CN113593033A - Three-dimensional model feature extraction method based on grid subdivision structure - Google Patents
Three-dimensional model feature extraction method based on grid subdivision structure Download PDFInfo
- Publication number
- CN113593033A CN113593033A CN202110649502.7A CN202110649502A CN113593033A CN 113593033 A CN113593033 A CN 113593033A CN 202110649502 A CN202110649502 A CN 202110649502A CN 113593033 A CN113593033 A CN 113593033A
- Authority
- CN
- China
- Prior art keywords
- grid
- model
- sampling
- mesh
- patch
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 101
- 238000005070 sampling Methods 0.000 claims abstract description 191
- 239000013598 vector Substances 0.000 claims abstract description 107
- 238000000034 method Methods 0.000 claims abstract description 65
- 240000004282 Grewia occidentalis Species 0.000 claims abstract description 7
- 230000004927 fusion Effects 0.000 claims description 19
- 238000013507 mapping Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 15
- 238000012549 training Methods 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 abstract description 12
- 238000010586 diagram Methods 0.000 description 11
- 238000010801 machine learning Methods 0.000 description 7
- 238000012545 processing Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000010146 3D printing Methods 0.000 description 1
- 241000283973 Oryctolagus cuniculus Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000010171 animal model Methods 0.000 description 1
- 210000000746 body region Anatomy 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
- G06T17/205—Re-meshing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Image Generation (AREA)
Abstract
The invention provides a three-dimensional model feature extraction method based on a grid subdivision structure, which comprises the following steps: determining a mesh representation of a three-dimensional model of features to be extracted; inputting the grid of the three-dimensional model into a feature extraction module, and outputting a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation; the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected. The method provided by the invention has the advantages that the extracted features can better describe the wide-range features, and the method is applied to grid classification, segmentation, retrieval, shape correspondence and the like, and can improve the accuracy and the robustness.
Description
Technical Field
The invention relates to the technical field of three-dimensional grids, in particular to a three-dimensional model feature extraction method based on a grid subdivision structure
Background
Mesh (Mesh) is a common description of three-dimensional shapes and is widely used for modeling, rendering, animation, 3D printing, and the like. Typically, a mesh defines vertices in three-dimensional space, and patches defined by the connections between the vertices. The most commonly used meshes are triangular meshes and quadrilateral meshes: all patches in the triangular mesh are triangles and all patches in the quadrilateral mesh are quadrilaterals.
The deep neural network has made a significant progress compared with the traditional method in the comprehension analysis and generation of data formats such as images, natural language processing, point clouds and the like, but the application of the deep neural network is less due to the complexity and irregularity of grid data.
The existing three-dimensional grid feature extraction method usually uses a machine learning mode to train a three-dimensional grid model, in the model training process, the existing three-dimensional grid model needs to be subjected to feature extraction of different rules to obtain grid features of different resolutions, the aggregation of local features or the propagation of global features is realized, which is equivalent to the way of obtaining feature maps of different scales in two-dimensional image processing, then fusion is carried out based on the feature maps of different scales to obtain the final overall feature vector of the image, the three-dimensional grid model also needs to extract grids of different resolutions, and then the final feature of each patch and the overall feature of the grid are determined based on the grid features of different resolutions.
The establishment of multi-resolution hierarchical representation for three-dimensional data can extract features from thin to thick and from local to global, and is one of the reasons why a neural network can be effectively applied to regular data such as images and sequences. While a typical grid lacks such a regular hierarchical structure. In the prior art, features are stored on edges, coarse-grained representation is obtained by continuously and dynamically deleting the edges on the basis of grid simplification, but the deleting method is uneven in detail and slow in speed.
Therefore, how to avoid the problem that the reasonable feature representation is difficult to obtain due to the lack of the multi-resolution hierarchical representation of the three-dimensional data in the conventional three-dimensional mesh feature extraction method is still a problem to be solved by those skilled in the art.
Disclosure of Invention
The invention provides a three-dimensional model feature extraction method based on a mesh subdivision structure, which is used for solving the problem that reasonable feature representation is difficult to obtain due to the lack of hierarchical representation of three-dimensional data multi-resolution in the conventional three-dimensional mesh feature extraction method.
The invention provides a three-dimensional model feature extraction method based on a grid subdivision structure, which comprises the following steps:
determining a mesh representation of a three-dimensional model of features to be extracted;
inputting the grid of the three-dimensional model into a feature extraction module, and outputting a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation;
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
According to the three-dimensional model feature extraction method based on the grid subdivision structure, provided by the invention, the feature extraction module comprises a re-gridding module and a grid up-down sampling module which are sequentially connected, and the method specifically comprises the following steps:
the regridding module performs regridding division on the grids input into the three-dimensional model to obtain a regridding model with a subdivision structure and outputs the regridding model to the grid upper and lower sampling module;
and the grid up-down sampling module performs down-sampling and up-sampling on the re-grid model for a plurality of times to obtain a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation.
