range of motion of the neck and head in the target object's body parts.

[0058] An object part set can be understood as a collection obtained by grouping multiple object parts. Taking the skirt as an example, the multiple object parts of the target object can be divided into three object part sets. These three object part sets represent the upper, middle, and lower parts of the target object's body skeleton, respectively. The upper object part set can include the neck, head, and arms; the middle object part set can include the pelvis, spine, and sternum; and the lower object part set can include the pelvis, spine, and legs.

[0059] Therefore, it's understandable that there can be a corresponding positional relationship between the set of object parts and its sub-related objects. For example, if the sub-related object is the hem of a skirt, since the skeleton of the hem mainly depends on the range of motion of the pelvis, spine, thigh, and calf, the sub-related skeletal data corresponding to this sub-related object can be determined based on the target skeletal features corresponding to the lower set of object parts. Alternatively, if the sub-related hem is the hem of a head skirt, since the skeleton of the head skirt mainly depends on the range of motion of the neck and head, the sub-related skeletal data corresponding to this sub-related object can be determined based on the target skeletal features corresponding to the upper set of object parts. In other words, the sub-related skeletal data corresponding to a sub-related object can be determined by either determining the set of object parts based on positional relationships or by determining the target skeletal features corresponding to the corresponding object parts.

[0060] Based on this, after inputting the target skeleton data corresponding to multiple object parts of the target object into the object processing model, the target skeleton features of the target skeleton data corresponding to each object part can be extracted by the encoder in the object processing model. Multiple object parts are grouped to obtain at least one set of object parts. Based on the target skeleton features corresponding to the object parts contained in each set of object parts, at least one sub-associated object corresponding to the sub-associated skeleton data is determined. The sub-associated skeleton data corresponding to at least one sub-associated object can be concatenated or added to obtain the associated skeleton data corresponding to the associated object.

[0061] In practical applications, the feature vector of the target bone data corresponding to each object part can be extracted by MLP as the target bone feature corresponding to each object part.

[0062] In summary, by using neural networks to map the bones of different limbs and integrating bone priors that depend on vertical association, the neural network generates the associated bone data of the corresponding related objects based on the target bone data of the input target object, thereby improving the efficiency of creating related objects.

[0063] Further, the step of determining at least one sub-associated object's sub-associated bone data based on the target bone features corresponding to the object parts contained in each object part set includes: The target bone features corresponding to the object parts contained in each set of object parts are fused to obtain the fused bone features corresponding to each set of object parts. Based on the fused skeletal features and target skeletal features corresponding to the at least one set of object parts, sub-associative skeletal data corresponding to at least one sub-associative object is determined.

[0064] Specifically, for each set of object parts, the target bone features corresponding to all object parts contained in each set can be fused to obtain the fused bone features corresponding to each set of object parts. Based on the fused bone features corresponding to the set of object parts and the target bone features corresponding to the object parts contained in the set of object parts, the sub-associative bone data corresponding to the sub-associative objects is determined.

[0065] In addition, when generating sub-associated skeleton data corresponding to a sub-associated object, the object part corresponding to the sub-associated object can also be determined, and the sub-associated skeleton data corresponding to the sub-associated object can be determined based on the target skeleton features corresponding to the object part.

[0066] In practical applications, taking the skirt hem as an example of the associated object, the sub-associated objects include the lower skirt hem, the upper skirt hem, the arm skirt hems, and the head skirt hem. See also... Figure 4 , Figure 4 A schematic diagram is shown of the process of generating each sub-associated object in an object processing method according to an embodiment of this specification.

[0067] When the child associated object is the lower skirt, since the lower skirt is created from the waist of the target object and distributed radially downwards along the skirt, each bone chain controls a part of the skirt area, making it more prominent in the presentation. Moreover, the lower skirt has more layers and more bone chains. Based on this, the lower skirt can be divided into the upper part of the lower skirt and the lower part of the lower skirt for calculation.

