Model generation method, model generation device, electronic equipment and storage medium

Abstract

The invention provides a model generation method, a model generation device, electronic equipment and a computer readable storage medium, and belongs to the technical field of computers. The method comprises the steps of obtaining a to-be-processed human body image, and performing image segmentation on the to-be-processed human body image to obtain a first part image of a first part of a human body; processing the first part image through a first network to obtain first three-dimensional feature information of the first part; processing the to-be-processed human body image through a second network, and determining second three-dimensional feature information of a second part of the human body; and generating a three-dimensional human body model of the human body according to the first three-dimensional feature information and the second three-dimensional feature information. According to the invention, the three-dimensional human body model can be accurately and effectively generated.

Description

Background Art

[0002] 3D human modeling is an important research issue in the field of computer vision and graphics. In recent years, 3D human reconstruction technology based on human images has made some progress in practical applications. For example, in scenarios where service robots interact with humans, robots can estimate the 3D human model of the interacting object from their own perspective; in augmented reality applications, the system can accurately estimate the 3D shape of the target human body from a specific angle.

[0003] However, the information that a single-view human body image can provide is often limited. For example, it only provides visual information of the front of the human body. This makes the three-dimensional reconstruction process highly dependent on the prior knowledge of human body geometry. In addition, when the human body is blocked by other obstacles, the visual information that can be obtained may also introduce other interference factors. The accuracy of the existing technology that directly uses human body images to construct a complete three-dimensional human body model may be affected to a certain extent, and the image information of the blocked object may also introduce interference, further increasing the difficulty of three-dimensional reconstruction. Therefore, how to improve the precision and accuracy of three-dimensional human body model reconstruction is a problem that needs to be solved urgently in the existing technology.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides a model generation method, a model generation device, an electronic device and a computer-readable storage medium, thereby overcoming the problem of low precision and accuracy in the generation of three-dimensional human body models in the prior art at least to a certain extent.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a model generation method is provided, comprising: acquiring a human body image to be processed, and performing image segmentation on the human body image to be processed to obtain a first part image of a first part of a human body; processing the first part image through a first network to obtain first three-dimensional feature information of the first part; processing the human body image to be processed through a second network to determine second three-dimensional feature information of a second part of the human body; and generating a three-dimensional human body model of the human body according to the first three-dimensional feature information and the second three-dimensional feature information.

[0008] According to one aspect of the present disclosure, a model generation device is provided, including: a human body image acquisition module, used to acquire a human body image to be processed, and perform image segmentation on the human body image to be processed to obtain a first part image of a first part of a human body; a first information determination module, used to process the first part image through a first network to obtain first three-dimensional feature information of the first part; a second information determination module, used to process the human body image to be processed through a second network to determine second three-dimensional feature information of a second part of the human body; and a human body model generation module, used to generate a three-dimensional human body model of the human body according to the first three-dimensional feature information and the second three-dimensional feature information.

[0009] According to one aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned methods by executing the executable instructions.

[0010] According to one aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above methods is implemented.

[0011] The exemplary embodiments of the present disclosure have the following beneficial effects:

[0012] A human body image to be processed is obtained, and image segmentation is performed on the human body image to obtain a first part image of a first part of the human body; the first part image is processed by a first network to obtain first three-dimensional feature information of the first part; the human body image to be processed is processed by a second network to determine second three-dimensional feature information of a second part of the human body; and a three-dimensional human body model is generated according to the first three-dimensional feature information and the second three-dimensional feature information. On the one hand, this exemplary embodiment proposes a new model generation method, which divides the human body into a first part and a second part based on the human body image to be processed, and determines the first three-dimensional feature information and the second three-dimensional feature information, and generates a three-dimensional human body model according to the first three-dimensional feature information and the second three-dimensional feature information, and can perform three-dimensional reconstruction on the first part and the second part of the human body respectively, and integrate the three-dimensional reconstruction results of the first part and the second part to obtain a complete three-dimensional human body model. Compared with directly reconstructing based on the human body image to be processed, this method of performing three-dimensional reconstruction and merging of different parts of the human body reduces the influence of interference information or partial information missing that may exist in the human body image to be processed on the generation of the three-dimensional human body model, and improves the efficiency and accuracy of the reconstruction of the three-dimensional human body model; on the other hand, this exemplary embodiment adopts different methods to perform three-dimensional reconstruction on the first part and the second part, and can provide targeted three-dimensional reconstruction methods for different parts, further ensuring the accuracy and detail expression of the reconstruction of the three-dimensional human body model.

[0013] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0015] Figure 1 A flow chart schematically illustrates a model generation method in this exemplary embodiment;

[0016] Figure 2 A schematic diagram of the structure of a three-dimensional hybrid attention module in this exemplary embodiment is schematically shown;

[0017] Figure 3 A schematic diagram of a network framework of a first network in this exemplary embodiment is schematically shown;

[0018] Figure 4 A schematic diagram of an architecture for determining second three-dimensional feature information in this exemplary embodiment is schematically shown;

[0019] Figure 5 A schematic diagram schematically illustrates an architecture of a model generation method in this exemplary embodiment;

[0020] Figure 6 A structural block diagram of a model generation device in this exemplary embodiment is schematically shown;

[0021] Figure 7 An electronic device for implementing the above method in this exemplary embodiment is schematically shown. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0023] The exemplary embodiment of the present disclosure first provides a model generation method. Figure 1 This exemplary embodiment is further described as follows. Figure 1As shown, the model generation method may include the following steps S110 to S140:

[0024] Step S110 , obtaining a human body image to be processed, and performing image segmentation on the human body image to be processed to obtain a first part image of a first part of the human body.

