Three-dimensional human body reconstruction method and device

Abstract

The present invention provides a three-dimensional human body reconstruction method and device, the method comprising: obtaining an image of a target human body instance as a target image; inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; wherein the three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each preset sampling point is evenly distributed in the three-dimensional space. The three-dimensional human body reconstruction method and device provided by the present invention can obtain the image of the target human body instance without constraints without relying on an image sensor with calibrated camera parameters, so that three-dimensional human body reconstruction can be performed more simply and efficiently, and three-dimensional human body reconstruction applicable to various scenes can be realized more widely.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a three-dimensional human body reconstruction method and device. Background Art

[0002] 3D human body reconstruction is a key issue in the field of computer graphics and computer vision. Currently, 3D human body reconstruction has been widely used in various fields such as virtual reality AR / VR, film and television entertainment, virtual try-on, and industrial manufacturing.

[0003] Existing 3D human body reconstruction methods can construct a 3D model of the human body instance based on a 2D image including the human body instance. However, existing 3D human body reconstruction methods have strong limitations. For example, the 2D image needs to be acquired by an image sensor with calibrated camera parameters; or the 2D image needs to be acquired by an expensive laser scanner or a multi-camera array system, which makes it difficult for existing 3D human body reconstruction methods to be widely applied to various scenarios. Summary of the invention

[0004] The present invention provides a three-dimensional human body reconstruction method and device, which are used to solve the defect that the prior art is difficult to be widely applied to various scenes, and achieve wider application to various scenes.

[0005] The present invention provides a three-dimensional human body reconstruction method, comprising:

[0006] Acquire an image of a target human instance as a target image;

[0007] Inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model;

[0008] The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance;

[0009] The three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each of the preset sampling points is evenly distributed in the three-dimensional space.

[0010] According to a 3D human body reconstruction method provided by the present invention, the 3D model of the sample human body instance includes a first sample model, a second sample model and a third sample model of the sample human body instance, the first sample model and the third sample model have the same posture, the posture of the second sample model is a preset standard posture, and the first sample model and the third sample model are obtained in different ways; the sample image includes the first sample model;

[0011] The three-dimensional human body reconstruction model includes: an image feature extraction layer, a skin parameter extraction layer, an attention mechanism layer and a result output layer;

[0012] Correspondingly, the step of inputting the target image into a three-dimensional human body reconstruction model and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model specifically includes:

[0013] Inputting the target image and the position information of each preset sampling point into the image feature extraction layer, and obtaining the image features corresponding to each preset sampling point output by the image feature extraction layer;

[0014] Inputting the position information and the corresponding image features of each preset sampling point into the skin parameter extraction layer, and obtaining the skin parameters corresponding to each preset sampling point output by the skin parameter extraction layer;

[0015] Inputting the position information, corresponding image features and skinning parameters of each preset sampling point into the attention mechanism layer, and obtaining identification information of each preset sampling point output by the attention mechanism layer, wherein the identification information is used to indicate whether each preset sampling point is located inside the three-dimensional human body model of the target human body instance or not;

[0016] The identification information of each preset sampling point is input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer, wherein the three-dimensional human body reconstruction result includes a first model of the target human body instance, and the posture of the first model is the standard posture.

[0017] According to a three-dimensional human body reconstruction method provided by the present invention, the image feature extraction layer comprises: a data processing unit, a position updating unit and a feature extraction unit;

[0018] Correspondingly, the step of inputting the target image and the position information of each preset sampling point into the image feature extraction layer to obtain the image feature corresponding to each preset sampling point specifically includes:

[0019] Inputting the target image into the data processing unit, obtaining a second model of the target human body instance output by the data processing unit, wherein the second model carries image posture parameters corresponding to the target image;

[0020] Inputting the target image, the image posture parameter and the position information of each preset sampling point into the position updating unit, obtaining the updated position information of each preset sampling point output by the position updating unit, wherein the updated position information corresponds to the real posture of the target human body instance in the target image;

[0021] The target image and the updated position information of each preset sampling point are input into the feature extraction unit, and the image feature corresponding to each preset sampling point output by the feature extraction unit is obtained.

[0022] According to a 3D human body reconstruction method provided by the present invention, after inputting the position information of each preset sampling point, the corresponding image features and the skinning parameters into the attention mechanism layer and obtaining the identification information of each preset sampling point output by the attention mechanism layer, the method further includes:

[0023] The identification information of each preset sampling point and the second model are input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer, wherein the three-dimensional human body reconstruction result includes a third model of the target human body instance, and the posture of the third model is the real posture of the target human body instance in the target image.

[0024] According to a 3D human body reconstruction method provided by the present invention, the loss function of the 3D human body reconstruction model includes: a skin parameter loss function;

[0025] The skin parameter loss function is determined based on the predicted skin parameters and skin parameter labels corresponding to the sample sampling points; the predicted skin parameters are input into the skin parameter extraction model in training with the position information and the corresponding image features of the sample sampling points, and are output by the skin parameter extraction model in training; the image features corresponding to the sample sampling points are determined based on the sample image.

[0026] According to a 3D human body reconstruction method provided by the present invention, the loss function of the 3D human body reconstruction model includes: a classification loss function;

[0027] The classification loss function is determined based on the predicted identification information and identification information label of the sample sampling point; the predicted identification information of the sample sampling point is input into the attention mechanism layer in training by the position information of the sample sampling point, the corresponding image features and the predicted skinning parameters, and output by the attention mechanism layer in training.