[0068] The process of calculating the sub-associated bone data at the upper part of the skirt hem based on the object processing model is as follows: ,in, For the child-related skeleton data at the upper part of the skirt hem, The target bone features are those corresponding to the left hip joint of the target object. The target skeletal features are the corresponding bones of the left lower leg of the target object. The target skeletal features are those corresponding to the right hip joint of the target object. The target skeletal features are the corresponding bones of the right lower leg of the target object. F represents the fused skeletal features corresponding to the set of object parts below. F can be understood as the input-to-output mapping function implemented by the object processing model. For example, F can be a feature processing network layer in the object processing model. This feature processing network layer can perform processes such as fusing, splicing, linear transformation, and attention mechanism processing on the input target skeletal features and fused skeletal features. The embodiments in this specification do not limit this.

[0069] Because the terminal bones of each skeletal chain have a greater range of motion during the movement of the target object, feature fusion can be used to combine the target bone features corresponding to the lower leg of the target object with the sub-related bone data of the upper part of the skirt hem to calculate the sub-related bone data of the lower part of the skirt hem. The process of calculating the sub-related bone data of the lower part of the skirt hem based on the object processing model is as follows: .in, This refers to the sub-associated bone data at the bottom of the skirt hem.

[0070] Therefore, the sub-related bone data at the upper end of the skirt hem and the sub-related bone data at the lower end of the skirt hem can be spliced ​​together to form a complete skirt hem. The specific process is as follows: ,in, This refers to the child-related skeletal data for the lower skirt hem.

[0071] The process of calculating the sub-associative bone data of the upper skirt based on the object processing model is as follows: ,in, For the child skeleton data of the upper skirt hem, The target skeletal features are those corresponding to level 2 (e.g., lumbar vertebrae) of the upper spine of the target object. The target skeletal features are the left clavicle of the target object. The target skeletal features are the right clavicle of the target object. The fused skeletal features are the corresponding set of object parts in the middle.

[0072] The process of calculating the sub-associative bone data of the arm and skirt based on the object processing model is as follows: ,in, For the child-related skeleton data of the arm and skirt hem, The target skeletal features are the corresponding features of the left upper arm of the target object. The target skeletal features are the corresponding features of the upper right arm of the target object. The target skeletal features are the corresponding features of the left forearm of the target object. The target skeletal features are the corresponding features of the right forearm of the target object.

[0073] The process of calculating the sub-associative bone data of the head skirt based on the object processing model is as follows: ,in, For the child-related skeleton data of the head and skirt, The target skeletal features corresponding to the neck of the target object. The target skeletal features corresponding to position point 1 on the neck of the target object. The target skeletal features are the head of the target object. This refers to the fused skeletal features corresponding to the set of object parts above.

[0074] Understandably, the object processing model can calculate the sub-associative skeletal data of sub-associated objects based on the object parts of the input target object. For example, if the target skeletal data of all object parts of the target object are input, the object processing model can calculate the sub-associative skeletal data of all sub-associated objects, such as the lower hem, upper hem, arm hem, and head hem. The resulting associated object will then be a complete skirt containing the lower hem, upper hem, arm hem, and head hem. If the target skeletal data of the upper object part of the target object is input, the object processing model can calculate the sub-associative skeletal data of the corresponding sub-associated objects. For example, the object processing model can calculate the sub-associative skeletal data of the arm hem and head hem, resulting in a skirt containing both the arm hem and head hem. Therefore, in practical applications, to generate a half-length skirt for a target object, the target skeletal data corresponding to the lower object part of the target object can be input into the object processing model, enabling the model to generate the sub-associative skeletal data for the lower and upper hems. The sub-association skeletal data of which sub-association objects the object processing model can generate can be adjusted according to actual needs. Specifically, this can be achieved by adjusting the input of the object processing model or by training and optimizing the object processing model. This specification does not limit this in the embodiments.

[0075] In summary, by generating sub-association skeleton data for each sub-association object, it is easier to generate complete association objects in the future, thus improving the efficiency of association object generation.

[0076] Step 106: Generate the associated object based on the associated skeletal data.