[0025] The human body image to be processed refers to an image that needs to be reconstructed into a three-dimensional human body model, including a human body object, such as a two-dimensional RGB (Red Green Blue, red, green and blue color mode) human body image. The human body image to be processed can be obtained in a variety of ways, such as a photo of a human body taken by a camera device, or an image of a human body downloaded from an image source, etc. The first part of the human body can include a part of the human body such as the head and hands, which has a simple structure and a small volume. The first part image is a partial image of the human body including the first part, such as a head image, a hand image, etc.

[0026] In this exemplary embodiment, after obtaining the human body image to be processed, the human body image to be processed can be segmented to obtain a first part image. The first part image can contain only one object, such as a head or a hand, or can include multiple objects, such as a pair of hands, etc. The present disclosure does not make specific limitations on this.

[0027] There are many ways to perform image segmentation on the human body image to be processed. For example, feature recognition can be performed on the human body image to be processed to determine the head or hand therefrom, and then image segmentation is performed based on the identified first part; or the area where the human body is located can be first identified from the human body image to be processed, and then the human body area can be divided into multiple sub-areas according to preset rules, and the first part image can be determined according to the sub-areas; the human body image to be processed can also be processed by a pre-trained neural network model to determine the key point information of the human body, and then image segmentation is performed based on the key point information, etc.

[0028] In an exemplary embodiment, in the above step S110, performing image segmentation on the human body image to be processed to obtain a first part image of a first part of the human body may include:

[0029] The human body image to be processed is processed by a third network to determine key point information of the human body;

[0030] The human body image to be processed is segmented according to the key point information to determine a first part image of a first part of the human body.

[0031] Among them, the third network can be a network for posture estimation of the human body in the human body image to be processed, such as a posture estimation network. This exemplary embodiment can use the human body image to be processed as input data of the third network. After being processed by the third network, key point information about the human body can be output. The key point can be a bone point or joint point used to reflect the structure of the human body, and the key point information can be the identification, position coordinates or posture data of the key point. For example, after the human body image to be processed is input into the third network, multiple joint points of the human body and the position information of each joint point can be obtained. Then, the human body image to be processed can be segmented according to the key point information, for example, the joint points at the head position and the hand position are respectively divided into different first part images, so as to segment the first part images of different first parts from the human body image to be processed.

[0032] Step S120: Process the first part image through the first network to obtain first three-dimensional feature information of the first part.

[0033] The first network refers to a network model used to process the head image and the hand image to perform three-dimensional reconstruction of the first part of the human body, such as a lightweight three-dimensional reconstruction network for the head and a lightweight three-dimensional reconstruction network for the hand. The first three-dimensional feature information refers to the three-dimensional feature information of the first part, such as the three-dimensional feature information of the head and the three-dimensional feature information of the hand. According to the first three-dimensional feature information, the first part of the human body can be three-dimensionally reconstructed to obtain a three-dimensional model of the first part, such as a three-dimensional model of the head and the three-dimensional model of the hand.

[0034] In this exemplary embodiment, a lightweight three-dimensional reconstruction network can be used to process different first part images to obtain first three-dimensional feature information, or different three-dimensional reconstruction networks can be used to process different first part images to obtain first three-dimensional feature information. For example, the same three-dimensional reconstruction network can be used to process head images and hand images, or a head three-dimensional reconstruction network can be used to process the first part image including the head, and a hand three-dimensional reconstruction network can be used to process the first part image including the hand, and so on.

[0035] In an exemplary embodiment, the first network includes a feature extraction subnetwork, a cost volume construction subnetwork, and a cost volume processing subnetwork; the above step S120 may include:

[0036] Processing the first part image through a feature extraction subnetwork to extract image feature data from the first part image;

[0037] The image feature data is processed through the cost body construction sub-network to construct the cost body;

[0038] The cost volume is processed by the cost volume processing subnetwork to obtain first three-dimensional feature information corresponding to the first part image.

[0039] In this exemplary embodiment, the first network may include a feature extraction subnetwork, a cost volume construction subnetwork, and a cost volume processing subnetwork, wherein the feature extraction subnetwork is used to extract image feature data from the part image, the cost volume construction subnetwork is used to construct a cost volume according to the image feature data, and the cost volume processing subnetwork is used to perform regularization processing on the cost volume, thereby determining the first three-dimensional feature information. In this exemplary embodiment, the first three-dimensional feature information may be a three-dimensional tensor with depth information.

[0040] This exemplary embodiment may use training data to pre-train the first network. Specifically, it may be to train the feature extraction subnetwork, the cost volume construction subnetwork and the cost volume processing subnetwork. For example, when training the feature extraction subnetwork, in order to ensure that the model can output three-dimensional feature information with depth information when applied, this exemplary embodiment may use sample images and corresponding reference images with depth information as input data for the feature extraction subnetwork to be trained. The difference between the output data obtained by processing the sample image and the reference image by the feature extraction subnetwork to be trained and the label data is used to optimize the parameters of the feature extraction subnetwork to be trained to obtain a trained feature extraction subnetwork.

[0041] In practical applications, after the first part image is input into the feature extraction subnetwork, the feature extraction subnetwork can process the first part image to extract multi-level image feature data from it. The image feature data refers to data that can reflect the image features, which can be the features of pixels in the image, or the features of objects included in the image, etc. Then, affine transformation can be applied to the image feature data under different viewing angles, and the processed image feature data can be used as input data of the cost volume construction subnetwork to construct the corresponding cost volume. Furthermore, the cost volume constructed from multiple stages from coarse to fine can be sent to the lightweight cost volume processing subnetwork to aggregate the depth information in the input image and improve its quality.