[0028] The present invention also provides a three-dimensional human body reconstruction device, comprising:

[0029] An image acquisition module, used for acquiring an image of a target human instance as a target image;

[0030] A three-dimensional reconstruction module, used for inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model;

[0031] The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance;

[0032] The three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each of the preset sampling points is evenly distributed in the three-dimensional space.

[0033] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-mentioned three-dimensional human body reconstruction methods are implemented.

[0034] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above-mentioned three-dimensional human body reconstruction methods are implemented.

[0035] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any one of the above-mentioned three-dimensional human body reconstruction methods are implemented.

[0036] The 3D human body reconstruction method and device provided by the present invention obtain an image of a target human body instance as a target image, input the target image into a 3D human body reconstruction model, the 3D human body reconstruction model obtains image features of each preset sampling point based on the target image, obtains skin parameters corresponding to each preset sampling point based on the image features corresponding to each preset sampling point, and performs 3D human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, thereby obtaining a 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model, the 3D human body reconstruction model is obtained after training based on the 3D model of the sample human body instance and the sample image including the 3D model of the sample human body instance, each preset sampling point is evenly distributed in the 3D space, and the image of the target human body instance can be obtained without constraints without relying on an image sensor with calibrated camera parameters, an expensive laser scanner or a multi-camera array system, thereby being able to perform 3D human body reconstruction more simply and efficiently, and being able to achieve 3D human body reconstruction that is more widely applicable to various scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 It is a schematic flow chart of the three-dimensional human body reconstruction method provided by the present invention;

[0039] Figure 2 is a schematic structural diagram of a three-dimensional human body reconstruction model in the three-dimensional human body reconstruction method provided by the present invention;

[0040] Figure 3 It is a schematic diagram of the process of obtaining a sample image in the three-dimensional human body reconstruction method provided by the present invention;

[0041] Figure 4 It is a schematic diagram of the process of training a three-dimensional human body reconstruction model in the three-dimensional human body reconstruction method provided by the present invention;

[0042] Figure 5 is a schematic structural diagram of a three-dimensional human body reconstruction device provided by the present invention;

[0043] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] In the description of the invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0046] Figure 1 3D human body reconstruction method provided by the present invention. Figure 1 The three-dimensional human body reconstruction method of the present invention is described. Figure 1 As shown, the method includes: step 101, obtaining an image of a target human instance as a target image.

[0047] Specifically, the target human body instance is a human body instance that needs to be reconstructed in three dimensions.

[0048] The target image can be obtained by using a conventional image sensor or an electronic device with an image acquisition function. For example, a mobile phone with an image acquisition function can be used to obtain an image of the target human body instance as the target image; or a conventional camera can be used to obtain an image of the target human body instance as the target image.

[0049] It should be noted that in the embodiments of the present invention, when acquiring the image of the target human body instance, there is no need to be constrained by camera parameters and other conditions. Compared with the traditional 3D human body reconstruction method, which needs to rely on image sensors with calibrated camera parameters, expensive laser scanners or multi-camera array systems to acquire the image of the target human body instance, the 3D human body reconstruction method provided by the present invention can acquire the image of the target human body instance without constraints, thereby improving the scope of application of the 3D human body reconstruction method, so that the 3D human body reconstruction method provided by the present invention can be more widely applied to various scenarios.

[0050] It should be noted that the number of images of the target human instance can be one or more. In the case of obtaining multiple frames of images or a video of the target human instance, each frame of the target human instance or each frame of the video can be used as the target image. Accordingly, the number of target images can be one or more.

[0051] Step 102: input the target image into the 3D human body reconstruction model, and obtain the 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model.

[0052] The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance.

[0053] The three-dimensional human body reconstruction model is used to reconstruct the three-dimensional human body of the target human body instance based on the image features and skin parameters corresponding to each preset sampling point. The image features are determined based on the target image, and the skin parameters are determined based on the corresponding image features. The preset sampling points are evenly distributed in the three-dimensional space.

[0054] It should be noted that before inputting the target image into the three-dimensional human body reconstruction model and obtaining the three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model, the three-dimensional human body reconstruction model can be trained based on the three-dimensional model of the sample human body instance and the sample image including the three-dimensional model of the above-mentioned sample human body instance to obtain a trained three-dimensional human body reconstruction model.

[0055] Optionally, the three-dimensional human body reconstruction model can be trained in the following manner: first, the three-dimensional model of the sample human body instance can be obtained in a variety of ways, for example, the three-dimensional model of the sample human body instance can be obtained based on a traditional laser scanner, SMPL algorithm, etc. Among them, the number of three-dimensional models of the sample human body instance can be multiple. Secondly, the three-dimensional model of the sample human body instance can be obtained as a sample image. Finally, based on the above sample image and the three-dimensional model of the above sample human body instance, the three-dimensional human body reconstruction model can be trained to obtain a trained three-dimensional human body reconstruction model.

[0056] After obtaining the trained 3D human body reconstruction model, the target image can be input into the trained 3D human body reconstruction model.

[0057] The trained 3D human body reconstruction model can obtain the image features corresponding to each preset sampling point based on the target image, obtain the skin parameters corresponding to each preset sampling point based on the image features corresponding to each preset sampling point, perform 3D human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, and obtain and output the 3D human body reconstruction result of the target human body instance.