[0077] Specifically, after generating the associated skeleton data corresponding to the associated object, the associated object can be generated based on the associated skeleton data corresponding to the associated object.

[0078] In specific implementation, generating the associated object based on the associated skeletal data includes: Skinning and binding are performed on the associated skeletal data to generate the associated object.

[0079] Specifically, the 3D mesh of the associated object can be skinned and bound to the associated skeletal data, and a weight can be assigned to each vertex of the associated skeletal data. This weight determines which bones affect each vertex, so that the mesh changes as the bones of the associated object move. Subsequently, by driving the skeletal motion of the associated object, the associated object can move with the target object, avoiding clipping and stiffness.

[0080] Furthermore, before inputting the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object, the method further includes: Obtain the sample target skeleton data corresponding to multiple object parts of the sample target object and the sample associated skeleton data corresponding to the sample associated objects; The object processing model is trained based on the target skeleton data and the associated skeleton data of the sample until the training stops, thus obtaining the trained object processing model.

[0081] In this context, the target skeleton data and associated skeleton data can be understood as training sample pairs used for training the object processing model. During supervised training of the object processing model, the target skeleton data can be used as training samples, and the associated skeleton data can be used as training labels. The training stopping condition can be understood as the model loss value reaching a preset loss threshold and / or the number of training iterations reaching a preset threshold.

[0082] Specifically, sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to the sample associated object can be obtained. The sample target skeleton data is input into the object processing model to obtain the predicted associated skeleton data output by the object processing model. Based on the sample associated skeleton data and the predicted associated skeleton data, the model loss value is calculated. The object processing model is trained based on the model loss value until the model loss value reaches a preset loss value threshold and / or the number of model training times reaches a preset number threshold, thus obtaining the trained object processing model.

[0083] It is understandable that the specific process of obtaining the target skeleton data and the associated skeleton data of the sample is similar to the process of extracting the target skeleton data described above. Both can be obtained by extracting the target action data corresponding to the target object and the associated object of the sample. The embodiments in this specification will not be repeated here.

[0084] In summary, by training the object processing model as described above, the ability to generate skeletal data for the object processing model is achieved.

[0085] Further, the step of obtaining sample target skeleton data corresponding to multiple object parts of the sample target object includes: Determine the sample target action data corresponding to the sample target object; Based on the target motion data, extract the sample bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the sample bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted sample bone rotation data as the sample target bone data.

[0086] Specifically, since the data is expected to be continuous during the training phase of the object processing model, and Euler angles, axis angles and quaternions are not continuous rotation representations, after extracting sample bone rotation data from sample action data, the sample bone rotation data represented by quaternions can be converted into a 6D three-dimensional rotation representation based on a six-dimensional vector.

[0087] It is understandable that the process of extracting sample bone rotation data and performing rotation encoding on the sample bone rotation data is similar to the process of extracting bone rotation data and performing rotation encoding on the bone rotation data described above, and will not be repeated in the embodiments of this specification.

[0088] Furthermore, before performing rotation encoding on the sample bone rotation data corresponding to the multiple object parts based on a six-dimensional vector to obtain the converted sample bone rotation data as the sample target bone data, the method further includes: Preset sample noise is added to the sample bone rotation data corresponding to the multiple object parts to obtain noisy sample bone rotation data; The step of performing rotation encoding on the sample bone rotation data corresponding to the multiple object parts based on a six-dimensional vector to obtain the converted sample bone rotation data as the sample target bone data includes: Based on a six-dimensional vector, the noisy sample bone rotation data is rotated and encoded to obtain the converted noisy sample bone rotation data as the sample target bone data.

[0089] The preset sample noise can be understood as noise perturbation added to the converted sample bone rotation data, such as Gaussian noise.

[0090] Specifically, based on the continuity of 6D three-dimensional rotation representation, a certain amount of noise perturbation (i.e., preset sample noise) can be added to the converted sample bone rotation data to represent minute rotations, thereby increasing the robustness of the object processing model. To ensure the correctness of the rotation after adding noise, the noisy sample bone rotation data obtained after adding preset sample noise can be converted into a rotation matrix, and then converted into a 6D three-dimensional rotation representation through the rotation matrix, thus ensuring orthogonality.

