A geometrically consistent and boundary-aware monocular video method for clothing human reconstruction
By using geometrically consistent deformation guided by normals and pose-driven mapping, boundary uncertainty modeling, and reliability-aware error propagation decoupled rendering, the geometric drift and boundary artifact problems in monocular video clothing human reconstruction are solved, resulting in more stable reconstruction results.
Patent Information
- Application Number
- CN202611314606.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-27
- Publication Date
- 2026-11-17
AI Technical Summary
Existing monocular video methods for reconstructing dressed human bodies are prone to geometric drift, boundary artifacts, and color overflow due to geometric ambiguity, boundary uncertainty, and noise propagation, resulting in unstable reconstruction results.
Geometrically consistent deformation is achieved through normal-guided constraints and pose-driven mapping, combined with boundary uncertainty modeling and boundary constraint optimization. Finally, reliability-aware and error propagation decoupled rendering is performed to output robust clothing human reconstruction results.
It improves the geometric stability, boundary consistency, and robustness of monocular video-based human body reconstruction, suppresses geometric drift and boundary error propagation, and enhances the stability and accuracy of the reconstruction results.
Smart Images

Figure CN123118905A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a geometrically consistent and boundary-aware monocular video method for reconstructing dressed human bodies, belonging to the fields of computer-aided design and graphics, and computer vision. Background Technology
[0002] Monocular video-driven clothing human reconstruction provides temporal observation information while maintaining low acquisition costs, better meeting practical application needs and possessing broad application value in fields such as virtual try-on, digital entertainment, film and television production, and medical rehabilitation. Known methods mainly include clothing human reconstruction methods based on explicit geometric representation, implicit functions, neural radiation fields, and 3D Gaussian representation. For clothing humans, geometric and appearance changes are mainly concentrated on the human body surface and clothing areas, thus requiring a representation form that combines explicit structural expression and efficient differentiable rendering capabilities. Compared to topology-constrained mesh representations and rendering-inefficient implicit representations, 3D Gaussian-based representation methods, combining Gaussian sets and Gaussian splashing, achieve efficient explicit scene expression and differentiable rendering, achieving a good balance between structural expression and rendering efficiency. For example, Hu (…<Conference on Computer Vision and Pattern Recognition > Liu (<International Conference on Computational Visual Media>, 2024) represents the human body as an animable 3D Gaussian set and combines pose-driven deformation and dynamic appearance modeling to construct a realistic 3D human body model from a single video. Addressing the issues of pose error and insufficient Gaussian surface alignment in monocular video 3D Gaussian human body reconstruction, Liu improved the quality of human body detail reconstruction through pose refinement and surface-guided reinitialization. Zhai (<Conference on ComputerVision and Pattern Recognition>, 2025) uses self-supervised learning to construct an animable Gaussian joint model without fixed templates, improving the flexibility of human joint deformation modeling in monocular videos.
[0003] Since monocular video-based human body reconstruction relies solely on a single RGB camera input, the depth relationships of the human body, self-occlusion areas, and non-rigid deformations of clothing are difficult to fully observe. Existing methods only use skeleton-driven or linear blending skinning for human body deformation, which easily leads to problems such as deformation direction drift and cross-frame structural inconsistencies. At the same time, existing methods usually rely on foreground masks to provide human body region supervision, without considering the uncertainty of foreground-background blending, which easily causes boundary offsets and local artifacts. In addition, existing methods use a unified error propagation mechanism in the differentiable rendering process, which fails to effectively distinguish the reliability of supervision information between boundary regions and the main human body region, easily causing color overflow and error accumulation, further affecting the stability of the reconstruction results. To this end, this invention first performs geometrically consistent deformation on the Gaussian residual displacements corresponding to the monocular video frame sequence and human pose parameters through normal-guided constraints and pose-driven mapping, obtaining a stable Gaussian set in the observation space; secondly, it performs boundary-aware optimization on the foreground mask and the Gaussian set in the observation space through boundary uncertainty modeling and boundary constraint optimization, obtaining the boundary uncertainty intensity and the boundary optimized Gaussian set, respectively; finally, it combines reliability perception and error propagation to perform decoupled Gaussian rendering on the boundary uncertainty intensity and the boundary optimized Gaussian set, outputting robust clothing human reconstruction results, thereby improving the robustness of monocular video clothing human reconstruction under geometrically underconstrained and boundary noise conditions. Summary of the Invention
[0004] This invention provides a geometrically consistent and boundary-aware monocular video method for clothing human reconstruction, which solves the problems of geometric drift, boundary artifacts and color overflow caused by geometric ambiguity, boundary uncertainty and noise propagation in existing monocular video clothing human reconstruction, and improves the geometric stability, boundary consistency and robust reconstruction capability of the reconstruction results.