[0058] It should be noted that all preset sampling points are evenly distributed in the three-dimensional space, and the distance between any two adjacent preset sampling points is equal, forming a three-dimensional dot matrix. The greater the density of the three-dimensional dot matrix, the more accurate the three-dimensional human body reconstruction result of the target human body instance is.

[0059] Optionally, in the embodiment of the present invention, each preset sampling point constitutes a three-dimensional cubic lattice, and the three-dimensional cubic lattice includes 256*256*256 preset sampling points.

[0060] The embodiment of the present invention obtains an image of a target human instance as a target image, and inputs the target image into a 3D human body reconstruction model. The 3D human body reconstruction model obtains image features of each preset sampling point based on the target image, obtains skin parameters corresponding to each preset sampling point based on the image features corresponding to each preset sampling point, and performs 3D human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, thereby obtaining a 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model. The 3D human body reconstruction model is obtained after training based on a 3D model of a sample human body instance and a sample image including the 3D model of the sample human body instance. The preset sampling points are evenly distributed in a 3D space, and the image of the target human body instance can be freely obtained without relying on an image sensor with calibrated camera parameters, an expensive laser scanner or a multi-camera array system, thereby enabling simpler and more efficient 3D human body reconstruction, and enabling a wider range of 3D human body reconstruction applicable to various scenarios.

[0061] Based on the contents of the above embodiments, the three-dimensional model of the sample human body instance includes a first sample model, a second sample model and a third sample model of the sample human body instance, the first sample model and the third sample model have the same posture, the posture of the second sample model is a preset standard posture, and the first sample model and the third sample model are acquired in different ways; the sample image includes the first sample model.

[0062] Figure 2 3D human body reconstruction model in the 3D human body reconstruction method provided by the present invention. Figure 2 As shown, the three-dimensional human body reconstruction model includes: image feature extraction layer, skin parameter extraction layer, attention mechanism layer and result output layer.

[0063] It should be noted that a sample space coordinate system can be constructed in the three-dimensional space of the three-dimensional model including the sample target human body instance, and the coordinates of any point in the sample space coordinate system can be used as the position information of the point.

[0064] Specifically, the sample human body instance can be scanned based on a laser scanner to obtain a first sample model of the sample human body instance. The third sample model of the sample human body instance can be constructed based on the SMPL algorithm. The first sample model and the third sample model need to be sealed models without holes. The third sample model constructed based on the SMPL algorithm carries the skin parameters corresponding to any point on the surface of the third sample model and the sample image posture parameters corresponding to the sample image.

[0065] It should be noted that the posture of the sample human body instance when the laser scanner scans the sample human body instance is the sample posture of the sample human body instance. The postures of the first sample model and the third sample model are both the above sample postures.

[0066] It should be noted that the first sample model may include the clothes wrinkles of the sample human body instance. The third sample model does not include the clothes wrinkles of the sample human body instance.

[0067] Optionally, images in different scenes can also be obtained as background images. In order to make the three-dimensional human body reconstruction model more generalizable, the obtained background images should reflect common scenes in daily life as realistically and diversely as possible.

[0068] Figure 3 FIG. 1 is a flow chart of obtaining a sample image in the three-dimensional human body reconstruction method provided by the present invention. Figure 3 As shown, the first sample human body model is subjected to data processing to obtain sample camera projection parameters corresponding to the first sample human body model and the first sample human body model after data processing. Any background image is selected, and after the background image is cropped and scaled to a preset size of 512*512, the first sample human body model after data processing can be rendered into the background image through orthogonal projection to obtain a sample image including the first sample human body model, and a silhouette image, a depth map and a normal vector map of the first sample human body model are simultaneously rendered.

[0069] After rendering the first sample human body model after data processing to the above-mentioned background image, it can be determined whether all background images are selected. If not all background images are selected, you can select any background image again. After processing the above-mentioned background image, render the first sample human body model after data processing to the above-mentioned background image until all background images are selected to obtain a sample image set.

[0070] Optionally, data processing of the first sample human body model specifically includes the following steps: first, normalizing the first sample model, translating the first sample model in the sample space coordinate system so that the center point of the first sample model coincides with a preset base point in the sample space coordinate system. The center point of the first sample model is determined based on the geometric features of the first sample model, and the preset base point can be the origin of the sample space coordinate system. Secondly, scaling the first sample model to [-0.5, 0.5]; finally, randomly rotating, translating and scaling the first sample model along the y-axis.

[0071] By using the SCANimate model, a second sample model of a standard posture of a sample human body instance can be constructed based on the first sample model, and the skinning parameters corresponding to any point on the second sample model can be obtained.

[0072] Figure 4 FIG. 1 is a flow chart of training a 3D human body reconstruction model in the 3D human body reconstruction method provided by the present invention, such as Figure 4 As shown, after obtaining the first sample model, the second sample model, the third sample model and the sample image including the first sample model of the sample human body instance, the three-dimensional human body reconstruction model can be trained based on the above-mentioned first sample model, the second sample model, the third sample model and the sample image.

[0073] It should be noted that since infinite precision sampling is possible in three-dimensional space, training a three-dimensional human body reconstruction model based on uniform sampling is very inefficient. Although sampling within a certain distance range of the three-dimensional model surface of the sample human body instance can enable the three-dimensional human body reconstruction model in training to better find the decision boundary, this sampling method will cause the three-dimensional human body reconstruction model to overfit. Therefore, in the embodiment of the present invention, uniform sampling is performed on the three-dimensional model surface of the sample human body instance and the three-dimensional space near the surface at a certain ratio, so that the three-dimensional human body reconstruction model can converge quickly and have better generalization ability.

