Three-dimensional dressing human body reconstruction method and device based on side view normal graph regularization and medium

Abstract

The invention discloses a three-dimensional dressing human body reconstruction method and device based on side view normal graph regularization and a medium, and the method comprises the steps: obtaining an input image, and obtaining a parameterized human body model aligned with the input image; obtaining voxel features, normal features and symbol distance features of the parameterized human body model; calculating the contrastive loss of the side view normal diagram of the reconstructed human body grid and the real normal diagram to update the parameters of the discriminator; an additional point is sampled at the sampling point along the coordinate axis, and first-order and second-order Eikonal losses of the additional point are calculated; the mean square error loss of the predicted value and the true value of the sampling point is calculated, and the total loss is calculated according to the antagonism loss, the first-order and second-order Eikonal loss and the mean square error loss and used for updating the parameters of the multi-layer perceptron; and the trained multi-layer perceptron is used for realizing three-dimensional dressing human body reconstruction. The method can ensure that the model reconstructs the side-looking geometry of the grid from global and local scales in a regularization manner, and effectively improves the accuracy of reconstructing the human body grid.

Description

Technical Field

[0001] The present invention relates to a three-dimensional clothed human body reconstruction technology, and in particular to a three-dimensional clothed human body reconstruction method, device and medium based on side view normal map regularization. Background Art

[0002] Creating realistic digital humans with complex clothing details plays a key role in mixed reality, autonomous driving simulation, and film production. Traditional methods usually require the use of expensive dedicated equipment to take multi-view images to customize high-fidelity three-dimensional digital images. Such methods consume a lot of human resources, are cost-ineffective, and are not easy to expand.

[0003] Recent methods extract 3D scan data from RGB images of clothed bodies, eliminating the need for expensive scanning equipment and making it easier for individuals to create personalized 3D images. A common approach for these methods is to sample points in 3D space and provide 2D / 3D features associated with these points to implicit functions to infer geometry. While recent implicit function-based methods are effective in handling clothed people in various scenarios, they do not fully address the issues of global visual plausibility and local surface continuity in side views. Due to the insufficient utilization of side view information of clothed bodies, these methods suffer from global or local artifacts such as incorrect depth or seams when unconstrained images are used as input. Summary of the invention

[0004] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the object of the present invention is to provide a method, device and medium for reconstructing a three-dimensional clothed human body based on side view normal map regularization.

[0005] The first technical solution adopted by the present invention is:

[0006] A method for reconstructing a three-dimensional clothed human body based on side view normal map regularization comprises the following steps:

[0007] Obtain an input image and obtain a parameterized human body model aligned with the input image;

[0008] Obtain voxel features, normal features, and signed distance features of the parametric human body model, concatenate the obtained features, and input them into a multi-layer perceptron to realize the reconstruction of the human body mesh;

[0009] Compute the adversarial loss between the side-view normal map of the reconstructed human mesh and the true normal map to update the parameters of the discriminator;

[0010] Sample additional points along the coordinate axis at the sampling point and calculate the first-order and second-order Eikonal losses of the additional points;

[0011] Calculate the mean square error loss between the predicted value and the true value of the sampling point, and calculate the total loss based on the adversarial loss, first-order and second-order Eikonal loss, and mean square error loss to update the parameters of the multilayer perceptron;

[0012] The trained multi-layer perceptron is used to realize three-dimensional reconstruction of the clothed human body.

[0013] Further, the obtaining of a parameterized human body model aligned with the input image comprises:

[0014] The SMPL model parameters aligned with the input image are obtained through the parameter regressor, and the corresponding parameterized human body model is obtained according to the SMPL model parameters.

[0015] Furthermore, the step of obtaining voxel features, normal features, and signed distance features of the parameterized human body model includes:

[0016] Voxelize the parameterized human body model to obtain the corresponding voxel model, and then use the voxel encoder to obtain the voxel features of the parameterized human body model.

[0017] Use a differentiable renderer to obtain the normal map of the parametric human body model, obtain the normal map of the clothed body through a fine normal generation network, and concatenate the normal map of the parametric human body model and the normal map of the clothed body to obtain the normal feature F. n (p);

[0018] A signed distance field is established for the parameterized human body model, and the signed distance value of each point of the parameterized human body model is obtained as the signed distance feature F s (p).