[0042] In an exemplary embodiment, after processing the feature data through the cost volume construction subnetwork to construct the cost volume, the above model generation method may further include:

[0043] Obtaining a first attention weight, and using the first attention weight to process the cost body to determine a first intermediate cost body;

[0044] Obtaining a second attention weight, and using the second attention weight to process the first intermediate cost body to determine a second intermediate cost body;

[0045] Obtaining a third attention weight, and using the third attention weight to process the second intermediate cost body to determine a target cost body;

[0046] Processing the cost volume through the cost volume regularization sub-network to obtain first three-dimensional feature information corresponding to the first part image may include:

[0047] The target cost volume is processed by the cost volume processing sub-network to obtain first three-dimensional feature information corresponding to the first part image.

[0048] In order to reduce the computational cost and control the amount of data while ensuring effective regularization of the cost body, this exemplary embodiment proposes a lightweight encoder-decoder structure combined with a deep separable convolution. In addition, in the decoding stage of the cost body regularization subnetwork, in order to reduce the impact of feature mismatch on the cost body regularization and ensure its accuracy, this exemplary embodiment adopts a three-dimensional hybrid attention module. The three-dimensional hybrid attention module can optimize the cost body, enhance important feature information and suppress the impact of abnormal data, remove noise and unnecessary data, and thus generate a purer and more accurate probability body.

[0049] Among them, the three-dimensional hybrid attention module can optimize the initial cost body through three attention modules respectively, and obtain the first intermediate cost body, the second intermediate cost body and the target cost body in turn. The first attention weight, the second attention weight and the third attention weight refer to the attention weights obtained in the three attention sub-modules in the three-dimensional hybrid attention module for processing the incoming cost body.

[0050] In this exemplary embodiment, the 3D hybrid attention module may include a channel attention submodule, a spatial attention submodule, and a depth attention submodule. By comprehensively considering the information of the entire cost body to calculate the attention weight, the 3D hybrid attention module can effectively enhance or suppress the similarity confidence at different depth levels, thereby significantly improving the accuracy of depth estimation. Figure 2 The structural diagram of a three-dimensional hybrid attention module in this exemplary embodiment is schematically shown. The initial cost body 210 can be processed by the channel attention sub-module 220 to obtain a first intermediate cost body 230; the spatial attention sub-module 240 can process the first intermediate cost body 230 to obtain a second intermediate cost body 250; the depth attention sub-module 260 can process the second intermediate cost body 250 to obtain a target cost body 270.

[0051] The channel attention submodule is used to perform attention operations on the channel information in the cost body. Specifically, the channel attention submodule can first perform a nonlinear dimensionality reduction operation on the cost body to obtain the channel attention weight W 1, that is, the first attention weight, the dimension reduction coefficient can be set according to actual needs, for example, set to 16. Then, by 1 Multiplying with the initial cost volume, we can get the first intermediate cost volume after channel refinement. In this channel attention submodule, the channel attention weight W 1 It is obtained by performing maximum pooling and average pooling operations on the cost volume, aggregating it through the convolution layer, and finally applying the activation function. It can be specifically expressed by the following formula:

[0052] W 1 =σ(M(MP(V))+M(AP(V)))

[0053] Among them, W 1 represents the first attention weight, σ is the activation function, M represents the multi-layer perceptron, MP represents the maximum pooling, AP represents the average pooling, and V represents the initial cost volume.

[0054] The spatial attention submodule can effectively perceive the cost volume information at different spatial positions. By using the spatial attention mechanism, the spatial attention submodule can identify the matching points that are most likely to correspond to the real geometric structure, thereby improving the accuracy of the matching. In this module, the size of the convolution kernel can be set according to the actual situation. For example, the present exemplary embodiment can use a convolution kernel of size 1×7×7 to filter the cost volume information along the spatial direction, aiming to enhance the feature representation while reducing noise interference. This design enables the convolution kernel to cover a larger spatial range, thereby capturing richer spatial context information. In this way, the spatial attention submodule can more accurately identify and emphasize key spatial positions related to deep matching, while suppressing information of low importance or that may be caused by noise, and providing accurate spatial information for subsequent processing stages. In the spatial attention submodule, the second attention weight W 2 It can be determined according to the first preset spatial convolution, the first intermediate cost body, the activation function, the maximum pooling and the average pooling. Taking the spatial convolution with a convolution kernel of 1×7×7 as an example, the second attention weight W 2 It can be determined, for example, by the following formula:

[0055] W 2 =σ(conv 1×7×7 (MP(V 1 )+AP(V 1 )))

[0056] Among them, W 2 represents the attention weight, σ is the activation function, conv 1×7×7 represents spatial convolution, MP represents maximum pooling, AP represents average pooling, V 1 Represents the first intermediate cost volume.