[0074] Specifically, the second sample model is translated in the sample space coordinate system so that the center point of the second sample model coincides with a preset base point in the sample space coordinate system.

[0075] Determine M on the surface of the second sample model 1 The first sample sampling points are uniformly determined in the three-dimensional space near the surface of the second sample model M 2 The first sample sampling point and the second sample sampling point are both used as sample sampling points. 1 >M 2 . Optionally, M 1 :M=16:1.

[0076] After the sample sampling points are determined, the first sample sampling point among the sample sampling points may be labeled as 1 as an identification information label of the first sample sampling point, indicating that the first sample sampling point is located on the surface of the second sample model. The second sample sampling point among the sample sampling points may be labeled as 0 as an identification information label of the second sample sampling point, indicating that the second sample sampling point is not located on the surface of the second sample model. Thus, the identification information labels of the sample sampling points may be obtained.

[0077] After the sample sampling points are determined, the coordinates of the first sample sampling point among the sample sampling points can be obtained, and a normal distribution noise disturbance with a mean of 0 and a variance of 0.05 is added to the coordinates of the first sample sampling point as the position information of the first sample sampling point. The coordinates of the second sample sampling point among the sample sampling points can be obtained as the position information of the second sample sampling point, thereby obtaining the position information of the sample sampling points.

[0078] After obtaining the position information of the sample sampling point, the position of the sample sampling point can be updated based on the third sample model to obtain the updated position information of the sample sampling point. The updated position information of the sample sampling point can correspond to the posture of the first sample model.

[0079] Specifically, the third sample model is translated in the sample space coordinate system so that the center point of the third sample model and the center point of the second sample model coincide with the preset base point in the sample space coordinate system, so that the second sample model, the third sample model and the sample sampling point are located in the same three-dimensional space.

[0080] For any sample sampling point, determine the closest point from the sample sampling point to the surface of the third sample model as the associated sampling point corresponding to the sample sampling point, and use the skin parameter corresponding to the associated sampling point as the skin parameter label corresponding to the sample sampling point.

[0081] The skin parameter labels, position information and sample image posture parameters corresponding to the sample sampling points are input into a linear hybrid skinning algorithm based on key point drive, and the updated position information of the sample sampling points is calculated.

[0082] The image feature extraction layer in the three-dimensional human body reconstruction model can be constructed based on the Hourglass network. The updated position information of the sample sampling points, the sample image and the camera projection parameters corresponding to the first sample human body model are input into the image feature extraction layer. The image feature extraction layer can project the sample sampling points onto the image features based on the above sample camera parameters and the updated position information of the sample sampling points, so as to obtain the image features corresponding to the sample sampling points output by the image feature extraction layer.

[0083] The skin parameter extraction layer in the 3D human body reconstruction model can be constructed based on the MLP network. After obtaining the image features corresponding to the sample sampling points, the position information and the corresponding image features of the sample sampling points can be input into the skin parameter extraction layer in training to obtain the predicted skin parameters corresponding to the sample sampling points output by the skin parameter extraction layer in training.

[0084] Optionally, the skin parameter extraction layer can obtain the predicted skin parameters corresponding to the sample sampling points based on the following formula:

[0085]

[0086] Among them, S represents the skin parameter extraction layer; p represents the location information of the sample sampling point, I represents the sample image; Φ represents the image feature extraction model, Represents an MLP network.

[0087] Based on the predicted skin parameters and skin parameter labels corresponding to the sample sampling points, the skin parameter extraction layer can be trained to obtain a trained skin parameter extraction layer.

[0088] The attention mechanism layer in the three-dimensional human body reconstruction model can be constructed based on the Transformer network. The position information of the sample sampling point, the corresponding predicted skinning parameters and the image features are input into the attention mechanism layer in training, and the predicted identification information of the sample sampling point output by the attention mechanism layer in training can be obtained. Among them, the predicted identification information of the sample sampling point can be 1 or 0, which is used to indicate that the sample sampling point is located or not located inside the second sample model. When the predicted identification information of the sample sampling point is 1, it means that the sample sampling point is located inside the second sample model; when the predicted identification information of the sample sampling point is 0, it means that the sample sampling point is not located inside the second sample model.

[0089] Optionally, the attention mechanism layer can obtain the predicted identification information of the sample sampling point based on the following formula:

[0090] F(p,I)=H[Φ(p,I),S(p,I),p]

[0091] Among them, H represents the attention mechanism model.

[0092] Based on the predicted identification information and identification information labels of the sample sampling points, the attention mechanism layer can be trained to obtain a trained attention mechanism layer.

[0093] Accordingly, the target image is input into the three-dimensional human body reconstruction model to obtain the three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model, specifically including: inputting the target image and the position information of each preset sampling point into the image feature extraction layer, and obtaining the image features corresponding to each preset sampling point output by the image feature extraction layer.

[0094] Specifically, a spatial coordinate system may be established in the three-dimensional space, and the position information of the preset sampling point may be represented by the coordinates of the preset sampling point in the spatial coordinate system (referred to as the coordinates of the preset sampling point).

[0095] After acquiring the target image, the target image and the coordinates of each preset sampling point may be input into an image feature extraction layer in the three-dimensional human body reconstruction model.