[0091] Understandably, similar noise addition and rotation encoding processes can be applied to the sample-associated skeletal data of the sample-associated object. Specifically, sample-associated skeletal data of multiple key points of the sample-associated object can be extracted from the target action data corresponding to the sample target object wearing the sample-associated object. This specification will not repeat the details in the embodiments.

[0092] Furthermore, to improve the generalization ability of the object processing model and extend it to more realistic scenarios, we can adopt... The scheme is used as the input and output of the object processing model, where Y is the predicted associated skeleton data output by the object processing model, and X is the sample target skeleton data input to the object processing model. Added preset sample noise, Predictive noise for the object processing model. Since it's difficult to guarantee that actions with similar motion trends will have identical rotation data during the generation of associated objects, the X data of each input sample can be understood as rotation data carrying noise. Predicting noise through the object processing model allows it to better identify noise deflections in samples, thereby improving its robustness against interference. See also Figure 5 , Figure 5 A schematic diagram illustrating object processing model training in an object processing method according to an embodiment of this specification is shown. Figure 5 As shown, the sample bone rotation data (including sample target bone data and sample associated bone data) in the training sample pair, represented by quaternions, Euler angles or axis angles, can be converted into a 6D rotation representation. The model is then trained based on the added preset sample noise and the 6D rotation representation, so that the object processing model can output predicted bone rotation data (i.e. predicted associated bone data) represented by quaternions, Euler angles or axis angles.

[0093] Therefore, the training data for training the object processing model includes both training sample pairs without noise and training sample pairs with noise. Based on this, the ratio of training sample pairs without noise to those with noise can be 1:3. By training the object processing model using a mixed dataset, a better balance between training accuracy and noise robustness can be achieved. The learning rate of the object processing model can be set to 0.0001, the optimizer can be Adam, the training epochs can be set to 100, the batch size can be set to 1024, and the loss function used to calculate the model's loss value can be SmoothL1 loss.

[0094] Furthermore, in the application phase, the trained object processing model benefits from the increasingly simplified motion capture technology resulting from the current advancements in artificial intelligence. Users can directly capture images of the target object to obtain motion capture data, which facilitates the subsequent extraction of target skeletal data for multiple parts of the target object. For example, 3D human reconstruction can be achieved based on SMPL, extracting relevant key points and directly generating skirt skeletons for the target object.

[0095] In summary, by rapidly adjusting the rotation of the main torso bones of the target object, the skirt bones corresponding to the frame sequence are automatically generated, with significant improvements in both generation quality and time, greatly increasing generation efficiency. This method can also be continuously optimized using an iterative approach; new motion resources can be further refined through neural network learning to optimize the generation effect.

[0096] In summary, the above methods utilize an object processing model to predict the associated skeletal data of related objects worn by the target object based on the target skeletal data corresponding to multiple object parts of the target object. Furthermore, the associated objects are generated based on this associated skeletal data, thus automatically generating the skeletal data of the associated objects and improving generation efficiency. Moreover, the object processing model can group multiple object parts of the target object to obtain an object part set. Based on the fused skeletal features corresponding to the object part set, sub-associated skeletal data of the corresponding sub-associated objects is obtained. The associated skeletal data of the associated objects is then generated based on the sub-associated skeletal data of the sub-associated objects. Finally, the generation of the associated skeletal data of the associated objects is achieved by analyzing the relationships between the split and grouped object part set and the sub-associated objects, further improving generation efficiency and reducing manpower and time consumption.

[0097] The following is in conjunction with the appendix Figure 6 Taking the application of the object processing method provided in this specification in model training as an example, the object processing method will be further explained. Figure 6 This specification illustrates a flowchart of an object processing method for model training according to an embodiment of the present invention, which specifically includes the following steps: Step 602: Determine the sample target action data corresponding to the sample target object.