[0057] In order to improve the accuracy of matching information of different depth hypotheses at the same spatial position in three-dimensional reconstruction, this exemplary embodiment may also set a depth domain attention submodule, namely, a depth attention submodule. The depth attention module can enhance the depth information by using a preset convolution kernel, such as a 7×1×1 convolution kernel, to operate along the depth dimension of the second intermediate cost volume. The depth domain attention module can capture and distinguish features at different depths more finely, effectively enhancing the network's ability to recognize and process depth information, thereby improving the accuracy of matching and the overall quality of reconstruction. In the depth domain attention module, the third attention weight W 3 It can be determined according to the second preset spatial convolution, the second intermediate cost body, the activation function, the maximum pooling and the average pooling. Taking the spatial convolution with a convolution kernel of 7×1×1 as an example, the third attention weight W 3 It can be determined, for example, by the following formula:

[0058] W 3 =σ(conv 7×1×1 (MP(V 2 )+AP(V 2 )))

[0059] Among them, W 3 represents the attention weight, σ is the activation function, conv 7×1×1 represents spatial convolution, MP represents maximum pooling, AP represents average pooling, V 2 Represents the second intermediate cost volume.

[0060] Finally, these three attention modules can be cascaded to form a three-dimensional hybrid attention module. This module combines the effects of channel attention, spatial attention, and deep attention, and can comprehensively enhance the network's attention to information of different dimensions. Combining the attention weights of these three dimensions enables the first network to generate a final cost volume tensor that integrates multiple aspects of information.

[0061] The first intermediate cost volume, the second intermediate cost volume and the target cost volume can be determined by the following formula:

[0062] V 1 =V×W 2

[0063] V 2 =V 1 ×W 2

[0064] V 3 =V 2 ×W 3

[0065] Among them, V 3 Represents the target cost volume.

[0066] Figure 3 A schematic diagram of the network framework of a first network in this exemplary embodiment is shown. The first network may include a feature extraction subnetwork 310, a cost body construction subnetwork 320, and a cost body processing subnetwork 330. The image to be processed may first be input into the feature extraction subnetwork 310 for multi-scale feature extraction; then, the extracted image feature data is input into the cost body construction subnetwork 320 to construct the cost body; further, the constructed initial cost body is input into the cost body processing subnetwork 330 for cost body regularization processing to output the first three-dimensional feature information of the first part. The feature extraction subnetwork 310 may include one or more convolutional layers 311; the cost body construction subnetwork 320 may include multiple feature bodies 321; the cost body processing subnetwork 330 may include a lightweight three-dimensional convolution 331, a three-dimensional mixed attention module 332, and a lightweight three-dimensional deconvolution 333. It should be noted that Figure 3 This is only an example, and the actual network structure and layout can be set or constructed according to actual needs.

[0067] Step S130: Processing the human body image to be processed by a second network to determine second three-dimensional feature information of a second part of the human body.

[0068] Among them, the second three-dimensional feature information refers to the three-dimensional feature information of the second part of the human body. In this exemplary embodiment, the second part can be a part other than the first part, such as the entire body except the head and hands, etc. The second part can be three-dimensionally reconstructed according to the second three-dimensional feature information to obtain a corresponding three-dimensional reconstructed model.

[0069] Considering that the parts of the human body other than the head and hands may be more complex, and in some cases there may be occlusion or missing information on the back of the human body, in order to improve the accuracy and efficiency of human body reconstruction, this exemplary embodiment can use the first network to perform three-dimensional reconstruction on the first part of the head and hands, and use the second network to perform three-dimensional reconstruction on the second part other than the head and hands to generate second three-dimensional feature information. The second network can include multiple sub-networks to perform their respective tasks.

[0070] Specifically, in an exemplary embodiment, the second network includes a normal feature generation subnetwork, a two-dimensional feature generation subnetwork, and a three-dimensional feature generation subnetwork; the above step S130 may include:

[0071] The human body image to be processed is processed by a normal feature generation subnetwork to determine normal feature data about the human body in the human body image to be processed;

[0072] The image to be processed is processed by a two-dimensional feature generation sub-network, and two-dimensional feature data of the human body is extracted from the human body image to be processed;

[0073] Processing the image to be processed by a three-dimensional feature generation subnetwork to determine three-dimensional feature data of a second part of the human body from the human body image to be processed;

[0074] Second three-dimensional feature information of a second part of the human body is determined based on the normal feature data, the two-dimensional feature data and the three-dimensional feature data.

[0075] In this exemplary embodiment, the human body image to be processed can be processed separately through the processing process of the three sub-networks to obtain normal feature information, two-dimensional feature data and three-dimensional feature data of the second part of the human body. Further, the second three-dimensional feature information of the second part can be determined based on this information. Among them, the three-dimensional feature data of the second part can be the three-dimensional feature points of the second part, such as the three-dimensional feature points of the skeleton except the head and the hands. In this exemplary embodiment, the difference between the three-dimensional feature data and the second three-dimensional feature information is that the three-dimensional feature data refers to the three-dimensional feature body of the human body generated based on the three-dimensional feature generation subnetwork, and the second three-dimensional feature information refers to the three-dimensional feature information including multiple human body information obtained by fusing the normal feature data, two-dimensional feature data and three-dimensional feature data of the human body. The three-dimensional reconstruction of the human body based on the second three-dimensional feature information can improve the accuracy of the reconstruction of the three-dimensional human body model.

[0076] In this exemplary embodiment, normal feature information can be extracted from the human body image to be processed. For example, the human body image to be processed is processed by a pre-trained normal feature generation subnetwork to obtain normal feature information about the human body in the human body image to be processed. The normal feature information can be a normal map of the front and back of the human body. The normal map can include the direction or depth information of each point on the human body surface.

[0077] Then, the image to be processed is processed by a two-dimensional feature generation subnetwork, and two-dimensional feature data of the human body is extracted from the human body image to be processed. The two-dimensional feature generation subnetwork may be a pre-trained two-dimensional convolutional network, which may encode the image to be processed and generate two-dimensional feature data, such as two-dimensional image features.