[0019] Furthermore, the calculation of the adversarial loss between the side view normal map of the reconstructed human body mesh and the real normal map to update the parameters of the discriminator includes:

[0020] Take the camera viewpoint of the input image as the front view;

[0021] Render and reconstruct the clothed human body from two preset side view angles to obtain the side view normal map;

[0022] Use the pre-trained normal generation network to obtain the normal map of the clothed body from front and back perspectives;

[0023] The side view normal map and the clothed body normal map are input into the discriminator, and the adversarial loss is calculated to update the parameters of the discriminator; wherein the discriminator is used to distinguish between real and generated normal maps.

[0024] Furthermore, the calculation formula of the adversarial loss is as follows:

[0025]

[0026] Where, L mse is the mean square error function; is the side view normal map, N c is the normal map of the clothed body; D represents the discriminator.

[0027] Furthermore, the step of sampling additional points along the coordinate axis at the sampling point and calculating the first-order and second-order Eikonal losses of the additional points includes:

[0028] Sample two additional points along each axis of the standard coordinates of the sampled points with a preset step size;

[0029] The predicted values ​​of the additional points are obtained through a multi-layer perceptron, and the first-order and second-order Eikonal losses of the additional points are calculated based on the predicted values ​​to promote the local continuity of the surface of the sampling points.

[0030] Furthermore, when along the z-axis, with a step size ∈ z When sampling additional points, the first-order and second-order Eikonal losses are calculated as follows:

[0031]

[0032] Where P i is the sampling point, φ is the multi-layer perceptron, represents differential; represents the first-order Eikonal loss, represents the second-order Eikonal loss.

[0033] Furthermore, the calculation of the mean square error loss between the predicted value and the true value of the sampling point, the calculation of the total loss based on the adversarial loss, the first-order and second-order Eikonal loss and the mean square error loss, and the updating of the parameters of the multilayer perceptron include:

[0034] Calculate the mean square error loss L between the predicted value and the true value of the sampling point a :

[0035]

[0036] In the formula, are the true and predicted signed distance values ​​respectively; L mse is the mean square error function;

[0037] Calculate the total training loss L:

[0038] L=L a +ω D L D +ω eik L eik +ω curv L curv

[0039] In the formula, ω D ,ω eik ,ω curv is the training hyperparameter; L D is the adversarial loss, L eik is the first-order Eikonal loss, L curv is the second-order Eikonal loss;

[0040] The total loss L is used to update the multilayer perceptron network parameters until the model converges.

[0041] The second technical solution adopted by the present invention is:

[0042] An electronic device comprises a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, the code set or the instruction set is loaded and executed by the processor to implement a three-dimensional clothed human body reconstruction method based on side view normal map regularization as described above.

[0043] The third technical solution adopted by the present invention is:

[0044] A computer-readable storage medium stores at least one instruction, at least one program, a code set or an instruction set, wherein the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement a three-dimensional clothed human body reconstruction method based on side view normal map regularization as described above.

[0045] The fourth technical solution adopted by the present invention is:

[0046] A computer program product or a computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above method.

[0047] The beneficial effects of the present invention are as follows: the present invention solves the problem that the existing three-dimensional reconstruction algorithm of a clothed human body cannot fully standardize the side structure of the human body and the side reconstruction results are poor. It can ensure that the model regularizes the side view geometry of the reconstructed mesh from a global and local scale, and effectively improves the accuracy of the reconstructed human body mesh. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 is a schematic diagram of a clothed 3D human body reconstruction algorithm guided by high-frequency local features in an embodiment of the present invention;

[0050] Figure 2 It is a flowchart of the steps of a method for reconstructing a three-dimensional clothed human body based on side view normal map regularization in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0052] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0053] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0054] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific content of the technical solution.

[0055] Terminology explanation:

[0056] SMPL model: Skinned Multi-Person Linear Model, is a statistically based 3D human model introduced by Loper et al. in 2015. It models human shape and posture through a linear mixture method, providing a simple but effective way to represent the human body.

[0057] In response to the existing technical problems, the present invention proposes a three-dimensional clothed human body reconstruction scheme that can well regularize the reconstruction of the human body's side view structure under unconstrained image input conditions. It is worth noting that the scheme does not require real side view texture information as training data.