[0098] Specifically, the sample target action data can be data on the target object performing target actions while wearing a sample associated object. The sample target object can be a 3D virtual character, and the sample associated object can be the skirt worn by the 3D virtual character.

[0099] Step 604: Based on the sample target action data, extract the sample bone rotation data corresponding to multiple object parts of the sample target object.

[0100] Step 606: Add preset sample noise to the sample bone rotation data corresponding to multiple object parts to obtain noisy sample bone rotation data.

[0101] Step 608: Based on the six-dimensional vector, perform rotation encoding on the noisy sample bone rotation data to obtain the converted noisy sample bone rotation data as the sample target bone data.

[0102] Step 610: Extract the sample-related skeletal data of the sample-related objects.

[0103] Specifically, the sample-related skeletal data of the sample-related object can be extracted according to the process described in steps 604 to 608 above.

[0104] Step 612: Input the sample target skeleton data into the object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object.

[0105] Step 614: Train the object processing model based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stops, and obtain the trained object processing model.

[0106] In summary, through the above training, the noise resistance of the object processing model is improved, the generalization ability of the object processing model is enhanced, and thus the versatility of the object processing model is improved.

[0107] Corresponding to the above method embodiments, this specification also provides embodiments of an object processing apparatus. Figure 7 A schematic diagram of an object processing apparatus according to an embodiment of this specification is shown. Figure 7 As shown, the device includes: The determination module 702 is configured to determine the target skeleton data corresponding to multiple object parts of the target object; Input module 704 is configured to input the target skeleton data into an object processing model to obtain associated skeleton data corresponding to the associated object of the target object, wherein the associated object is an object worn by the target object, the associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-associated object, the sub-associated skeleton data is determined based on the fused skeleton features corresponding to the object part set, and the fused skeleton features are determined based on the target skeleton data corresponding to the object parts included in the object part set; The generation module 706 is configured to generate the associated object based on the associated skeletal data.

[0108] In an optional embodiment, the input module 704 is further configured to: The target bone data is input into an object processing model, and feature encoding is performed on the target bone data in the object processing model to obtain the target bone features corresponding to the multiple object parts. The multiple object parts are grouped to obtain at least one set of object parts; Based on the target bone features corresponding to the object parts contained in each object part set, determine the sub-association bone data corresponding to at least one sub-association object, wherein the sub-association object and the object part set correspond to each other and have a positional association relationship; Based on the at least one sub-associated skeleton data, determine the associated skeleton data corresponding to the associated object.

[0109] In an optional embodiment, the input module 704 is further configured to: The target bone features corresponding to the object parts contained in each set of object parts are fused to obtain the fused bone features corresponding to each set of object parts. Based on the fused skeletal features and target skeletal features corresponding to the at least one set of object parts, sub-associative skeletal data corresponding to at least one sub-associative object is determined.

[0110] In an optional embodiment, the determining module 702 is further configured to: Determine the target action data corresponding to the target object; Based on the target motion data, the bone rotation data corresponding to multiple object parts of the target object are extracted as the target bone data.

[0111] In an optional embodiment, the determining module 702 is further configured to: Based on the target motion data, extract the bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted bone rotation data as the target bone data.

[0112] In an optional embodiment, the generation module 706 is further configured to: Skinning and binding are performed on the associated skeletal data to generate the associated object.

[0113] In an optional embodiment, the apparatus further includes a training module, which is further configured to: Obtain the sample target skeleton data corresponding to multiple object parts of the sample target object and the sample associated skeleton data corresponding to the sample associated objects; The object processing model is trained based on the target skeleton data and the associated skeleton data of the sample until the training stops, thus obtaining the trained object processing model.

[0114] In an optional embodiment, the training module is further configured to: Determine the sample target action data corresponding to the sample target object; Based on the target motion data, extract the sample bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the sample bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted sample bone rotation data as the sample target bone data.

[0115] In an optional embodiment, the training module is further configured to: Preset sample noise is added to the sample bone rotation data corresponding to the multiple object parts to obtain noisy sample bone rotation data; Based on a six-dimensional vector, the noisy sample bone rotation data is rotated and encoded to obtain the converted noisy sample bone rotation data as the sample target bone data.