[0096] The image feature extraction layer constructed based on the Hourglass network can extract features of the target image, and can obtain the image features corresponding to each preset sampling point through numerical calculation methods based on the camera projection parameters corresponding to the first sample human body model, the coordinates of each preset sampling point and the extracted image features, so as to obtain the image features corresponding to each preset sampling point output by the image feature extraction layer.

[0097] The position information of each preset sampling point and the corresponding image feature are input into the skin parameter extraction layer, and the skin parameter corresponding to each preset sampling point output by the skin parameter extraction layer is obtained.

[0098] Specifically, after obtaining the image feature corresponding to each preset sampling point, the position information of each preset sampling point and the corresponding image feature may be input into the trained skin parameter extraction layer.

[0099] The input skin parameter extraction layer constructed based on the MLP network can obtain the skin parameters corresponding to each preset sampling point based on the position information of each preset sampling point and the corresponding image features, so as to obtain the skin parameters corresponding to each preset sampling point output by the skin parameter extraction layer.

[0100] The position information, corresponding image features and skinning parameters of each preset sampling point are input into the attention mechanism layer, and the identification information of each preset sampling point output by the attention mechanism layer is obtained, where the identification information is used to indicate whether the preset sampling point is located inside the three-dimensional human body model of the target human instance or not.

[0101] Specifically, after obtaining the skin parameters corresponding to each preset sampling point, the position information of each preset sampling point, the corresponding image features and the skin parameters can be input into the trained attention mechanism layer.

[0102] The attention mechanism layer constructed based on the Transformer network can perform network forward calculation based on the position information of each preset sampling point, the corresponding image features and the skinning parameters, and obtain the identification information of each preset sampling point, so as to obtain the identification information of each preset sampling point output by the attention mechanism layer. Among them, the identification information of the preset sampling point can be 1 or 0. When the identification information of the preset sampling point is 1, it indicates that the preset sampling point is located inside the three-dimensional human body model of the target human body instance; when the identification information of the preset sampling point is 0, it indicates that the preset sampling point is not located inside the three-dimensional human body model of the target human body instance.

[0103] The identification information of each preset sampling point is input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer. The three-dimensional human body reconstruction result includes a first model of the target human body instance, and the posture of the first model is a standard posture.

[0104] Specifically, after obtaining the identification information of each preset sampling point, the identification information of each preset sampling point may be input into the result output layer.

[0105] The result output layer can generate a first model of the target human body instance through the MarchingCubic algorithm based on a preset threshold and the identification information of each preset sampling point as the three-dimensional human body reconstruction result of the target human body instance, and then the three-dimensional human body reconstruction result of the target human body instance output by the result output layer can be obtained.

[0106] It should be noted that the posture of the first model of the target human body instance is the above-mentioned standard posture.

[0107] Optionally, the preset threshold may be 0.5.

[0108] The embodiment of the present invention inputs the target image and the position information of each preset sampling point into the image feature extraction layer, obtains the image feature corresponding to each preset sampling point, inputs the position information of each preset sampling point and the corresponding image feature into the skin parameter extraction layer, obtains the skin parameter corresponding to each preset sampling point output by the skin parameter extraction layer, inputs the position information of each preset sampling point, the corresponding image feature and the skin parameter into the attention mechanism layer, obtains the identification information output by the attention mechanism layer for indicating that each preset sampling point is located or not located inside the three-dimensional human body model of the target human body instance, and inputs the identification information of each preset sampling point into the result output The three-dimensional human body reconstruction result of the first model including the target human body instance output by the result output layer can be obtained by the skin parameter extraction layer constructed based on the MLP network, and the three-dimensional human body reconstruction can be driven based on the skin parameters of each preset sampling point. Compared with the traditional three-dimensional human body reconstruction method in which the skin parameters are obtained based on the parameterized human body model, the driving can be smoother and more flexible. The fusion of multi-frame graphic features is realized through the attention mechanism layer constructed based on the Transformer network. Compared with the direct calculation of the average image features, the image features can be adaptively fused, thereby improving the accuracy of three-dimensional human body reconstruction.

[0109] Based on the contents of the above embodiments, the image feature extraction layer includes: a data processing unit, a position updating unit and a feature extraction unit.

[0110] Correspondingly, the target image and the position information of each preset sampling point are input into the image feature extraction layer to obtain the image features corresponding to each preset sampling point, specifically including: inputting the target image into the data processing unit, obtaining the second model of the target human body instance output by the data processing unit, and the second model carries the image posture parameters corresponding to the target image.

[0111] Specifically, the target image is input into the data processing unit in the image feature extraction layer, and the data processing unit can generate a second model of the target human body instance based on the SMPL algorithm, so that the above-mentioned second model output by the data processing unit can be obtained.

[0112] It should be noted that although the posture of the second model generated based on the SMPL algorithm is the same as the actual posture of the target human instance in the target image, the second model is a rough three-dimensional human body model, which is prone to problems such as distortion and missing details. Therefore, the second model cannot be used as the three-dimensional human body reconstruction result of the target human instance.

[0113] It should be noted that when the data processing unit generates the second model based on the SMPL algorithm, the image posture parameters corresponding to the target image can be obtained, and the second model output by the data processing unit can carry the above-mentioned image posture parameters.

[0114] The target image, image posture parameters and position information of each preset sampling point are input into a position update unit, and the updated position information of each preset sampling point output by the position update unit is obtained, wherein the updated position information corresponds to the real posture of the target human instance in the target image.