[0005] The technical solution of this invention is: a geometrically consistent and boundary-aware monocular video method for reconstructing a dressed human body, comprising the following steps:
[0006] Step 1: Apply normal-guided constraints and pose-driven mapping to the monocular video frame sequences in the dressed human dataset. Gaussian residual displacement of human posture parameters By performing geometrically consistent deformation, we obtain the Gaussian set of the observation space. ;
[0007] Step 2: Based on the foreground mask and the Gaussian set of the observation space Boundary-aware optimization is performed through boundary uncertainty modeling and boundary constraint optimization, respectively, to obtain the boundary uncertainty intensity. and boundary optimization Gaussian sets ;
[0008] Step 3: Based on boundary uncertainty intensity and boundary optimization Gaussian sets By combining reliability perception and error propagation, decoupled Gaussian rendering is performed to output robust clothing human reconstruction results. .
[0009] The specific process of Step 1 is as follows:
[0010] First, for human body surface sampling points in the standardized space According to the Human pose parameters in frames Constructing human posture conditional features Through the shared attitude condition mapping function The attitude-related Gaussian residual displacement is obtained. ;
[0011] Secondly, according to Corresponding local normal direction break down The normal residual components were obtained respectively. and tangential residual components Its decomposition form is: , ;
[0012] Then, to Perform residual update to obtain the Gaussian center point in the normalized space. , with tangential residual components Constructing geometrically consistent deformation loss ,constraint The direction of updates;
[0013] Finally, according to the first The frame human pose parameter calculation standard space to observation space skeleton transformation relationship, combined with The corresponding skin weights will Mapping to the observation space yields the Gaussian set of the observation space. .
[0014] The specific process of Step 2 is as follows:
[0015] Step 2.1 Boundary Uncertainty Modeling:
[0016] First, obtain the foreground mask corresponding to the current video frame. Constructing the neighborhood boundary zone of the human body contour Its form is: ,in, and These represent expansion and corrosion operations, respectively. Indicates the first Foreground mask of the frame, Represents the boundary zone of the human body contour neighborhood. Indicates the width of the boundary band;
[0017] Then, according to The distances from pixels to the foreground and background regions are calculated based on screen-space Euclidean distance transform, and the boundary uncertainty intensity is obtained through normalized fusion and inversion mapping. .
[0018] Step 2.2 Boundary Constraint Optimization:
[0019] First, according to Generate boundary space weights Weighted supervision is applied to the pixel reconstruction error within the uncertain boundary region to obtain the boundary-aware weighted residual: ,in, Represents pixels The original reconstruction error at the location, This represents the reconstruction error after boundary space weight modulation;
[0020] Then, in screen space statistics Internally Gaussian set formed by projection scale Local scale regularization constraints are applied to Gaussians with excessively large scale responses within the boundary zone. ,in, It is a monotonically increasing function;
[0021] Finally, combined with boundary-aware weighted residuals With local scale regularization right Joint optimization is performed to obtain the boundary optimization Gaussian set. .