[0078] Furthermore, the image to be processed can be processed by a three-dimensional feature generation subnetwork to determine the three-dimensional feature data of the second part of the human body from the human body image to be processed. The three-dimensional feature generation subnetwork can be a parameterized model, which can be used to construct a preliminary three-dimensional human body model based on the human body image to be processed and generate three-dimensional feature data.

[0079] Finally, feature aggregation processing may be performed on the normal feature data, the two-dimensional feature data, and the three-dimensional feature data, and they may be integrated together, and finally the second three-dimensional feature data of the second part may be determined by the implicit function reconstruction module.

[0080] In an exemplary embodiment, the above-mentioned processing of the image to be processed by the three-dimensional feature generation sub-network to determine the three-dimensional feature data of the second part of the human body from the human body image to be processed may include:

[0081] The human body image to be processed is processed through a three-dimensional feature generation sub-network to obtain three-dimensional feature data of the human body;

[0082] The three-dimensional feature data of the second part of the human body is extracted from the three-dimensional feature data of the human body.

[0083] Among them, the three-dimensional feature generation subnetwork can be a human body parameterized model (SMPL, Skinned Multi-Person Linear) for three-dimensional human body reconstruction. In this exemplary embodiment, the human body image to be processed can be used as input data of the three-dimensional feature generation subnetwork for processing, and the three-dimensional feature data of the entire human body in the human body image to be processed, such as the three-dimensional feature points of the human body, can be obtained first. Then, according to the preset human body structure information, the three-dimensional feature data of the second part of the human body can be extracted from the three-dimensional feature data of the human body. In this exemplary embodiment, the three-dimensional feature data of the second part can include structural prior information of the human skeleton, and the structural prior information of the skeleton can be the three-dimensional feature data used for three-dimensional reconstruction of the second part of the human body.

[0084] In this exemplary embodiment, the implementation of a human body parameterized model is taken as an example to specifically illustrate the processing of a human body image to be processed by a three-dimensional feature generation subnetwork.

[0085] The three-dimensional spatial layout information of the human skeleton can be determined through the human body parametric model. Specifically, this exemplary embodiment uses the two-dimensional projection coordinates of the three-dimensional information in the three-dimensional spatial layout information of the human skeleton to extract corresponding features from corresponding two-dimensional feature data, such as a two-dimensional feature map. In order to enhance the richness and accuracy of the features, this exemplary embodiment can splice the extracted two-dimensional features with the original three-dimensional features. Finally, these spliced ​​features are sent to a multi-layer perceptron to determine whether a specific sampling point is located inside or outside the human body. This method of combining two-dimensional and three-dimensional features effectively improves the ability to recognize human skeleton features.

[0086] The specific implementation of skeleton-based feature expression is as follows: the human body parameterized model M and the custom spatial query point q are used as input, and the relative distance r between the query point and the human body key point is calculated according to the human body key point obtained above. k . Where pk is the spatial feature point, G is the threshold function, and m is the number of key points of the human body.

[0087] First, calculate the spatial feature point p k The relative position r of each key point of the human body k,j Based on the proximity principle, that is, the spatial points are mainly affected by the adjacent human key points, and those human key points that are farther away and have weaker correlation are filtered out. In order to achieve this, a threshold function G can be defined, which determines whether to include the relative position feature of the human key point based on the distance between the query point and the human key point. Relative position r k,j It can be determined by the following formula:

[0088] r k，j =G(||p k -J j ||2∈R), j∈(1, m)

[0089] The skeleton can be expressed as:

[0090] r k =[r k，1 ,……,r k，m ]

[0091] The structural prior information based on the skeleton is expressed as follows:

[0092] H & (P ( )=r k ,k=1,……,N+K

[0093] Where N+K is the training sampling point obtained based on the skeleton sampling strategy, r k It is a skeleton-based feature expression.

[0094] In an exemplary embodiment, the above-mentioned processing of the image to be processed by the two-dimensional feature generation sub-network to extract the two-dimensional feature data of the human body from the human body image to be processed may include:

[0095] The human body image to be processed is processed through a two-dimensional feature generation sub-network to obtain initial two-dimensional feature data of the human body;

[0096] According to the three-dimensional feature data of the human body, the initial two-dimensional feature data is aligned to determine the two-dimensional feature data of the human body.

[0097] Among them, the two-dimensional feature generation subnetwork may refer to a two-dimensional convolutional network. In this exemplary embodiment, the human body image to be processed may be encoded through a two-dimensional convolutional network to generate initial two-dimensional feature data, which may be a two-dimensional image feature. The two-dimensional image feature includes a two-dimensional feature point, which has a pixel position. The three-dimensional feature data may refer to a three-dimensional feature point in a three-dimensional space. Usually, there may be a certain correspondence between the two-dimensional feature point and the three-dimensional feature point, for example, there is a back-projection relationship between the two-dimensional feature point and the three-dimensional feature point. In this exemplary embodiment, the back-projection relationship between the two-dimensional feature data of the human body and the three-dimensional feature data may be used to align the initial two-dimensional feature data and determine the aligned two-dimensional feature data of the human body.

[0098] In this exemplary embodiment, determining second three-dimensional feature information of a second part of the human body according to the normal feature data, the two-dimensional feature data, and the three-dimensional feature data may include:

[0099] The normal feature data, the two-dimensional feature data and the three-dimensional feature data are feature aggregated and then processed based on an implicit function to obtain second three-dimensional feature information.