[0115] Specifically, the target image, the above-mentioned image posture parameters and the position information of each preset sampling point are input into the position update unit in the image feature extraction layer, and the position update unit can input the target image, the above-mentioned image posture parameters and the position information of each preset sampling point into the linear mixed skinning algorithm to calculate and obtain the updated position information of each preset sampling point. The updated position information of each preset sampling point can correspond to the real posture of the target human body instance in the target image.

[0116] The updated position information of the target image and each preset sampling point is input into the feature extraction unit, and the image feature corresponding to each preset sampling point output by the feature extraction unit is obtained.

[0117] Specifically, after obtaining the updated position information of each preset sampling point, the target image and the updated position information of each preset sampling point may be input into a feature extraction unit in the image feature extraction layer.

[0118] The feature extraction unit can perform feature extraction on the target image, and project each preset sampling point onto the image feature through orthogonal projection, so as to obtain the image feature corresponding to each preset sampling point output by the feature extraction unit.

[0119] The embodiment of the present invention inputs the target image into a data processing unit in an image feature extraction layer, obtains a second model of a target human body instance output by the data processing unit and image posture parameters corresponding to the target image carried by the second model, inputs the target image, the image posture parameters and the position information of each preset sampling point into a position updating unit in the image feature extraction layer, obtains updated position information of each preset sampling point output by the position updating unit, the updated position information corresponds to the real posture of the target human body instance in the target position, inputs the target image and the updated position information of each preset sampling point into a feature extraction unit in the image feature extraction layer, obtains image features corresponding to each preset sampling point output by the feature extraction unit, and uses the rough second model as prior knowledge to provide richer three-dimensional semantic information for the preset sampling points, thereby further improving the accuracy of three-dimensional human body reconstruction.

[0120] Based on the contents of the above embodiments, the position information, corresponding image features and skinning parameters of each preset sampling point are input into the attention mechanism layer, and the identification information of each preset sampling point output by the attention mechanism layer is obtained. The above method also includes: inputting the identification information of each preset sampling point and the second model into the result output layer, and obtaining the three-dimensional human body reconstruction result output by the result output layer. The three-dimensional human body reconstruction result includes a third model of the target human body instance, and the posture of the third model is the real posture of the target human body instance in the target image.

[0121] Specifically, after obtaining the identification information of each preset sampling point output by the attention mechanism layer, the identification information of each preset sampling point and the second model of the target human body instance can be input into the result output layer together.

[0122] The result output layer can generate a first model of the target human body instance through the Marching Cubic algorithm based on the identification information of each preset sampling point, and perform posture deformation on the first model based on the second model to obtain a third model of the target human body instance. The posture of the third model is the same as that of the second model, and both are the true postures of the target human body instance in the target image.

[0123] Optionally, the preset threshold may be 0.5.

[0124] The embodiment of the present invention inputs the identification information of each preset sampling point and the second model of the target human body instance into the result output layer, and obtains the three-dimensional human body reconstruction result of the third model including the target human body instance output by the result output layer. On the basis of obtaining the first model with a standard posture, the first model can be deformed in posture based on the rough second model, so that the third model with the same actual posture as the target human body instance in the target image can be obtained more accurately and efficiently.

[0125] Based on the contents of the above embodiments, the loss function of the three-dimensional human body reconstruction model includes: a skin parameter loss function.

[0126] The skin parameter loss function is determined based on the predicted skin parameters and skin parameter labels corresponding to the sample sampling points; the predicted skin parameters are output by the skin parameter extraction model in training by inputting the position information of the sample sampling points and the corresponding image features; the image features corresponding to the sample sampling points are determined based on the sample images.

[0127] Specifically, the skin parameter loss function can be expressed by the following formula:

[0128]

[0129] Among them, Lskin represents the skin parameter loss value of the sample sampling point; P represents the set of sample sampling points; S(p,I) represents the predicted skin parameter corresponding to the sample sampling point; S * (p,I) represents the skin parameter label corresponding to the sample sampling point. The training target is the skin parameter loss value L of the sample sampling point. skin minimize.

[0130] It should be noted that the specific process of obtaining the predicted skin parameters and skin parameter labels corresponding to the sample sampling points can be found in the above embodiments, which will not be described again here.

[0131] The embodiment of the present invention can improve the calculation accuracy of the three-dimensional human body reconstruction model by determining a skin parameter loss function based on the predicted skin parameters and skin parameter labels corresponding to sample sampling points, and training the three-dimensional human body reconstruction model based on the skin parameter loss function.

[0132] Based on the contents of the above embodiments, the loss function of the three-dimensional human body reconstruction model includes: a classification loss function.

[0133] The classification loss function is determined based on the predicted identification information and identification information label of the sample sampling point; the predicted identification information of the sample sampling point is input into the attention mechanism layer in training by inputting the position information of the sample sampling point, the corresponding image features and the predicted skinning parameters, and is output by the attention mechanism layer in training.

[0134] Specifically, the classification loss function can be expressed by the following formula:

[0135]

[0136] Among them, L 3D represents the classification loss value of the sample sampling point; P represents the set of sample sampling points; F(p,I) represents the predicted identification information of the sample sampling point, F * (p,I) represents the identification information label of the sample sampling point. The training target is the classification loss value L of the sample sampling point 3D minimize.

[0137] It should be noted that the specific process of obtaining the predicted identification information and the identification information label of the sampling point can refer to the content of the above embodiment, which will not be repeated here.