[0022] The specific process of Step 3 is as follows:
[0023] First, optimize Gaussian sets based on boundaries. Perform differentiable Gaussian rendering to obtain the first... Frame pixels The original Gaussian rendering color at the location ;
[0024] Secondly, based on the boundary uncertainty strength Construct pixel reliability weights , used to represent pixels The degree of credibility during the rendering and optimization process, among which Indicates the pixel position in screen space. ;
[0025] Then, combine , Construct reliability-aware synthetic results The calculation form is: ,in, Indicates the reference background color;
[0026] Finally, during the backpropagation process, As a boundary error propagation factor, for based on The calculated reconstruction residuals are gradient modulated to update the 3D Gaussian geometric parameters. and appearance parameters Its form is: , Output robust dressed human reconstruction results .
[0027] The present invention also provides a geometrically consistent and boundary-aware monocular video-based human body reconstruction system, the system comprising: a module for performing the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0028] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0029] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0030] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0031] The beneficial effects of this invention are:
[0032] 1. Existing methods use linear hybrid skinning to drive human body deformation, relying solely on skeletal transformation to describe the human movement process. This makes it difficult to accurately represent the complex non-rigid deformation of clothing areas, thus easily leading to geometric drift and temporal jitter, affecting the stability of the reconstruction results. This invention constructs a geometrically consistent deformation module, establishing a continuous mapping relationship between human posture changes and Gaussian residual displacements. Furthermore, it utilizes normal and tangential residual decomposition to constrain the update direction of the Gaussian center point, thereby improving the geometric stability under monocular under-constraint conditions.
[0033] 2. Most known methods employ unified reconstruction supervision to optimize the human body region. However, they lack effective modeling mechanisms for the foreground-background color mixing problem in the human body contour region, easily leading to phenomena such as boundary blurring, contour loss, and artifacts. This invention designs a boundary-aware optimization module. By constructing a boundary uncertainty distribution map, it models the uncertainty of the human body boundary region and combines boundary-aware weights and local scale regularization constraints to perform differentiated optimization of the Gaussian of the boundary region, thereby improving the reliability of contour neighborhood supervision and the stability of the boundary structure.
[0034] 3. Existing methods employ a unified error propagation mechanism during differentiable rendering. Unreliable errors in boundary regions can easily propagate to the interior of the human body, affecting geometric and appearance parameter updates and leading to problems such as color mixing errors and color overflow. This invention constructs pixel reliability weights and combines reliability awareness and error propagation modulation mechanisms for decoupled Gaussian rendering. This allows pixels in different regions to have different contribution weights during parameter optimization, thereby suppressing the propagation of boundary errors to the main human body region and improving the boundary consistency and overall robustness of the reconstruction results. Attached Figure Description
[0035] Figure 1 This is a flowchart of the present invention;
[0036] Figure 2 This is a flowchart illustrating the geometrically consistent deformation process of the present invention.
[0037] Figure 3 This is a flowchart illustrating the boundary uncertainty modeling process of the present invention.
[0038] Figure 4 This is a flowchart of the boundary constraint optimization process of the present invention;
[0039] Figure 5 This is a flowchart of the decoupled Gaussian rendering process of the present invention;
[0040] Figure 6 This is an example diagram of the clothing-reconstructed human body reconstruction result of the present invention. Detailed Implementation
[0041] Example 1: As Figures 1-6 As shown, a geometrically consistent and boundary-aware monocular video method for clothing human reconstruction includes:
[0042] Step 1: Apply normal-guided constraints and pose-driven mapping to the monocular video frame sequences in the dressed human dataset. Gaussian residual displacement of human posture parameters By performing geometrically consistent deformation, we obtain the Gaussian set of the observation space. ;
[0043] Step 2: Based on the foreground mask and the Gaussian set of the observation space Boundary-aware optimization is performed through boundary uncertainty modeling and boundary constraint optimization, respectively, to obtain the boundary uncertainty intensity. and boundary optimization Gaussian sets ;
[0044] Step 3: Based on boundary uncertainty intensity and boundary optimization Gaussian sets By combining reliability perception and error propagation, decoupled Gaussian rendering is performed to output robust clothing human reconstruction results. .