[0116] In the aforementioned device, an object processing model can predict the associated bone data of related objects worn by the target object based on the target bone data corresponding to multiple object parts of the target object. Then, associated objects are generated based on this associated bone data, achieving automatic generation of the bone data for associated objects and thus improving generation efficiency. Furthermore, the object processing model can group multiple object parts of the target object to obtain an object part set. Based on the fused bone features corresponding to the object part set, sub-associated bone data of the corresponding sub-associated objects is obtained. The associated bone data of the associated objects is then generated based on the sub-associated bone data of the sub-associated objects. Further, by analyzing the association relationships between the split and grouped object part set and the sub-associated objects, the associated bone data of the associated objects is generated, further improving generation efficiency and reducing manpower and time consumption.

[0117] The above is an illustrative scheme of an object processing apparatus according to this embodiment. It should be noted that the technical solution of this object processing apparatus and the technical solution of the object processing method described above belong to the same concept. For details not described in detail in the technical solution of the object processing apparatus, please refer to the description of the technical solution of the object processing method described above.

[0118] Corresponding to the above-described method embodiments, this specification also provides a data processing method. Figure 8 A flowchart of a data processing method provided in one embodiment of this specification is shown, which specifically includes the following steps.

[0119] Step 802: Determine the target skeleton data corresponding to multiple parts of the target virtual 3D character; Step 804: Input the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the clothing object of the target virtual 3D character. The associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-clothing object. The sub-associated skeleton data is determined based on the fused skeleton features corresponding to the character part set. The fused skeleton features are determined based on the target skeleton data corresponding to the character parts contained in the character part set. Step 806: Generate the clothing object based on the associated skeletal data.

[0120] Specifically, the target virtual 3D character can be understood as the target object, and the clothing object can be understood as the associated object. The process of generating the associated skeletal data corresponding to the clothing object is similar to the process of generating the associated skeletal data corresponding to the associated object described above, and will not be repeated in the embodiments of this specification.

[0121] The above is an illustrative scheme of a data processing method according to this embodiment. It should be noted that the technical solution of this data processing method and the technical solution of the object processing method described above belong to the same concept. For details not described in detail in the technical solution of the data processing method, please refer to the description of the technical solution of the object processing method described above.

[0122] Corresponding to the above method embodiments, this specification also provides data processing apparatus embodiments. Figure 9 A schematic diagram of the structure of a data processing apparatus provided in one embodiment of this specification is shown. Figure 9 As shown, the device includes: The determination module 902 is configured to determine the target skeleton data corresponding to multiple parts of the target virtual 3D character; Input module 904 is configured to input the target skeleton data into an object processing model to obtain associated skeleton data corresponding to the clothing object of the target virtual 3D character. The associated skeleton data is determined based on sub-associated skeleton data corresponding to at least one sub-clothing object. The sub-associated skeleton data is determined based on fused skeleton features corresponding to the character part set. The fused skeleton features are determined based on target skeleton data corresponding to the character parts included in the character part set. The generation module 906 is configured to generate the clothing object based on the associated skeletal data.

[0123] The above is an illustrative scheme of a data processing apparatus according to this embodiment. It should be noted that the technical solution of this data processing apparatus and the technical solution of the object processing method described above belong to the same concept. For details not described in detail in the technical solution of the data processing apparatus, please refer to the description of the technical solution of the object processing method described above.

[0124] Corresponding to the above-described method embodiments, this specification also provides a method for training an object processing model. Figure 10 A flowchart of an object processing model training method provided in one embodiment of this specification is shown, which specifically includes the following steps.

[0125] Step 1002: Obtain sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to sample associated objects, wherein the sample associated objects are the objects worn by the sample target object; Step 1004: Input the sample target skeleton data into the object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object, wherein the predicted associated skeleton data is determined based on the sample sub-associated skeleton data corresponding to at least one sample sub-associated object, the sample sub-associated skeleton data is determined based on the sample fusion skeleton features corresponding to the object part set, and the sample fusion skeleton features are determined based on the sample target skeleton data corresponding to the object parts included in the object part set. Step 1006: Train the object processing model based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stops, and obtain the trained object processing model.