[0138] Optionally, a weighted summation of the skin parameter loss function and the classification loss function may be performed to obtain a target loss function for training the 3D human body reconstruction model. The target loss function may be expressed by the following formula:

[0139]

[0140] Among them, L represents the target loss function value; γ 1 and γ 2 They represent weight parameters respectively; θ represents learnable model parameters. The training goal is to minimize the target loss function value L.

[0141] It should be noted that the weight parameter γ 1 and γ 2 It can be predetermined based on prior knowledge.

[0142] When training the three-dimensional human body reconstruction model based on the target loss function, the model parameter θ can be optimized based on the obtained target loss function value. When the target loss function value has not converged, the three-dimensional human body reconstruction model continues to be trained based on the target loss function; when the target loss function value converges, the training of the three-dimensional human body reconstruction model is stopped, the model parameter θ is output, and the trained three-dimensional human body reconstruction model is obtained.

[0143] The embodiment of the present invention can improve the calculation accuracy of the three-dimensional human body reconstruction model by determining the classification loss function based on the predicted identification information and identification information labels of the sample sampling points, and training the three-dimensional human body reconstruction model based on the classification loss function.

[0144] Figure 5 Schematic diagram of the structure of the three-dimensional human body reconstruction device provided by the present invention. Figure 5 The three-dimensional human body reconstruction device provided by the present invention is described. The three-dimensional human body reconstruction device described below and the three-dimensional human body reconstruction method provided by the present invention described above can be referred to each other. Figure 5 As shown, the device includes: an image acquisition module 501 and a three-dimensional reconstruction module 502.

[0145] The image acquisition module 501 is used to acquire an image of a target human instance as a target image.

[0146] The 3D reconstruction module 502 is used to input the target image into the 3D human body reconstruction model, and obtain the 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model.

[0147] The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance.

[0148] The three-dimensional human body reconstruction model is used to reconstruct the three-dimensional human body of the target human body instance based on the image features and skin parameters corresponding to each preset sampling point. The image features are determined based on the target image, and the skin parameters are determined based on the corresponding image features. The preset sampling points are evenly distributed in the three-dimensional space.

[0149] Specifically, the image acquisition module 501 and the three-dimensional reconstruction module 502 are electrically connected.

[0150] The image acquisition module 501 can be used to acquire an image of a target human instance as a target image by relying on a conventional image sensor or an electronic device with an image acquisition function.

[0151] The 3D reconstruction module 502 may be used to input a target image into a trained 3D human body reconstruction model, and obtain a 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model.

[0152] The embodiment of the present invention obtains an image of a target human instance as a target image, and inputs the target image into a 3D human body reconstruction model. The 3D human body reconstruction model obtains image features of each preset sampling point based on the target image, obtains skin parameters corresponding to each preset sampling point based on the image features corresponding to each preset sampling point, and performs 3D human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, thereby obtaining a 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model. The 3D human body reconstruction model is obtained after training based on a 3D model of a sample human body instance and a sample image including the 3D model of the sample human body instance. The preset sampling points are evenly distributed in a 3D space, and the image of the target human body instance can be freely obtained without relying on an image sensor with calibrated camera parameters, an expensive laser scanner or a multi-camera array system, thereby enabling simpler and more efficient 3D human body reconstruction, and enabling a wider range of 3D human body reconstruction applicable to various scenarios.

[0153] Figure 6 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 6As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute a three-dimensional human body reconstruction method, the method comprising: obtaining an image of a target human body instance as a target image; inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; wherein the three-dimensional human body reconstruction model is obtained after training based on a three-dimensional model of a sample human body instance and a sample image, and the sample image includes a three-dimensional model of the sample human body instance; the three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each preset sampling point is evenly distributed in the three-dimensional space.

[0154] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0155] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the three-dimensional human body reconstruction method provided by the above-mentioned methods, which includes: obtaining an image of a target human body instance as a target image; inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; wherein the three-dimensional human body reconstruction model is obtained after training based on a three-dimensional model of a sample human body instance and a sample image, and the sample image includes the three-dimensional model of the sample human body instance; the three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each preset sampling point is evenly distributed in the three-dimensional space.

[0156] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the three-dimensional human body reconstruction method provided by the above-mentioned methods, the method comprising: obtaining an image of a target human body instance as a target image; inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; wherein the three-dimensional human body reconstruction model is obtained after training based on a three-dimensional model of a sample human body instance and a sample image, and the sample image includes the three-dimensional model of the sample human body instance; the three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each preset sampling point is evenly distributed in the three-dimensional space.

[0157] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0158] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A three-dimensional human body reconstruction method, It is characterized in that include: Acquire an image of a target human instance as a target image; Inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance; The three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each of the preset sampling points is evenly distributed in the three-dimensional space; The three-dimensional model of the sample human body instance includes a first sample model, a second sample model and a third sample model of the sample human body instance, the first sample model and the third sample model have the same posture, the posture of the second sample model is a preset standard posture, and the first sample model and the third sample model are acquired in different ways; the sample image includes the first sample model; The three-dimensional human body reconstruction model includes: an image feature extraction layer, a skin parameter extraction layer, an attention mechanism layer and a result output layer; Correspondingly, the step of inputting the target image into a three-dimensional human body reconstruction model and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model specifically includes: Inputting the target image and the position information of each preset sampling point into the image feature extraction layer, and obtaining the image features corresponding to each preset sampling point output by the image feature extraction layer; Inputting the position information and the corresponding image features of each preset sampling point into the skin parameter extraction layer, and obtaining the skin parameters corresponding to each preset sampling point output by the skin parameter extraction layer; Inputting the position information, corresponding image features and skinning parameters of each preset sampling point into the attention mechanism layer, and obtaining identification information of each preset sampling point output by the attention mechanism layer, wherein the identification information is used to indicate whether each preset sampling point is located inside the three-dimensional human body model of the target human body instance or not; The identification information of each preset sampling point is input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer, wherein the three-dimensional human body reconstruction result includes a first model of the target human body instance, and the posture of the first model is the standard posture.