[0045] The specific process of Step 1 is as follows:
[0046] First, for human body surface sampling points in the standardized space According to the Human pose parameters in frames Constructing human posture conditional features Through the shared attitude condition mapping function The attitude-related Gaussian residual displacement is obtained. ;
[0047] Secondly, according to Corresponding local normal direction break down The normal residual components were obtained respectively. and tangential residual components Its decomposition form is: , ;
[0048] Then, to Perform residual update to obtain the Gaussian center point in the normalized space. , with tangential residual components Constructing geometrically consistent deformation loss ,constraint The direction of updates;
[0049] Finally, according to the first The frame human pose parameter calculation standard space to observation space skeleton transformation relationship, combined with The corresponding skin weights will Mapping to the observation space yields the Gaussian set of the observation space. .
[0050] The specific process of Step 2 is as follows:
[0051] Step 2.1 Boundary Uncertainty Modeling:
[0052] First, obtain the foreground mask corresponding to the current video frame. Constructing the neighborhood boundary zone of the human body contour Its form is: ,in, and These represent expansion and corrosion operations, respectively. Indicates the first Foreground mask of the frame, Represents the boundary zone of the human body contour neighborhood. Indicates the width of the boundary band;
[0053] Then, according to The distances from pixels to the foreground and background regions are calculated based on screen-space Euclidean distance transform, and the boundary uncertainty intensity is obtained through normalized fusion and inversion mapping. .
[0054] Step 2.2 Boundary Constraint Optimization:
[0055] First, according to Generate boundary space weights Weighted supervision is applied to the pixel reconstruction error within the uncertain boundary region to obtain the boundary-aware weighted residual: ,in, Represents pixels The original reconstruction error at the location, This represents the reconstruction error after boundary space weight modulation;
[0056] Then, in screen space statistics Internally Gaussian set formed by projection scale Local scale regularization constraints are applied to Gaussians with excessively large scale responses within the boundary zone. ,in, It is a monotonically increasing function;
[0057] Finally, combined with boundary-aware weighted residuals With local scale regularization right Joint optimization is performed to obtain the boundary optimization Gaussian set. .
[0058] The specific process of Step 3 is as follows:
[0059] First, optimize Gaussian sets based on boundaries. Perform differentiable Gaussian rendering to obtain the first... Frame pixels The original Gaussian rendering color at the location ;
[0060] Secondly, based on the boundary uncertainty strength Construct pixel reliability weights ,in Indicates the pixel position in screen space. ;
[0061] Then, combine , Construct reliability-aware synthetic results The calculation form is: ,in, Indicates the reference background color;
[0062] Finally, during the backpropagation process, As a boundary error propagation factor, for based on The calculated reconstruction residuals are gradient modulated to update the 3D Gaussian geometric parameters. and appearance parameters Its form is: , Output robust dressed human reconstruction results .
[0063] Example 2: As Figure 1-6 As shown, a geometrically consistent and boundary-aware monocular video method for clothing human reconstruction is presented. The specific steps of this method are as follows:
[0064] Step 1: Apply normal-guided constraints and pose-driven mapping to the monocular video frame sequences in the dressed human dataset. Gaussian residual displacement of human posture parameters By performing geometrically consistent deformation, we obtain the Gaussian set of the observation space. ;
[0065] Step 2: Based on the foreground mask and the Gaussian set of the observation space Boundary-aware optimization is performed through boundary uncertainty modeling and boundary constraint optimization, respectively, to obtain the boundary uncertainty intensity. and boundary optimization Gaussian sets ;
[0066] Step 3: Based on boundary uncertainty intensity and boundary optimization Gaussian sets By combining reliability perception and error propagation, decoupled Gaussian rendering is performed to output robust clothing human reconstruction results. .