According to the three-dimensional model feature extraction method based on the mesh subdivision structure provided by the invention, the mesh of the input original three-dimensional model is subjected to re-mesh division to obtain a re-mesh model with the subdivision structure, and the method specifically comprises the following steps:
simplifying the input three-dimensional model original grid to obtain a simplified grid, and establishing mapping between the simplified grid and the three-dimensional model original grid;
subdividing the simplified grid by using a preset subdivision rule to obtain a grid model with a subdivided structure;
and mapping the grid model back to the original grid of the three-dimensional model based on the mapping to obtain a re-grid model with a subdivision structure of the original grid of the three-dimensional model.
According to the three-dimensional model feature extraction method based on the mesh subdivision structure provided by the invention, the heavy mesh model is subjected to down sampling and up sampling for a plurality of times to obtain a target mesh representation after feature extraction and a target feature vector of each patch in the target mesh representation, and the method specifically comprises the following steps:
the m-th sampling layer of the grid up-down sampling module samples the m-1 th sampling layer characteristic vector and the m-1 th sampling layer grid model of each patch according to a preset rule, outputs the m-th sampling layer characteristic vector and the m-th sampling layer grid model of each patch to an m +1 th sampling layer, outputs the L-th layer characteristic vector, and is an up-sampling layer or a down-sampling layer, and the preset rule corresponding to the m-th sampling layer is the preset up-sampling rule or the preset down-sampling rule;
and m is 1,2, …, and L is the total number of sampling layers in the grid up-down sampling module, the zeroth sampling layer feature vector is formed by the patch shape description and the patch posture description of the heavy grid model, and the zeroth sampling layer grid model is the heavy grid model.
According to the three-dimensional model feature extraction method based on the grid subdivision structure provided by the invention, the m-th sampling layer samples the input m-1-th sampling layer feature vector and the m-1-th sampling layer grid model of each patch according to a preset rule, and outputs the m-th sampling layer feature vector and the m-th sampling layer grid model of each patch, and the method specifically comprises the following steps:
if the mth sampling layer is the upsampling layer,
splitting each patch input into the m-th sampling layer into four patches according to a preset subdivision rule to obtain an m-th sampling layer grid model; determining the m sampling layer characteristic vectors of the four surface patches according to a preset propagation rule based on the input m-1 sampling layer characteristic vector of each surface patch;
if the mth sampling layer is a downsampling layer,
combining four surface patches in the same subdivision structure in the grid model input into the m-th sampling layer to obtain a grid model of the m-th sampling layer; and determining the m sampling layer characteristic vectors of the four patches according to a preset fusion rule based on the input m-1 sampling layer characteristic vector of each patch.
According to the three-dimensional model feature extraction method based on the mesh subdivision structure, provided by the invention, if the mesh is a triangular mesh, the preset subdivision rule is a Loop subdivision rule;
and if the grid is a quadrangle grid, the preset subdivision rule is a Catmull-Clerk subdivision rule.
According to the three-dimensional model feature extraction method based on the grid subdivision structure, provided by the invention, the preset fusion rule is an average value, a maximum value or a minimum value of each dimension of feature vectors of four surface patches.
According to the three-dimensional model feature extraction method based on the grid subdivision structure, provided by the invention, the preset propagation rule is nearest upsampling or bilinear upsampling.
The invention also provides a multi-resolution deep neural network for three-dimensional grids, which comprises the feature extraction module and the specific task module in any three-dimensional model feature extraction method based on the grid subdivision structure, and is characterized in that:
in the course of the training process,
the feature extraction module extracts a sample feature vector of each patch of the input sample three-dimensional grid model and outputs the sample feature vector to the specific task module;
the specific task module processes the sample feature vector by a specific rule to obtain a prediction result;
constructing a loss function by using a preset learning target based on the prediction result and the patch label construction; the preset learning target is grid classification or triangular grid model object type identification.
The invention also provides a three-dimensional model feature extraction device based on the mesh subdivision structure, which comprises the following steps:
an input unit for determining a mesh representation of a three-dimensional model of features to be extracted;
the output unit is used for inputting the grids of the three-dimensional model into the feature extraction module, outputting the target grid representation after feature extraction and the target feature vector of each patch in the target grid representation;
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
The invention also provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the steps of any one of the three-dimensional model feature extraction methods based on the grid subdivision structure.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when being executed by a processor, carries out the steps of any of the above-mentioned three-dimensional model feature extraction methods based on a mesh subdivision structure.