[0100] Specifically, the acquired normal feature data and the three-dimensional feature data of the human body can be input into the aggregation model together with the two-dimensional feature data, for example, the front and back normal maps of the human body, the structural prior information of the skeleton, and the two-dimensional image features can be input into the aggregation model. The aggregation model can extract key features that can accurately reflect the human body shape through effective information fusion. Finally, these features can be decoded using implicit functions to generate a detailed three-dimensional human body model. This exemplary embodiment not only makes full use of the front and back normal maps and the prior information of the skeleton, but also realizes the accurate conversion from the two-dimensional image to the three-dimensional model through feature aggregation and implicit function decoding.

[0101] The formula of the pixel alignment implicit function of the human body 3D reconstruction method based on geometric prior information can be expressed as:

[0102]

[0103] Where x = π(X) is its 2D projection, F L , B L are the front normal map and the back normal map respectively, I is the input human body image to be processed, H & (P k ) is the structural prior information based on the skeleton, Z is the depth value in the coordinate space, g is the convolutional neural network used for feature extraction, and ψ is the sampling function, which is used to obtain the feature corresponding to x on the obtained feature map according to the projection point x.

[0104] Figure 4A schematic diagram of an architecture for determining second three-dimensional feature information is shown. Determining the second three-dimensional feature information of a second part of a human body may include the following steps: inputting a human body image 410 to be processed into a normal feature generation subnetwork 420 for processing to generate normal feature data; inputting the human body image 410 to be processed into a two-dimensional feature generation subnetwork 430 for processing to obtain initial two-dimensional image features, and processing the initial two-dimensional image features and the three-dimensional feature data of the human body through a feature alignment module 440 to determine the aligned two-dimensional image features; inputting the human body image 410 to be processed into a human body parameterized model estimation module 450 to determine the three-dimensional feature data of the human body, and then processing the three-dimensional feature data of the human body through a skeleton prior information acquisition module 460 to determine the three-dimensional feature data of the second part of the human body; further, inputting the normal feature data, the two-dimensional feature data and the three-dimensional feature data into a feature aggregation module 470 to obtain aggregated features; and then processing the aggregated features through an implicit function 480 to determine the second three-dimensional feature information 490 of the second part.

[0105] Based on the above-mentioned implementation modes, this exemplary embodiment enhances the model's ability to capture image normal information by introducing additional normal feature information, and integrates the three-dimensional feature bodies extracted by the human body parametric model, the pixel alignment features of the two-dimensional image, and the additional normal feature information as additional parameters of the deep implicit function. By utilizing the constraints provided by the multi-attribute prior, the reconstruction process of the three-dimensional human body is effectively standardized, the integrity and coherence of the reconstruction are ensured, and the accurate conversion from the two-dimensional image to the three-dimensional model is achieved.

[0106] Step S140: generating a three-dimensional human body model according to the first three-dimensional feature information and the second three-dimensional feature information.

[0107] Through the aforementioned steps, the first three-dimensional feature information and the second three-dimensional feature information can be obtained, and then they can be input into multiple modules for integration processing. The corresponding relationship between the mesh vertices is used to calculate the rotation or translation transformation according to the selected feature key points to integrate the first three-dimensional feature information with the second three-dimensional feature information, thereby integrating the first part and the second part into the human body model and sharing the topological mesh relationship of the model. This exemplary embodiment can convert the body reconstruction result in the form of a human body parameterized mesh into a full-body parameterized model mesh form including facial expressions and hand gestures, and then integrate the independent reconstruction results of each part together. The above integration process can not only ensure the integrity and coherence of the reconstruction, but also significantly improve the visual effect of the reconstruction result.

[0108] Figure 5The schematic diagram of the architecture of a model generation method in this exemplary embodiment is shown, which may specifically include: an image acquisition module 510, a posture estimation module 520, a first part reconstruction module 530, a second part reconstruction module 540, an integration module 550, and a human body model generation module 560. Among them, the image acquisition module 510 is used to acquire a human body image to be processed; the posture estimation module 520 is used to process the human body image to be processed through a third network to determine the key point information of the human body; and to perform image segmentation on the human body image to be processed according to the key point information to determine a first part image of a first part of the human body; the first part reconstruction module 530 may include a hand lightweight reconstruction network 531 and a head lightweight reconstruction network 532, which are used to process the first part image of the hand and the first part image of the head to obtain first three-dimensional feature information of the first part being the hand and the head; the second part reconstruction module 540 at least includes a normal feature generation module 541, a two-dimensional feature generation module 542, and a second part reconstruction module 543. The generation module 542, the three-dimensional feature generation module 543 and the implicit function module 544 process the human body image to be processed based on the normal feature generation subnetwork, the two-dimensional feature generation subnetwork and the three-dimensional feature generation subnetwork to obtain the normal feature information, the two-dimensional feature data and the three-dimensional feature data of the human body, and the normal feature information, the two-dimensional feature data and the three-dimensional feature data of the human body are processed by the implicit function module 544 to determine the second three-dimensional feature information of the second part; the information integration module 550 is used to integrate the first three-dimensional feature information and the second three-dimensional feature information; the human body model generation module 560 is used to generate a complete three-dimensional human body model according to the integrated feature information. It should be noted that Figure 5 This is only an exemplary description, wherein the structures of the hand lightweight reconstruction network 531 and the head lightweight reconstruction network 532 may be the same or different, and may be adjusted according to actual needs, and the present disclosure does not make any specific limitations on this.