[0058] Example 1

[0059] like Figure 2 As shown, this embodiment provides a method for reconstructing a 3D clothed human body based on side view normal map regularization, which can reconstruct the posture and clothing details of the human body in the input image under the input condition of a single image, and regularize the side view geometry of the reconstructed mesh at the global and local scales. The method includes the following steps:

[0060] S1. Obtain an input image and obtain a parameterized human body model aligned with the input image.

[0061] Specifically, the SMPL model parameters aligned with the input image are obtained using a parameter-based regressor to obtain the corresponding parameterized human body model.

[0062] S2. Obtain voxel features, normal features, and signed distance features of the parameterized human body model, concatenate the obtained features, and input them into a multi-layer perceptron to realize the reconstruction of the human body mesh.

[0063] In some embodiments, the parameterized human body model is voxelized to obtain a corresponding voxel model, and then a voxel encoder is used to obtain voxel features of the parameterized human body model; a differentiable renderer is used to obtain a normal map of the parameterized human body model, a clothed human body normal map is obtained through a fine normal generation network, and the parameterized human body model normal map and the clothed human body normal map are spliced ​​to obtain normal features; a signed distance field is established to obtain signed distance features of each point of the parameterized human body model.

[0064] S3. Calculate the adversarial loss between the side view normal map of the reconstructed human mesh and the true normal map to update the parameters of the discriminator.

[0065] Exemplarily, normal maps of a reconstructed human body are rendered from two side view angles (such as 90° and 270°), the side view normal map and the clothed human body normal map are input into a side view normal map discriminator, the discriminator is used to distinguish between real and generated normal maps, the adversarial loss is calculated, and the discriminator parameters are updated.

[0066] S4. Sample additional points along the coordinate axis at the sampling point and calculate the first-order and second-order Eikonal losses of the additional points.

[0067] Specifically, two additional points are sampled at a certain distance along each axis of the standard coordinates of the sampling points, the predicted values ​​of the additional points are obtained through the multi-layer perceptron, and the first-order and second-order Eikonal losses of the additional points are calculated.

[0068] S5. Calculate the mean square error loss between the predicted value and the true value of the sampling point, and calculate the total loss based on the adversarial loss, first-order and second-order Eikonal loss and mean square error loss, which is used to update the parameters of the multilayer perceptron.

[0069] S6. Use the trained multi-layer perceptron to achieve three-dimensional reconstruction of the clothed human body.

[0070] The following is combined with Figure 1 The above method is supplemented with explanation.

[0071] like Figure 1 As shown in the figure, this embodiment designs a 3D clothed human reconstruction algorithm based on side view normal map regularization. First, the SMPL model parameter regressor is used to obtain a parameterized human body model aligned with the input image; then, voxel features, normal features, and signed distance features are extracted based on the parameterized human body model and spliced; finally, the spliced ​​features are used to input the implicit function to predict the 3D human body occupancy field. It consists of the following modules:

[0072] (a) Feature extraction

[0073] First, a parametric regressor is used to obtain a parametric human body model aligned with the input image. Then, the feature extraction module extracts voxel features, normal features, and signed distance features from the parametric human body model.

[0074] Specifically, using the public parameter regressor, from the input human image I in Regress the human body parameters and obtain the same in Aligned SMPL model M body ; Secondly, use the encoder to convert M body Perform voxelization and encoding to obtain voxel features Then, M is rendered using a differentiable renderer. body The normal map of the front and back surfaces is used to obtain the normal feature F n (p); Finally, for M body Establish a signed distance field and obtain the signed distance value F of each point s (p).

[0075] (b) Global side view normal map identification

[0076] The normal map of the human body is reconstructed by rendering from two side view angles of 90° and 270°. The side view normal map discriminator is then used to distinguish the side view normal map from the clothed human body normal map, and the adversarial loss is calculated to promote the multi-layer perceptron to generate a more accurate side view structure of the human body.

[0077] Specifically, first, the input image I in The camera viewpoint is used as the front view, that is, the angle is 0°; then, the reconstructed clothed human body is rendered from two side view angles (i.e., 90° and 270°) to obtain the side view normal map Then use the pre-trained normal generation network to obtain the normal map N of the clothed human body in the front and back perspectives c ; Finally, a block-based discriminator network is introduced to regularize the normal map at the scale of the block and calculate the adversarial loss:

[0078]

[0079] Among them, L mse is the mean square error function; is the side view normal map, N c is the normal map of the clothed body; D represents the discriminator.