The invention provides a three-dimensional model feature extraction method based on a mesh subdivision structure, which comprises the steps of determining mesh representation of a three-dimensional model with features to be extracted, inputting meshes of the three-dimensional model into a feature extraction module, outputting the mesh representation after feature extraction and feature vectors of each patch of the meshes, and adding a subdivision structure into the meshes of an original three-dimensional model through a preset surface subdivision rule by a re-meshing module. The new multi-resolution layering method is provided as the re-gridding module and the grid up-down sampling module which are sequentially connected are built in the feature extraction module, the subdivision structure is added to carry out re-gridding on the original grid model in a surface subdivision mode, and then the re-gridding module is subjected to up-down sampling to obtain different resolution features of each patch, so that the method can support the traditional machine learning and deep learning, is applied to three-dimensional model classification, segmentation, retrieval, shape correspondence and the like, and can improve the high accuracy and robustness of the system.
Drawings
In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.
FIG. 1 is a schematic flow chart of a three-dimensional model feature extraction method based on a mesh subdivision structure according to the present invention;
FIG. 2 is an exemplary diagram of a re-gridding result and a downsampling process provided by the present invention;
FIG. 3 is an exemplary diagram of a grid of subdivision structures provided by the present invention;
FIG. 4 is a schematic diagram of up-down sampling computation of a three-dimensional grid according to the present invention;
FIG. 5 is a schematic structural diagram of a three-dimensional model feature extraction device based on a mesh subdivision structure according to the present invention;
fig. 6 is a schematic physical structure diagram of an electronic device provided in the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Because the existing three-dimensional grid feature extraction method lacks the multi-resolution hierarchical representation of three-dimensional data, the problem that reasonable feature representation is difficult to obtain is caused. The following describes a three-dimensional model feature extraction method based on a mesh subdivision structure in accordance with the present invention with reference to fig. 1 to 4. Fig. 1 is a three-dimensional model feature extraction method based on a mesh subdivision structure, as shown in fig. 1, the method includes:
Specifically, a mesh representation of a three-dimensional model of the features to be extracted is first determined. The three-dimensional model can be various real objects or scenes, such as a character animal model, a clothing fabric model, a human body organ model, a furniture model, an industrial product model, a film and television special effect model, a sculpture building model, an indoor scene model and the like.
The three-dimensional model mesh representation is composed of polygonal patches, and the mesh representation referred to by the invention refers to triangular meshes or quadrilateral meshes. There are various ways to determine the triangular mesh of the three-dimensional model, for example, a triangular mesh artificially constructed by using modeling software or automatically generated according to a certain algorithm, or a three-dimensional point cloud constructed based on multi-viewpoint images of objects or scenes acquired in various directions and then subjected to gridding processing, or a three-dimensional point cloud of objects or scenes obtained based on omnidirectional radar scanning and then subjected to gridding processing, and the like, which are not limited in detail herein. The three-dimensional model feature extraction method provided by the present invention extracts feature vectors for the three-dimensional model on the basis of the mesh of the original three-dimensional model determined in the current step 110.
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
The feature vectors output by the feature extraction module are obtained after a plurality of times of downsampling or upsampling in the feature extraction module, each downsampling can enable input feature vectors to be aggregated, and each upsampling can enable the feature vectors to be spread, so that the feature vectors output by the feature extraction module are aggregated of wide-range input feature vectors, and can describe the features of a wide-range input patch.
The output feature vector can support a traditional machine learning method, such as a support vector machine. Specifically, the output feature vector is used as input data for training a machine learning model, and the learning objective of the machine learning model can be tasks of solving three-dimensional model classification, three-dimensional model segmentation, three-dimensional model retrieval, three-dimensional model shape correspondence and the like. The method can also be applied to deep learning. Specifically, one or more feature extraction modules are used as a network layer of the deep neural network, and are used for outputting feature vectors output after passing through the feature extraction module as input of a next layer of the deep neural network by taking the depth features of each patch as input.
The invention provides a three-dimensional model feature extraction method based on a mesh subdivision structure, which comprises the steps of determining mesh representation of a three-dimensional model with features to be extracted, inputting meshes of the three-dimensional model into a feature extraction module, outputting the mesh representation after feature extraction and feature vectors of each patch of the meshes, and adding a subdivision structure into the meshes of an original three-dimensional model through a preset surface subdivision rule by a re-meshing module. The new multi-resolution layering method is provided as the re-gridding module and the grid up-down sampling module which are sequentially connected are built in the feature extraction module, the subdivision structure is added to carry out re-gridding on the original grid model in a surface subdivision mode, and then the re-gridding module is subjected to up-down sampling to obtain different resolution features of each patch, so that the method can support the traditional machine learning and deep learning, is applied to three-dimensional model classification, segmentation, retrieval, shape correspondence and the like, and can improve the high accuracy and robustness of the system.