[0126] The above is an illustrative scheme of an object processing model training method according to this embodiment. It should be noted that the technical solution of this object processing model training method belongs to the same concept as the object processing method described above. For details not described in detail in the technical solution of the object processing model training method, please refer to the description of the technical solution of the object processing method described above.

[0127] Corresponding to the above method embodiments, this specification also provides embodiments of an object processing model training apparatus. Figure 11 A schematic diagram of the structure of an object processing model training device provided in one embodiment of this specification is shown. Figure 11 As shown, the device includes: The acquisition module 1102 is configured to acquire sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to sample associated objects, wherein the sample associated objects are objects worn by the sample target object; Input module 1104 is configured to input the sample target skeleton data into an object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object. The predicted associated skeleton data is determined based on the sample sub-associated skeleton data corresponding to at least one sample sub-associated object. The sample sub-associated skeleton data is determined based on the sample fusion skeleton features corresponding to the object part set. The sample fusion skeleton features are determined based on the sample target skeleton data corresponding to the object parts included in the object part set. The training module 1106 is configured to train the object processing model based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stop condition is met, thereby obtaining the trained object processing model.

[0128] The above is a schematic scheme of an object processing model training device according to this embodiment. It should be noted that the technical solution of this object processing model training device and the technical solution of the object processing method described above belong to the same concept. For details not described in detail in the technical solution of the object processing model training device, please refer to the description of the technical solution of the object processing method described above.

[0129] Figure 12 A structural block diagram of a computing device 1200 according to an embodiment of this specification is shown. The components of the computing device 1200 include, but are not limited to, a memory 1210 and a processor 1220. The processor 1220 is connected to the memory 1210 via a bus 1230, and a database 1250 is used to store data.

[0130] The computing device 1200 also includes an access device 1240, which enables the computing device 1200 to communicate via one or more networks 1260. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1240 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0131] In one embodiment of this specification, the aforementioned components of the computing device 1200 and Figure 12 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 12 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0132] The computing device 1200 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1200 can also be a mobile or stationary server.

[0133] The processor 1220 is configured to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above method.

[0134] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above method belong to the same concept, and all details not described in detail in the technical solution of the computing device can be referred to the description of the technical solution of the above method.

[0135] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0136] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the method described above belong to the same concept, and all details not described in detail in the technical solution of the storage medium can be referred to the description of the technical solution of the method described above.

[0137] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0138] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the above method belong to the same concept, and all details not described in detail in the technical solution of the computer program product can be referred to the description of the technical solution of the above method.

[0139] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0140] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0141] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this specification is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this specification. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this specification.

[0142] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0143] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. These embodiments have been selected and specifically described in this specification to better explain the principles and practical applications of this specification, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An object processing method, characterized in that, include: Determine the target skeleton data corresponding to multiple object parts of the target object; The target skeleton data is input into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object. The associated object is the object worn by the target object. The associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-associated object. The sub-associated skeleton data is determined based on the fused skeleton features corresponding to the object part set. The fused skeleton features are determined based on the target skeleton data corresponding to the object parts included in the object part set. The associated object is generated based on the associated skeletal data.

2. The method according to claim 1, characterized in that, The step of inputting the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object includes: The target bone data is input into an object processing model, and feature encoding is performed on the target bone data in the object processing model to obtain the target bone features corresponding to the multiple object parts. The multiple object parts are grouped to obtain at least one set of object parts; Based on the target bone features corresponding to the object parts contained in each object part set, determine the sub-association bone data corresponding to at least one sub-association object, wherein the sub-association object and the object part set correspond to each other and have a positional association relationship; Based on the at least one sub-associated skeleton data, determine the associated skeleton data corresponding to the associated object.