2. The three-dimensional human body reconstruction method according to claim 1, It is characterized in that The image feature extraction layer includes: a data processing unit, a position updating unit and a feature extraction unit; Correspondingly, the step of inputting the target image and the position information of each preset sampling point into the image feature extraction layer to obtain the image feature corresponding to each preset sampling point specifically includes: Inputting the target image into the data processing unit, obtaining a second model of the target human body instance output by the data processing unit, wherein the second model carries image posture parameters corresponding to the target image; Inputting the target image, the image posture parameter and the position information of each preset sampling point into the position updating unit, obtaining the updated position information of each preset sampling point output by the position updating unit, wherein the updated position information corresponds to the real posture of the target human body instance in the target image; The target image and the updated position information of each preset sampling point are input into the feature extraction unit, and the image feature corresponding to each preset sampling point output by the feature extraction unit is obtained.

3. The three-dimensional human body reconstruction method according to claim 2, It is characterized in that After inputting the position information of each preset sampling point, the corresponding image features and the skinning parameters into the attention mechanism layer and obtaining the identification information of each preset sampling point output by the attention mechanism layer, the method further includes: The identification information of each preset sampling point and the second model are input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer, wherein the three-dimensional human body reconstruction result includes a third model of the target human body instance, and the posture of the third model is the real posture of the target human body instance in the target image.

4. The three-dimensional human body reconstruction method according to claim 1, It is characterized in that The loss function of the three-dimensional human body reconstruction model includes: a skin parameter loss function; The skin parameter loss function is determined based on the predicted skin parameters and skin parameter labels corresponding to the sample sampling points; the predicted skin parameters are input into the skin parameter extraction model in training with the position information and the corresponding image features of the sample sampling points, and are output by the skin parameter extraction model in training; the image features corresponding to the sample sampling points are determined based on the sample image.

5. The three-dimensional human body reconstruction method according to claim 4, It is characterized in that The loss function of the three-dimensional human body reconstruction model includes: a classification loss function; The classification loss function is determined based on the predicted identification information and identification information label of the sample sampling point; the predicted identification information of the sample sampling point is input into the attention mechanism layer in training by the position information of the sample sampling point, the corresponding image features and the predicted skinning parameters, and output by the attention mechanism layer in training.

6. A three-dimensional human body reconstruction device, It is characterized in that include: An image acquisition module, used for acquiring an image of a target human instance as a target image; A three-dimensional reconstruction module, used for inputting the target image into a three-dimensional human body reconstruction model, and obtaining a three-dimensional human body reconstruction result of the target human body instance output by the three-dimensional human body reconstruction model; The three-dimensional human body reconstruction model is obtained after training based on the three-dimensional model of the sample human body instance and the sample image, and the sample image includes the three-dimensional model of the sample human body instance; The three-dimensional human body reconstruction model is used to perform three-dimensional human body reconstruction on the target human body instance based on the image features and skin parameters corresponding to each preset sampling point, the image features are determined based on the target image, the skin parameters are determined based on the corresponding image features, and each of the preset sampling points is evenly distributed in the three-dimensional space; The three-dimensional model of the sample human body instance includes a first sample model, a second sample model and a third sample model of the sample human body instance, the first sample model and the third sample model have the same posture, the posture of the second sample model is a preset standard posture, and the first sample model and the third sample model are acquired in different ways; the sample image includes the first sample model; The three-dimensional human body reconstruction model includes: an image feature extraction layer, a skin parameter extraction layer, an attention mechanism layer and a result output layer; Correspondingly, the 3D reconstruction module inputs the target image into a 3D human body reconstruction model, and obtains a 3D human body reconstruction result of the target human body instance output by the 3D human body reconstruction model, specifically including: Inputting the target image and the position information of each preset sampling point into the image feature extraction layer, and obtaining the image features corresponding to each preset sampling point output by the image feature extraction layer; Inputting the position information and the corresponding image features of each preset sampling point into the skin parameter extraction layer, and obtaining the skin parameters corresponding to each preset sampling point output by the skin parameter extraction layer; Inputting the position information, corresponding image features and skinning parameters of each preset sampling point into the attention mechanism layer, and obtaining identification information of each preset sampling point output by the attention mechanism layer, wherein the identification information is used to indicate whether each preset sampling point is located inside the three-dimensional human body model of the target human body instance or not; The identification information of each preset sampling point is input into the result output layer to obtain the three-dimensional human body reconstruction result output by the result output layer, wherein the three-dimensional human body reconstruction result includes a first model of the target human body instance, and the posture of the first model is the standard posture.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the program, the steps of the three-dimensional human body reconstruction method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the steps of the three-dimensional human body reconstruction method according to any one of claims 1 to 5 are implemented.

9. A computer program product comprising a computer program, It is characterized in that When the computer program is executed by a processor, the steps of the three-dimensional human body reconstruction method according to any one of claims 1 to 5 are implemented.

Images