Based on the above embodiment, in the method, the feature extraction module includes a re-gridding module and a grid up-down sampling module that are sequentially connected, and specifically includes:
the regridding module performs regridding division on the grids input into the three-dimensional model to obtain a regridding model with a subdivision structure and outputs the regridding model to the grid upper and lower sampling module;
and the grid up-down sampling module performs down-sampling and up-sampling on the re-grid model for a plurality of times to obtain a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation.
Specifically, the regridding module and the grid up-down sampling module are combined to extract different resolution features of each patch in the original three-dimensional grid.
Based on the above embodiment, in the method, the re-meshing of the mesh of the input original three-dimensional model to obtain the re-meshed model with the subdivided structure specifically includes:
simplifying the input three-dimensional model original grid to obtain a simplified grid, and establishing mapping between the simplified grid and the three-dimensional model original grid;
subdividing the simplified grid by using a preset subdivision rule to obtain a grid model with a subdivided structure;
and mapping the grid model back to the original grid of the three-dimensional model based on the mapping to obtain a re-grid model with a subdivision structure of the original grid of the three-dimensional model.
Specifically, fig. 2 is an exemplary diagram of a re-gridding result and a down-sampling process provided by the present invention, and as shown in fig. 2, a self-parameterization method is used to simplify a mesh input into a three-dimensional model, and establish a parameterization mapping between the simplified mesh and an original mesh. There are many ways to establish mapping, preferably, a multi-resolution adaptive parameterized MAPS method is adopted, and the method is simplified by deleting vertices, so that a multilayer conformal mapping is established between the simplified mesh and the original mesh. And subdividing the simplified grid, and mapping the subdivided grid back to the original three-dimensional grid to obtain a regridded result of the original three-dimensional grid. At this time, the connection relation of the nodes of the mesh conforms to the tessellation rule. It should be noted here that a patch in the original mesh model may be a triangle or a quadrangle, and if the input original three-dimensional mesh is a triangular mesh, the used subdivision rule is Loop subdivision, and if the input original three-dimensional mesh is a quadrangular mesh, the used subdivision rule is Catmull-Clerk subdivision. In the following description, a triangle patch is taken as an example, and the case of a quadrilateral mesh is similar to the case of a triangle patch.
Fig. 3 is an exemplary diagram of a mesh with a subdivision structure provided by the present invention, and as shown in fig. 3, the left and right sides are three-dimensional mesh models of the same rabbit, the three-dimensional mesh model on the left side is obtained by adding the subdivision structure on the basis of the three-dimensional mesh model on the right side, the left and right mesh models both draw out a same large patch at the same position, and four small patches are filled into the large patch on the left side relative to the same patch on the right side, which is an example of adding the subdivision structure. If the mesh has a subdivided structure, then adjacent patches can be merged into one large patch, resulting in a coarser-grained, low-resolution mesh. Meanwhile, the low-resolution grid can also obtain a high-resolution grid by splitting the patches.
Based on the above embodiment, in the method, the performing downsampling and upsampling on the re-mesh model for several times to obtain a target mesh representation after feature extraction and a target feature vector of each patch in the target mesh representation specifically includes:
the m-th sampling layer of the grid up-down sampling module samples the m-1 th sampling layer characteristic vector and the m-1 th sampling layer grid model of each patch according to a preset rule, outputs the m-th sampling layer characteristic vector and the m-th sampling layer grid model of each patch to an m +1 th sampling layer, outputs the L-th layer characteristic vector, and is an up-sampling layer or a down-sampling layer, and the preset rule corresponding to the m-th sampling layer is the preset up-sampling rule or the preset down-sampling rule;
and m is 1,2, …, and L is the total number of sampling layers in the grid up-down sampling module, the zeroth sampling layer feature vector is formed by the patch shape description and the patch posture description of the heavy grid model, and the zeroth sampling layer grid model is the heavy grid model.
Specifically, the grid up-down sampling module comprises a plurality of sampling layers, the connection relationship of the sampling layers is a cascade form, namely, a connection relationship in sequence, each sampling layer can be an up-sampling layer or a down-sampling layer, and the arrangement combination mode and the number of the layers of the up-sampling layer and the down-sampling layer in the grid up-down sampling module are determined according to the requirement of feature extraction.
Based on the above embodiment, in the method, the sampling at the m-th sampling layer is performed by sampling, by the m-th sampling layer, the input m-1-th sampling layer feature vector and m-1-th sampling layer mesh model of each patch according to a preset rule, and outputting the m-th sampling layer feature vector and the m-th sampling layer mesh model of each patch, which specifically includes:
if the mth sampling layer is the upsampling layer,
splitting each patch input into the m-th sampling layer into four patches according to a preset subdivision rule to obtain an m-th sampling layer grid model; determining the m sampling layer characteristic vectors of the four surface patches according to a preset propagation rule based on the input m-1 sampling layer characteristic vector of each surface patch;
if the mth sampling layer is a downsampling layer,
combining four surface patches in the same subdivision structure in the grid model input into the m-th sampling layer to obtain a grid model of the m-th sampling layer; and determining the m sampling layer characteristic vectors of the four patches according to a preset fusion rule based on the input m-1 sampling layer characteristic vector of each patch.