[0067] The specific process of geometrically consistent deformation in Step 1 is as follows:
[0068] First, construct the Gaussian representation of the human body in the normal space. ,in, Indicates the first The location of the Gaussian center. According to the first... Human pose parameters in frames Human posture conditional features are constructed using a posture encoder. ;
[0069] Then, for the human body surface sampling points in the standardized space and human posture characteristics ,like Figure 2 As shown, through the shared attitude condition mapping function By studying the correspondence between changes in human posture and local non-rigid deformation of clothing, the posture-related Gaussian residual displacement is obtained. Among them, the attitude condition mapping function Implemented by a multi-layer residual sensing network, the computational form is as follows: ;
[0070] Secondly, according to Corresponding local normal direction break down The normal residual components were obtained respectively. and tangential residual components Its decomposition form is: , ;
[0071] Then, to Perform residual update to obtain the Gaussian center point in the normalized space. , with tangential residual components Constructing geometrically consistent deformation loss ,constraint The direction of updates;
[0072] Finally, according to the first The frame human pose parameter calculation standard space to observation space skeleton transformation relationship, combined with The corresponding skin weights will Mapping to the observation space yields the Gaussian set of the observation space. .
[0073] Step 2.1 Boundary Uncertainty Modeling:
[0074] First, obtain the foreground mask corresponding to the current video frame. Constructing the neighborhood boundary zone of the human body contour Its form is: ,in, and These represent expansion and corrosion operations, respectively. Indicates the first Foreground mask of the frame, Represents the boundary zone of the human body contour neighborhood. Indicates the width of the boundary band;
[0075] Then, in The distance from each pixel to the foreground region is calculated based on the screen space Euclidean distance transform. Distance from pixel to background area The boundary uncertainty intensity is obtained through normalized fusion and inversion mapping. ,like Figure 3 As shown. For the main human body region, due to sufficient foreground coverage, its boundary uncertainty is low; for the transition region of the human body contour, because both foreground and background participate in imaging, its boundary uncertainty gradually increases, therefore corresponding to... The value is higher;
[0076] Step 2.2 Boundary Constraint Optimization:
[0077] First, according to Generate boundary-aware weights : ,in, Indicates the boundary constraint weight coefficient. A binary identifier function representing an uncertain boundary region. Utilizing boundary-aware weights. Constructing boundary-aware weighted residuals: ,in, Represents pixels The original reconstruction error at the location, This represents the reconstruction error after boundary space weight modulation;
[0078] Then, in screen space statistics Internally Gaussian set formed by projection scale Local scale regularization constraints are applied to Gaussians with excessively large scale responses within the boundary zone. ,in, It is a monotonically increasing function;
[0079] Finally, combined with boundary-aware weighted residuals With local scale regularization right Perform joint optimization, such as Figure 4 As shown, the boundary-optimized Gaussian set is obtained. .
[0080] The specific process of Step 3 is as follows:
[0081] First, optimize Gaussian sets based on boundaries. Perform differentiable Gaussian rendering, such as Figure 5 As shown, the first... Frame pixels The original Gaussian rendering color at the location ;
[0082] Secondly, based on the boundary uncertainty strength The boundary confidence level is used to construct pixel reliability weights. ,in Indicates the pixel position in screen space. Pixels with higher reliability will contribute more gradient to the Gaussian parameter update, while unreliable boundary pixels will contribute less.
[0083] Then, combine , Construct reliability-aware synthetic results The calculation form is: ,in, Indicates the reference background color;
[0084] Finally, during the backpropagation process, As a boundary error propagation factor, for based on The calculated reconstruction residuals are gradient modulated to update the 3D Gaussian geometric parameters. and appearance parameters Its form is: , Output robust dressed human reconstruction results, see Figure 6 Compared with related known methods, the method of the present invention has stronger robustness in non-rigid deformation modeling of dressed human bodies, boundary structure preservation, and boundary error suppression.
[0085] The present invention also provides a geometrically consistent and boundary-aware monocular video-based human body reconstruction system, the system comprising: a module for performing the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0086] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0087] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0088] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the geometrically consistent and boundary-aware monocular video-based human body reconstruction method.
[0089] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A method for reconstructing a dressed human body from a monocular video using geometric consistency and boundary awareness, characterized in that: Includes the following steps: Step 1: Apply normal-guided constraints and pose-driven mapping to the monocular video frame sequences in the dressed human dataset. Gaussian residual displacement of human posture parameters By performing geometrically consistent deformation, we obtain the Gaussian set of the observation space. ; Step 2: Based on the foreground mask and the Gaussian set of the observation space Boundary-aware optimization is performed through boundary uncertainty modeling and boundary constraint optimization, respectively, to obtain the boundary uncertainty intensity. and boundary optimization Gaussian sets ; Step 3: Based on boundary uncertainty intensity and boundary optimization Gaussian sets By combining reliability perception and error propagation to perform decoupled Gaussian rendering, robust clothing human reconstruction results are output. .