[0080] (c) Local many-to-one gradient computation

[0081] Two additional points are sampled at a certain distance along each axis of the standard coordinates of the sampling points, and the predicted values ​​of the additional points are obtained by inputting into the multilayer perceptron. The first-order and second-order Eikonal losses of the additional points are calculated to promote the local continuity of the surface of the sampling points.

[0082] Along the sampling point P i Each axis of the standard coordinate (x / y / z axis) samples two additional points with a step size of ∈, inputs the multilayer perceptron φ to obtain the predicted value of the additional point, and calculates the first-order Eikonal loss of the additional point and the second-order Eikonal loss To promote the local continuity of the surface of the sampling point. Taking the z-axis as an example, the first-order and second-order Eikonal loss calculation formulas are as follows:

[0083]

[0084] As an optional implementation, in order to prevent over-smoothing in the many-to-one optimization process, this embodiment introduces coarse-to-fine sampling. Specifically, as the training epoch n progresses, we exponentially reduce the size of the step size ∈:

[0085] ∈ n =∈0·(1 / 2) n

[0086] Among them, ∈0 is the preset initial value.

[0087] (d) Three-dimensional occupancy field prediction

[0088] Calculate the mean square error loss between the predicted value and the true value of the sampling point, calculate the total training loss, and update the multi-layer perceptron network parameters.

[0089] Specifically, calculate the mean square error loss L between the predicted value and the true value of the sampling point a :

[0090]

[0091] in, are the true and predicted signed distance values, respectively.

[0092] Calculate the total training loss L:

[0093] L=L a +ω D L D +ω eik L eik +ω curv L curv

[0094] Among them, ω D ,ω eik ,ω curv is the training hyperparameter.

[0095] The total loss L is used to update the multilayer perceptron network parameters until the model converges.

[0096] The trained multi-layer perceptron can predict the signed distance value of the sampling point and obtain the reconstructed clothed human body mesh M through the Marching Cube algorithm. cloth , for M cloth Hand replacement and texture mapping can be performed separately to obtain a more detailed hand structure. rep and M with texture color tex .

[0097] Among them, module (a) provides input features corresponding to the sampling points for modules (b) and (c), which can be used for model training; module (b) reconstructs the side view structure of the human body from a global regularization perspective, which promotes the visual rationality of the reconstruction results of module (d); module (c) locally smoothes the significant differences near the sampling points, which promotes the surface continuity of the reconstruction results of module (d).

[0098] In summary, compared with the prior art, the present invention has at least the following advantages and beneficial effects:

[0099] (1) The present invention introduces a side-view normal map discriminator to distinguish the rendered side-view normal map from the clothing normal map, regularizes the global side-view geometry of the reconstruction result, and improves the visual rationality of the reconstruction result.

[0100] (2) The present invention introduces a many-to-one gradient calculation strategy to integrate the gradients of points near the sampling point, smoothing the local irregularities to obtain a more consistent surface reconstruction and reducing local artifacts in the reconstruction results, such as seams and jagged edges.

[0101] Example 2

[0102] An embodiment of the present invention further provides an electronic device, the electronic device comprising a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the following Figure 2 A three-dimensional clothed human body reconstruction method based on side view normal map regularization is shown.

[0103] It is understood that the memory may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory may be used to store instructions, programs, codes, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data created according to the use of the server, etc.

[0104] The processor may include one or more processing cores. The processor uses various interfaces and lines to connect the various parts of the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory. Optionally, the processor can be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor can integrate one or a combination of a central processing unit (CPU) and a modem. Among them, the CPU mainly processes the operating system and application programs; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor, but implemented separately through a chip.

[0105] Since the electronic device is an electronic device corresponding to a method for reconstructing a three-dimensional clothed human body based on side-view normal map regularization in an embodiment of the present invention, and the principle of solving the problem by the electronic device is similar to that of the method, the implementation of the electronic device can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0106] Example 3

[0107] The embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the following Figure 2 A three-dimensional clothed human body reconstruction method based on side view normal map regularization is shown.