Specifically, the above specifically defines the specific operations of each sampling layer, including how to process the mesh model and the corresponding feature vectors of each patch. The downsampling layer downsamples according to a preset downsampling rule, and specifically comprises the following steps: combining four surface patches in the same subdivision structure in a current input grid model to obtain an output grid, wherein the feature vectors of the input grid are fused according to the feature vectors of the four surface patches by a preset fusion rule; the upsampling layer performs upsampling according to a preset upsampling rule, and specifically comprises: splitting each patch in a current input mesh model into four patches according to a preset subdivision rule, and obtaining feature vectors of the four patches according to a preset propagation rule based on the feature vectors of the input patches
Based on the above embodiment, in the method, if the mesh is a triangular mesh, the preset subdivision rule is a Loop subdivision rule;
and if the grid is a quadrangle grid, the preset subdivision rule is a Catmull-Clerk subdivision rule.
Specifically, if the processing object is a triangular mesh, the processing object uses a Loop subdivision rule to divide one triangular patch into four triangular patches; if the processing object is a quadrilateral mesh, the processing object uses a Catmull-Clerk subdivision rule to split one quadrilateral patch into four quadrilateral patches.
Based on the above embodiment, in the method, the preset fusion rule is an average value, a maximum value, or a minimum value of each dimension of the feature vectors of the four patches.
Specifically, the preset fusion rule is to take an average value, a maximum value or a minimum value of each dimension of the feature vector. The three fusion rules belong to a fusion mode with the minimum calculated amount, and the characteristics of the fused low-resolution patch are generated in the simplest mode under the condition that the characteristic information of the original patch is kept. Fig. 4 is a schematic diagram of up-down sampling calculation of a three-dimensional mesh provided by the present invention, and as shown in fig. 4, the first diagram from left shows a down-sampling process for integrating patches containing subdivision structures into a large lower resolution patch (the large patch in fig. 4)A large triangle including 4 small triangles is integrated into a final large triangle), and the feature vectors of the fused large patch are obtained by fusing the feature vectors of the four small triangle patches included in the subdivision structure, that is, the feature vectors of the 4 small patches are assumed to be e1,e2,e3,e4The dimension of the feature is d; assuming that the feature vector of the fused large patch is E, the feature dimension is d, and the fusion rule of each dimension taking the mean value is as follows:
similarly, the fusion rule for each dimension taking the maximum value is,
Ei＝max{e1,i，e2,i，e3,i，e4,i}
the fusion rule for each dimension taking the minimum value is,
Ei＝min{e1,i，e2,i，e3,i，e4,i}
based on the above embodiment, in the method, the preset propagation rule is nearest neighbor upsampling or bilinear upsampling.
Specifically, if nearest neighbor upsampling is used, the feature vectors of the four patches obtained by splitting are all the feature vectors of the input patches; if bilinear upsampling is used, the feature vector of the split patch is obtained by interpolating the feature vectors of the patch before splitting and the patch adjacent to the patch, and the interpolation proportion is the inverse proportion of the distance from the center of the patch before splitting to the patch after splitting. And during upsampling, transmitting the characteristics of the low-resolution grid patch to the corresponding high-resolution grid patch, wherein the transmission mode comprises nearest neighbor upsampling, bilinear upsampling and the like. Nearest neighbor upsampling as the second "nearest neighbor upsampling" display process from left to right in fig. 4, each low-resolution patch feature is directly propagated to the corresponding high-resolution patch; bilinear upsampling as shown in the second "bilinear upsampling" display process from left to right in fig. 4, the high resolution feature is obtained by interpolation of adjacent low resolution patches, and the interpolation ratio is the inverse ratio of the distance from the high resolution patch to the low resolution patch.
The method provided by the invention is mainly characterized in that the method for extracting the layering features of the three-dimensional grids with different resolutions is improved, and firstly, the input grids are preprocessed to have a subdivision structure. Specifically, a self-parameterization method is used, the mesh of the input three-dimensional model is simplified firstly, parameterization mapping of the simplified mesh and the original mesh is established, and the mesh of the input three-dimensional model is regridded, so that the connection relation of mesh nodes accords with a surface subdivision rule. In this network, the features of the mesh are stored on a patch. For a mesh that has been re-gridded, its patch input features are two types including shape features and pose features. The shape characteristics comprise the area of the patch, the size of the inner angle of the patch and the curvature of the vertex of the patch; the pose characteristics comprise the center coordinates of the surface patch and the normal phase direction of the surface patch. Based on the nature of the subdivision structure, every fourth patch can be merged into one patch, so regular downsampling and upsampling can be performed. During down-sampling, finding out a plurality of patches of the high-resolution grid, merging the corresponding patches of the low-resolution grid according to a subdivision connection structure, aggregating the features of the high-resolution grid, and transmitting the aggregated features to the low-resolution grid, wherein the aggregation mode of the features comprises calculation of an average value, a maximum value, a minimum value and the like; and during upsampling, transmitting the characteristics of the low-resolution grid patch to the corresponding high-resolution grid patch, wherein the transmission mode comprises nearest neighbor upsampling, bilinear upsampling and the like.