2. The method for geometrically consistent and boundary-aware monocular video-based human body reconstruction according to claim 1, characterized in that: The specific process of geometrically consistent deformation in Step 1 is as follows: First, for human body surface sampling points in the standardized space According to the Human pose parameters in frames Constructing human posture conditional features Through the shared attitude condition mapping function The attitude-related Gaussian residual displacement is obtained. ; Secondly, according to Corresponding local normal direction break down The normal residual components were obtained respectively. and tangential residual components Its decomposition form is: , ; Then, to Perform residual update to obtain the Gaussian center point in the normalized space. , with tangential residual components Constructing geometrically consistent deformation loss ,constraint The direction of updates; Finally, according to the first The frame human pose parameter calculation standard space to observation space skeleton transformation relationship, combined with The corresponding skin weights will Mapping to the observation space yields the Gaussian set of the observation space. .
3. The method for geometrically consistent and boundary-aware monocular video-based clothing human reconstruction according to claim 1, characterized in that: The specific process of boundary-aware optimization in Step 2 is as follows: Step 2.1 Boundary Uncertainty Modeling: First, obtain the foreground mask corresponding to the current video frame. Constructing the neighborhood boundary zone of the human body contour Its form is: ,in, and These represent expansion and corrosion operations, respectively. Indicates the first Foreground mask of the frame, Represents the boundary zone of the human body contour neighborhood. Indicates the width of the boundary band; Then, according to The distances from pixels to the foreground and background regions are calculated based on screen-space Euclidean distance transform, and the boundary uncertainty intensity is obtained through normalized fusion and inversion mapping. ; Step 2.2 Boundary Constraint Optimization: First, according to Generate boundary space weights Weighted supervision is applied to the pixel reconstruction error within the uncertain boundary region to obtain the boundary-aware weighted residual: ,in, Represents pixels The original reconstruction error at the location, This represents the reconstruction error after boundary space weight modulation; Then, statistics Internally Gaussian set formed by projection scale Local scale regularization constraints are applied to Gaussians with excessively large scale responses within the boundary zone. ,in, It is a monotonically increasing function; Finally, combined with boundary-aware weighted residuals With local scale regularization right Joint optimization is performed to obtain the boundary optimization Gaussian set. .
4. The method for geometrically consistent and boundary-aware monocular video-based clothing human reconstruction according to claim 1, characterized in that: The specific process of decoupling Gaussian rendering in Step 3 is as follows: First, optimize Gaussian sets based on boundaries. Perform differentiable Gaussian rendering to obtain the first... Frame pixels The original Gaussian rendering color at the location ; Secondly, based on the intensity of boundary uncertainty Construct pixel reliability weights ,in Indicates the pixel position in screen space. ; Then, combine , Construct reliability-aware synthetic results The calculation form is: ,in, Indicates the reference background color; Finally, during the backpropagation process, As a boundary error propagation factor, for based on The calculated reconstruction residuals are gradient modulated to update the 3D Gaussian geometric parameters. and appearance parameters Its form is: , Output robust dressed human reconstruction results .
5. A geometrically consistent and boundary-aware monocular video-based human body reconstruction system, characterized in that, The system includes a module for performing a monocular video-based human body reconstruction method with geometric consistency and boundary awareness as described in any one of claims 1 to 4.
6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the geometrically consistent and boundary-aware monocular video method for reconstructing a dressed human body as described in any one of claims 1 to 4.
7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the geometrically consistent and boundary-aware monocular video method for clothing reconstruction as described in any one of claims 1 to 4.
8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the geometrically consistent and boundary-aware monocular video method for clothing reconstruction as described in any one of claims 1 to 4.