3. The method according to claim 2, characterized in that, The step of determining at least one sub-associated object's sub-associated bone data based on the target bone features corresponding to the object parts contained in each object part set includes: The target bone features corresponding to the object parts contained in each set of object parts are fused to obtain the fused bone features corresponding to each set of object parts. Based on the fused skeletal features and target skeletal features corresponding to the at least one set of object parts, sub-associative skeletal data corresponding to at least one sub-associative object is determined.

4. The method according to claim 1, characterized in that, The determination of target skeletal data corresponding to multiple object parts of the target object includes: Determine the target action data corresponding to the target object; Based on the target motion data, the bone rotation data corresponding to multiple object parts of the target object are extracted as the target bone data.

5. The method according to claim 4, characterized in that, The step of extracting bone rotation data corresponding to multiple object parts of the target object as the target bone data based on the target motion data further includes: Based on the target motion data, extract the bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted bone rotation data as the target bone data.

6. The method according to any one of claims 1-5, characterized in that, The step of generating the associated object based on the associated skeletal data includes: Skinning and binding are performed on the associated skeletal data to generate the associated object.

7. The method according to any one of claims 1-5, characterized in that, Before inputting the target skeleton data into the object processing model to obtain the associated skeleton data corresponding to the associated object of the target object, the method further includes: Obtain the sample target skeleton data corresponding to multiple object parts of the sample target object and the sample associated skeleton data corresponding to the sample associated objects; The object processing model is trained based on the target skeleton data and the associated skeleton data of the sample until the training stops, thus obtaining the trained object processing model.

8. The method according to claim 7, characterized in that, The acquisition of sample target skeleton data corresponding to multiple object parts of the sample target object includes: Determine the sample target action data corresponding to the sample target object; Based on the target motion data, extract the sample bone rotation data corresponding to multiple object parts of the target object; Based on a six-dimensional vector, the sample bone rotation data corresponding to the multiple object parts are rotated and encoded to obtain the converted sample bone rotation data as the sample target bone data.

9. The method according to claim 8, characterized in that, Before performing rotation encoding on the sample bone rotation data corresponding to the multiple object parts based on a six-dimensional vector to obtain the converted sample bone rotation data as the sample target bone data, the method further includes: Preset sample noise is added to the sample bone rotation data corresponding to the multiple object parts to obtain noisy sample bone rotation data; The step of performing rotation encoding on the sample bone rotation data corresponding to the multiple object parts based on a six-dimensional vector to obtain the converted sample bone rotation data as the sample target bone data includes: Based on a six-dimensional vector, the noisy sample bone rotation data is rotated and encoded to obtain the converted noisy sample bone rotation data as the sample target bone data.

10. A data processing method, characterized in that, include: Determine the target skeleton data corresponding to multiple parts of the target virtual 3D character; The target skeleton data is input into the object processing model to obtain the associated skeleton data corresponding to the clothing object of the target virtual 3D character. The associated skeleton data is determined based on the sub-associated skeleton data corresponding to at least one sub-clothing object. The sub-associated skeleton data is determined based on the fused skeleton features corresponding to the character part set. The fused skeleton features are determined based on the target skeleton data corresponding to the character parts included in the character part set. The clothing object is generated based on the associated skeletal data.

11. A method for training an object processing model, characterized in that, include: Obtain sample target skeleton data corresponding to multiple object parts of the sample target object and sample associated skeleton data corresponding to sample associated objects, wherein the sample associated objects are the objects worn by the sample target object; The sample target skeleton data is input into the object processing model to obtain the predicted associated skeleton data corresponding to the sample associated object. The predicted associated skeleton data is determined based on the sample sub-associated skeleton data corresponding to at least one sample sub-associated object. The sample sub-associated skeleton data is determined based on the sample fused skeleton features corresponding to the object part set. The sample fused skeleton features are determined based on the sample target skeleton data corresponding to the object parts included in the object part set. The object processing model is trained based on the sample-associated skeletal data and the predicted-associated skeletal data until the training stops, thus obtaining a fully trained object processing model.

12. A computing device, characterized in that, include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 11.

13. A computer-readable storage medium storing a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 11.

14. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1 to 11.

Images