And constructing a multi-resolution grid neural network based on the above to learn the characteristics of the three-dimensional grid surface patch. The network comprises a down-sampling layer and an up-sampling layer, which respectively enable the grid resolution to be reduced or increased, thereby realizing the aggregation of local features or the propagation of global features.
Based on the above embodiments, the present invention further provides a multi-resolution deep neural network for three-dimensional mesh, comprising the feature extraction module and the task-specific module as in any of the above embodiments,
in the course of the training process,
the feature extraction module extracts a sample feature vector of each patch of the input sample three-dimensional grid model and outputs the sample feature vector to the specific task module;
the specific task module processes the sample feature vector by a specific rule to obtain a prediction result;
constructing a loss function by using a preset learning target based on the prediction result and the patch label construction; the preset learning target is grid classification or triangular grid model object type identification.
Specifically, for example, if the preset learning target is a human body mesh segmentation method, the human body mesh segmentation method includes the following steps: determining an original three-dimensional grid of the human body to be detected; inputting the original three-dimensional mesh into a segmentation model, and outputting the region to which each patch in the original three-dimensional mesh belongs, wherein the region comprises a head, four limbs and a trunk; the segmentation model is obtained by training based on a sample original three-dimensional grid and a corresponding human body region label, and comprises the re-meshing module, the grid up-down sampling module and the classification module which are sequentially connected, wherein the re-meshing module is used for adding a subdivision structure into the original three-dimensional grid through a preset surface subdivision rule. In the object class identification method, in the training process of the classification model, a feature extraction module performs convolution calculation on the features of each patch in an input sample original three-dimensional grid according to a preset rule to obtain a feature vector of each patch and outputs the feature vector to a feature fusion module; the feature fusion module fuses the feature vectors of each patch according to a preset fusion rule to obtain the overall feature vector of the original three-dimensional grid of the sample and outputs the overall feature vector to the classified full-connection layer; the classification full-connection layer determines a predicted object type corresponding to the sample original three-dimensional grid based on the integral feature vector; and the predicted object type is used for constructing a loss function during the training of the classification model, and convolution calculation is performed on any patch according to a preset rule to aggregate neighborhood patch characteristics of any patch so as to extract local characteristics of any patch. Further, the segmentation model includes a regrooving module, a grid up-down sampling module and a classification module which are connected in sequence, and specifically includes: in the training process of the segmentation model, the regrooving module performs regrooving division on an input original three-dimensional grid of the sample to obtain a regrooving model comprising a subdivision structure and outputs the regrooving model to the grid up-down sampling module; the grid up-down sampling module sequentially performs a plurality of down-sampling and up-sampling of the heavy grid model in any combination form to obtain a target feature vector of each patch and outputs the target feature vector to the classification module; the classification module determines a prediction region to which each patch belongs by using a fully-connected layer sharing parameters based on the target feature vector of each patch; and the prediction region is used for constructing a loss function in the training of the segmentation model.
The three-dimensional model feature extraction device based on the mesh subdivision structure provided by the invention is described below, and the three-dimensional model feature extraction device described below and the three-dimensional model feature extraction method described above can be referred to in a mutually corresponding manner.
Fig. 5 is a schematic structural diagram of a three-dimensional model feature extraction apparatus based on a mesh subdivision structure according to the present invention, as shown in fig. 5, the apparatus includes an input unit 510 and an output unit 520, wherein,
the input unit 510 is configured to determine a mesh representation of a three-dimensional model of features to be extracted;
the output unit 520 is configured to input the mesh of the three-dimensional model into the feature extraction module, and output a target mesh representation after feature extraction and a target feature vector of each patch in the target mesh representation;
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
According to the device provided by the invention, the grid representation of the three-dimensional model of the features to be extracted is determined, the grid of the three-dimensional model is input into the feature extraction module, the grid representation after the feature extraction and the feature vector of each surface patch of the grid are output, and the re-gridding module is used for adding a subdivision structure into the grid of the original three-dimensional model through a preset surface subdivision rule. The new multi-resolution layering method is provided as the re-gridding module and the grid up-down sampling module which are sequentially connected are built in the feature extraction module, the subdivision structure is added to carry out re-gridding on the original grid model in a surface subdivision mode, and then the re-gridding module is subjected to up-down sampling to obtain different resolution features of each patch, so that the method can support the traditional machine learning and deep learning, is applied to three-dimensional model classification, segmentation, retrieval, shape correspondence and the like, and can improve the high accuracy and robustness of the system.