[0109] Based on the above description, in this exemplary embodiment, a human body image to be processed is obtained, and image segmentation is performed on the human body image to be processed to obtain a first part image of a first part of the human body; the first part image is processed by a first network to obtain first three-dimensional feature information of the first part; the human body image to be processed is processed by a second network to determine second three-dimensional feature information of a second part of the human body; and a three-dimensional human body model is generated based on the first three-dimensional feature information and the second three-dimensional feature information. On the one hand, this exemplary embodiment proposes a new model generation method, which divides the human body into a first part and a second part based on the human body image to be processed, and determines the first three-dimensional feature information and the second three-dimensional feature information, and generates a three-dimensional human body model according to the first three-dimensional feature information and the second three-dimensional feature information, and can perform three-dimensional reconstruction on the first part and the second part of the human body respectively, and integrate the three-dimensional reconstruction results of the first part and the second part to obtain a complete three-dimensional human body model. Compared with directly reconstructing based on the human body image to be processed, this method of performing three-dimensional reconstruction and merging of different parts of the human body reduces the influence of interference information or partial information missing that may exist in the human body image to be processed on the generation of the three-dimensional human body model, and improves the efficiency and accuracy of the reconstruction of the three-dimensional human body model; on the other hand, this exemplary embodiment adopts different methods to perform three-dimensional reconstruction on the first part and the second part, and can provide targeted three-dimensional reconstruction methods for different parts, further ensuring the accuracy and detail expression of the reconstruction of the three-dimensional human body model.

[0110] The exemplary embodiments of the present disclosure also provide a model generation device. Figure 6 The device 600 may include a human body image acquisition module 610, which is used to acquire a human body image to be processed and perform image segmentation on the human body image to be processed to obtain a first part image of a first part of a human body; a first information determination module 620, which is used to process the first part image through a first network to obtain first three-dimensional feature information of the first part; a second information determination module 630, which is used to process the human body image to be processed through a second network to determine second three-dimensional feature information of a second part of the human body; and a human body model generation module 640, which is used to generate a three-dimensional human body model of the human body according to the first three-dimensional feature information and the second three-dimensional feature information.

[0111] In an exemplary embodiment, a human body image acquisition module includes: a key point information determination unit, used to process the human body image to be processed through a third network to determine the key point information of the human body; an image segmentation unit, used to perform image segmentation on the human body image to be processed according to the key point information to determine a first part image of a first part of the human body.

[0112] In an exemplary embodiment, the first network includes a feature extraction subnetwork, a cost volume construction subnetwork, and a cost volume processing subnetwork; the first information determination module includes: a feature extraction unit, which is used to process the first part image through the feature extraction subnetwork and extract image feature data from the first part image; a cost volume construction unit, which is used to process the image feature data through the cost volume construction subnetwork and construct a cost volume; and a cost volume processing unit, which is used to process the cost volume through the cost volume processing subnetwork to obtain first three-dimensional feature information corresponding to the first part image.

[0113] In an exemplary embodiment, the image feature data is processed by a cost body construction subnetwork. After the cost body is constructed, the model generation device also includes: a first weight processing unit, which is used to obtain a first attention weight, and use the first attention weight to process the cost body to determine a first intermediate cost body; a second weight processing unit, which is used to obtain a second attention weight, and use the second attention weight to process the first intermediate cost body to determine a second intermediate cost body; a third weight processing unit, which is used to obtain a third attention weight, and use the third attention weight to process the second intermediate cost body to determine a target cost body; and a cost body processing unit, which is used to process the target cost body through a cost body processing subnetwork to obtain first three-dimensional feature information corresponding to the first part image.

[0114] In an exemplary embodiment, the second network includes a normal feature generation subnetwork, a two-dimensional feature generation subnetwork, and a three-dimensional feature generation subnetwork; the second information determination module includes: a normal feature determination unit, which is used to process the human body image to be processed through the normal feature generation subnetwork, and determine the normal feature data about the human body in the human body image to be processed; a two-dimensional feature determination unit, which is used to process the image to be processed through the two-dimensional feature generation subnetwork, and extract the two-dimensional feature data of the human body from the human body image to be processed; a three-dimensional feature determination unit, which is used to process the image to be processed through the three-dimensional feature generation subnetwork, and determine the three-dimensional feature data of the second part of the human body from the human body image to be processed; and a second three-dimensional feature determination unit, which is used to determine the second three-dimensional feature information of the second part of the human body based on the normal feature data, the two-dimensional feature data and the three-dimensional feature data.

[0115] In an exemplary embodiment, the three-dimensional feature determination unit includes: a three-dimensional feature network processing subunit, used to process the human body image to be processed through a three-dimensional feature generation subnetwork to obtain three-dimensional feature data of the human body; and a data extraction subunit, used to extract three-dimensional feature data of a second part of the human body from the three-dimensional feature data of the human body.

[0116] In an exemplary embodiment, the two-dimensional feature determination unit includes: an initial two-dimensional feature data acquisition subunit, which is used to process the human body image to be processed through a two-dimensional feature generation subnetwork to obtain initial two-dimensional feature data of the human body; and a two-dimensional feature alignment subunit, which is used to align the initial two-dimensional feature data according to the three-dimensional feature data of the human body to determine the two-dimensional feature data of the human body.

[0117] The specific details of each module / unit in the above-mentioned device have been described in detail in the embodiments of the method part. The undisclosed details can be found in the embodiments of the method part, so they will not be repeated here.

[0118] Exemplary embodiments of the present disclosure also provide an electronic device capable of implementing the above method.