[0108] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0109] Since the storage medium is a storage medium corresponding to a method for reconstructing a three-dimensional clothed human body based on side view normal map regularization in an embodiment of the present invention, and the principle of solving the problem by the storage medium is similar to that of the method, the implementation of the storage medium can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0110] Example 4

[0111] In some possible implementations, various aspects of the method of the embodiment of the present invention may also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of a three-dimensional clothed human body reconstruction method based on side view normal map regularization according to various exemplary embodiments of the present application described above in this specification. Among them, the executable computer program code or "code" for executing each embodiment can be written in a high-level programming language such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, structured query language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0112] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0113] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0114] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for reconstructing a three-dimensional clothed human body based on side view normal map regularization, characterized in that: The following steps are involved: Obtain an input image and obtain a parameterized human body model aligned with the input image; Obtain voxel features, normal features, and signed distance features of the parametric human body model, concatenate the obtained features, and input them into a multi-layer perceptron to realize the reconstruction of the human body mesh; Compute the adversarial loss between the side-view normal map of the reconstructed human mesh and the true normal map to update the parameters of the discriminator; Sample additional points along the coordinate axis at the sampling point and calculate the first-order and second-order Eikonal losses of the additional points; Calculate the mean square error loss between the predicted value and the true value of the sampling point, and calculate the total loss based on the adversarial loss, first-order and second-order Eikonal loss, and mean square error loss to update the parameters of the multilayer perceptron; The trained multi-layer perceptron is used to realize three-dimensional reconstruction of the clothed human body.

2. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 1, characterized in that: The step of obtaining a parameterized human body model aligned with the input image comprises: The SMPL model parameters aligned with the input image are obtained through the parameter regressor, and the corresponding parameterized human body model is obtained according to the SMPL model parameters.

3. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 1, characterized in that: The step of obtaining voxel features, normal features, and signed distance features of the parameterized human body model includes: Voxelize the parameterized human body model to obtain the corresponding voxel model, and then use the voxel encoder to obtain the voxel features of the parameterized human body model. Use a differentiable renderer to obtain the normal map of the parametric human body model, obtain the normal map of the clothed body through a fine normal generation network, and concatenate the normal map of the parametric human body model and the normal map of the clothed body to obtain the normal feature F. n (p); A signed distance field is established for the parameterized human body model, and the signed distance value of each point of the parameterized human body model is obtained as the signed distance feature F s (p).

4. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 1, characterized in that: The adversarial loss between the side view normal map of the reconstructed human body mesh and the true normal map is calculated to update the parameters of the discriminator, including: Take the camera viewpoint of the input image as the front view; Render and reconstruct the clothed human body from two preset side view angles to obtain the side view normal map; Use the pre-trained normal generation network to obtain the normal map of the clothed body from front and back perspectives; Input the side view normal map and the clothed body normal map into the discriminator and calculate the adversarial loss to update the parameters of the discriminator; The discriminator is used to distinguish between real and generated normal maps.

5. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 4, characterized in that: The calculation formula of the adversarial loss is as follows: Where, L mse is the mean square error function; is the side view normal map, N c is the normal map of the clothed body; D represents the discriminator.

6. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 1, characterized in that: The step of sampling additional points along the coordinate axis at the sampling point and calculating the first-order and second-order Eikonal losses of the additional points includes: Sample two additional points along each axis of the standard coordinates of the sampled points with a preset step size; The predicted values ​​of the additional points are obtained through a multi-layer perceptron, and the first-order and second-order Eikonal losses of the additional points are calculated based on the predicted values ​​to promote the local continuity of the surface of the sampling points.

7. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 6, characterized in that: When along the z-axis, with a step size ∈ z When sampling additional points, the first-order and second-order Eikonal losses are calculated as follows: Where P i is the sampling point, φ is the multi-layer perceptron, Indicates differential; L eik is the first-order Eikonal loss, L curv is the second-order Eikonal loss.

8. The method for reconstructing a three-dimensional clothed human body based on side view normal map regularization according to claim 1, characterized in that: The calculation of the mean square error loss between the predicted value and the true value of the sampling point, the calculation of the total loss based on the adversarial loss, the first-order and second-order Eikonal loss and the mean square error loss, and the updating of the parameters of the multilayer perceptron include: Calculate the mean square error loss L between the predicted value and the true value of the sampling point a : In the formula, S, are the true and predicted signed distance values ​​respectively; L mse is the mean square error function; Calculate the total training loss L: L=L a +oh D L D +oh eik L eik +oh curv L curv In the formula, ω D ,ω eik ,ω curv is the training hyperparameter; L D is the adversarial loss, L eik is the first-order Eikonal loss, L curv is the second-order Eikonal loss; The total loss L is used to update the multilayer perceptron network parameters until the model converges.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method according to any one of claims 1 to 8.

Images