Based on the above embodiment, in the apparatus, the feature extraction module includes a re-gridding module and a grid up-down sampling module that are connected in sequence, and specifically includes:
the regridding module performs regridding division on the grids input into the three-dimensional model to obtain a regridding model with a subdivision structure and outputs the regridding model to the grid upper and lower sampling module;
and the grid up-down sampling module performs down-sampling and up-sampling on the re-grid model for a plurality of times to obtain a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation.
Based on the above embodiment, in the apparatus, the re-meshing the mesh of the input original three-dimensional model to obtain a re-meshed model with a subdivided structure specifically includes:
simplifying the input three-dimensional model original grid to obtain a simplified grid, and establishing mapping between the simplified grid and the three-dimensional model original grid;
subdividing the simplified grid by using a preset subdivision rule to obtain a grid model with a subdivided structure;
and mapping the grid model back to the original grid of the three-dimensional model based on the mapping to obtain a re-grid model with a subdivision structure of the original grid of the three-dimensional model.
Based on the above embodiment, in the apparatus, the performing downsampling and upsampling on the re-mesh model for several times to obtain a target mesh representation after feature extraction and a target feature vector of each patch in the target mesh representation specifically includes:
the m-th sampling layer of the grid up-down sampling module samples the m-1 th sampling layer characteristic vector and the m-1 th sampling layer grid model of each patch according to a preset rule, outputs the m-th sampling layer characteristic vector and the m-th sampling layer grid model of each patch to an m +1 th sampling layer, outputs the L-th layer characteristic vector, and is an up-sampling layer or a down-sampling layer, and the preset rule corresponding to the m-th sampling layer is the preset up-sampling rule or the preset down-sampling rule;
and m is 1,2, …, and L is the total number of sampling layers in the grid up-down sampling module, the zeroth sampling layer feature vector is formed by the patch shape description and the patch posture description of the heavy grid model, and the zeroth sampling layer grid model is the heavy grid model.
Based on the above embodiment, in the apparatus, the sampling at the m-th layer samples the input m-1-th sampling layer feature vector and the m-1-th sampling layer mesh model of each patch according to a preset rule, and outputs the m-th sampling layer feature vector and the m-th sampling layer mesh model of each patch, which specifically includes:
if the mth sampling layer is the upsampling layer,
splitting each patch input into the m-th sampling layer into four patches according to a preset subdivision rule to obtain an m-th sampling layer grid model; determining the m sampling layer characteristic vectors of the four surface patches according to a preset propagation rule based on the input m-1 sampling layer characteristic vector of each surface patch;
if the mth sampling layer is a downsampling layer,
combining four surface patches in the same subdivision structure in the grid model input into the m-th sampling layer to obtain a grid model of the m-th sampling layer; and determining the m sampling layer characteristic vectors of the four patches according to a preset fusion rule based on the input m-1 sampling layer characteristic vector of each patch.
Based on the above embodiment, in the apparatus, if the mesh is a triangular mesh, the preset subdivision rule is a Loop subdivision rule;
and if the grid is a quadrangle grid, the preset subdivision rule is a Catmull-Clerk subdivision rule.
Based on the above embodiment, in the apparatus, the preset fusion rule is an average value, a maximum value, or a minimum value of each dimension of the feature vectors of the four patches.
Based on the above embodiment, in the apparatus, the preset propagation rule is nearest neighbor upsampling or bilinear upsampling.
Fig. 6 is a schematic entity structure diagram of an electronic device provided in the present invention, and as shown in fig. 6, the electronic device may include: a processor (processor)610, a communication Interface (Communications Interface)620, a memory (memory)630 and a communication bus 640, wherein the processor 610, the communication Interface 620 and the memory 630 communicate with each other via the communication bus 640. The processor 610 may invoke logic instructions in the memory 630 to perform a method for three-dimensional model feature extraction based on mesh subdivision, the method comprising: giving a grid representation of a three-dimensional model of features to be extracted, inputting the grid into a feature extraction module, and outputting feature vectors of all surface patches of the grid; the feature extraction module comprises a regridding module and a grid up-down sampling module which are sequentially connected. .
In addition, the logic instructions in the memory 630 may be implemented in software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, the computer being capable of executing the three-dimensional model feature extraction method based on a mesh subdivision structure provided by the above methods, the method including: giving a grid representation of a three-dimensional model of features to be extracted, inputting the grid into a feature extraction module, and outputting feature vectors of all surface patches of the grid; the feature extraction module comprises a regridding module and a grid up-down sampling module which are sequentially connected.