[0119] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0120] Refer to the following Figure 7 An electronic device 700 according to such an exemplary embodiment of the present disclosure is described below. Figure 7 The electronic device 700 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0121] like Figure 7 As shown, the electronic device 700 is in the form of a general computing device. The components of the electronic device 700 may include, but are not limited to: the at least one processing unit 710, the at least one storage unit 720, a bus 730 connecting different system components (including the storage unit 720 and the processing unit 710), and a display unit 740.

[0122] The storage unit stores program codes, which can be executed by the processing unit 710, so that the processing unit 710 executes the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 710 can execute Figure 1 Steps shown etc.

[0123] The storage unit 720 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 721 and / or a cache memory unit 722 , and may further include a read-only memory unit (ROM) 723 .

[0124] The storage unit 720 may also include a program / utility 724 having a set (at least one) of program modules 725, such program modules 725 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0125] Bus 730 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0126] The electronic device 700 may also communicate with one or more external devices 800 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 700, and / or communicate with any device that enables the electronic device 700 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 750. Furthermore, the electronic device 700 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 760. As shown, the network adapter 760 communicates with other modules of the electronic device 700 via a bus 730. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0127] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the exemplary embodiment of the present disclosure.

[0128] The exemplary embodiments of the present disclosure also provide a computer-readable storage medium on which a program product capable of implementing the above-mentioned method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above-mentioned "Exemplary Method" section of the present specification.

[0129] The exemplary embodiments of the present disclosure also provide a program product for implementing the above method, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0130] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0131] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0132] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0133] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0134] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0135] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to exemplary embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0136] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0137] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A model generation method, characterized in that: include: Acquire a human body image to be processed, and perform image segmentation on the human body image to be processed to obtain a first part image of a first part of the human body; Processing the first part image through a first network to obtain first three-dimensional feature information of the first part; Processing the human body image to be processed through a second network to determine second three-dimensional feature information of a second part of the human body; A three-dimensional human body model of the human body is generated according to the first three-dimensional feature information and the second three-dimensional feature information.

2. The method according to claim 1, characterized in that The step of performing image segmentation on the human body image to be processed to obtain a first part image of a first part of the human body includes: Processing the human body image to be processed through a third network to determine key point information of the human body; The human body image to be processed is segmented according to the key point information to determine a first part image of the first part of the human body.

3. The method according to claim 1, characterized in that The first network includes a feature extraction subnetwork, a cost volume construction subnetwork, and a cost volume processing subnetwork; the first part image is processed by the first network to obtain first three-dimensional feature information of the first part, including: Processing the first part image through a feature extraction sub-network to extract image feature data from the first part image; Processing the image feature data through a cost body construction sub-network to construct a cost body; The cost volume is processed by a cost volume processing subnetwork to obtain first three-dimensional feature information corresponding to the first part image.

4. The method according to claim 3, characterized in that The image feature data is processed by constructing a sub-network through a cost body. After the cost body is constructed, the method further includes: Obtaining a first attention weight, and using the first attention weight to process the cost body to determine a first intermediate cost body; Acquire a second attention weight, and use the second attention weight to process the first intermediate cost volume to determine a second intermediate cost volume; Obtaining a third attention weight, and using the third attention weight to process the second intermediate cost volume to determine a target cost volume; The step of processing the cost volume through the cost volume regularization sub-network to obtain first three-dimensional feature information corresponding to the first part image includes: The target cost volume is processed by a cost volume processing subnetwork to obtain first three-dimensional feature information corresponding to the first part image.

5. The method according to claim 1, characterized in that The second network includes a normal feature generation subnetwork, a two-dimensional feature generation subnetwork, and a three-dimensional feature generation subnetwork; The step of processing the human body image to be processed by a second network to determine second three-dimensional feature information of a second part of the human body includes: Processing the human body image to be processed by a normal feature generation subnetwork to determine normal feature data about the human body in the human body image to be processed; Processing the image to be processed by a two-dimensional feature generation subnetwork, and extracting two-dimensional feature data of the human body from the human body image to be processed; Processing the image to be processed by a three-dimensional feature generation subnetwork, and determining three-dimensional feature data of a second part of the human body from the human body image to be processed; Second three-dimensional feature information of a second part of the human body is determined according to the normal feature data, the two-dimensional feature data and the three-dimensional feature data.

6. The method according to claim 5, characterized in that The step of processing the image to be processed by a three-dimensional feature generation subnetwork to determine the three-dimensional feature data of the second part of the human body from the human body image to be processed includes: Processing the human body image to be processed by the three-dimensional feature generation subnetwork to obtain three-dimensional feature data of the human body; The three-dimensional feature data of the second part of the human body is extracted from the three-dimensional feature data of the human body.

7. The method according to claim 6, characterized in that The step of processing the image to be processed by a two-dimensional feature generation subnetwork to extract two-dimensional feature data of the human body from the human body image to be processed includes: Processing the human body image to be processed by a two-dimensional feature generation subnetwork to obtain initial two-dimensional feature data of the human body; According to the three-dimensional feature data of the human body, the initial two-dimensional feature data is aligned to determine the two-dimensional feature data of the human body.

8. A model generation device, characterized in that: include: A human body image acquisition module, used for acquiring a human body image to be processed, and performing image segmentation on the human body image to be processed to obtain a first part image of a first part of the human body; A first information determination module, configured to process the first part image through a first network to obtain first three-dimensional feature information of the first part; A second information determination module, used for processing the to-be-processed human body image through a second network to determine second three-dimensional feature information of a second part of the human body; A human body model generation module is used to generate a three-dimensional human body model of the human body according to the first three-dimensional feature information and the second three-dimensional feature information.

9. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium comprising a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images