In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program, when being executed by a processor, implementing a three-dimensional model feature extraction method based on a mesh subdivision structure, the method being provided by the above methods, the method including: giving a grid representation of a three-dimensional model of features to be extracted, inputting the grid into a feature extraction module, and outputting feature vectors of all surface patches of the grid; the feature extraction module comprises a regridding module and a grid up-down sampling module which are sequentially connected.
The above-described server embodiments are only illustrative, and the units described as separate components may or may not be physically separate, and components displayed as units may or may not be physical units, may be located in one place, or 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 solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (12)
1. A three-dimensional model feature extraction method based on a mesh subdivision structure is characterized by comprising the following steps:
determining a mesh representation of a three-dimensional model of features to be extracted;
inputting the grid of the three-dimensional model into a feature extraction module, and outputting a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation;
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
2. The method for extracting features of a three-dimensional model based on a mesh subdivision structure according to claim 1, wherein the feature extraction module includes a re-meshing module and a mesh up-down sampling module which are sequentially connected, and specifically includes:
the regridding module performs regridding division on the grids input into the three-dimensional model to obtain a regridding model with a subdivision structure and outputs the regridding model to the grid upper and lower sampling module;
and the grid up-down sampling module performs down-sampling and up-sampling on the re-grid model for a plurality of times to obtain a target grid representation after feature extraction and a target feature vector of each patch in the target grid representation.
3. The method for extracting features of a three-dimensional model based on a mesh subdivision structure according to claim 1 or 2, wherein the re-meshing of the mesh of the input original three-dimensional model to obtain a re-meshed model with a subdivision structure comprises:
simplifying the input three-dimensional model original grid to obtain a simplified grid, and establishing mapping between the simplified grid and the three-dimensional model original grid;
subdividing the simplified grid by using a preset subdivision rule to obtain a grid model with a subdivided structure;
and mapping the grid model back to the original grid of the three-dimensional model based on the mapping to obtain a re-grid model with a subdivision structure of the original grid of the three-dimensional model.
4. The method for extracting features of a three-dimensional model based on a mesh subdivision structure according to claim 1 or 2, wherein the performing down-sampling and up-sampling on the re-mesh model for several times to obtain a target mesh representation after feature extraction, and a target feature vector of each patch in the target mesh representation specifically includes:
the m-th sampling layer of the grid up-down sampling module samples the m-1 th sampling layer characteristic vector and the m-1 th sampling layer grid model of each patch according to a preset rule, outputs the m-th sampling layer characteristic vector and the m-th sampling layer grid model of each patch to an m +1 th sampling layer, outputs the L-th layer characteristic vector, and is an up-sampling layer or a down-sampling layer, and the preset rule corresponding to the m-th sampling layer is the preset up-sampling rule or the preset down-sampling rule;
and m is 1,2, …, and L is the total number of sampling layers in the grid up-down sampling module, the zeroth sampling layer feature vector is formed by the patch shape description and the patch posture description of the heavy grid model, and the zeroth sampling layer grid model is the heavy grid model.
5. The method for extracting features of a three-dimensional model based on a mesh subdivision structure according to claim 4, wherein the m-th sampling layer samples the input m-1-th sampling layer feature vector and the m-1-th sampling layer mesh model of each patch according to a preset rule, and outputs the m-th sampling layer feature vector and the m-th sampling layer mesh model of each patch, specifically comprising:
if the mth sampling layer is the upsampling layer,
splitting each patch input into the m-th sampling layer into four patches according to a preset subdivision rule to obtain an m-th sampling layer grid model; determining the m sampling layer characteristic vectors of the four surface patches according to a preset propagation rule based on the input m-1 sampling layer characteristic vector of each surface patch;
if the mth sampling layer is a downsampling layer,
combining four surface patches in the same subdivision structure in the grid model input into the m-th sampling layer to obtain a grid model of the m-th sampling layer; and determining the m sampling layer characteristic vectors of the four patches according to a preset fusion rule based on the input m-1 sampling layer characteristic vector of each patch.
6. The method of extracting features of a three-dimensional model based on a mesh subdivision structure according to claim 5,
if the grid is a triangular grid, the preset subdivision rule is a Loop subdivision rule;
and if the grid is a quadrangle grid, the preset subdivision rule is a Catmull-Clerk subdivision rule.
7. The method of claim 5, wherein the preset fusion rule is an average value, a maximum value or a minimum value of each dimension of the feature vectors of the four patches.
8. The method of claim 5, wherein the predetermined propagation rule is nearest neighbor upsampling or bilinear upsampling.
9. A multi-resolution deep neural network for three-dimensional mesh, comprising the feature extraction module and the task-specific module in the mesh-subdivision-based three-dimensional model feature extraction method according to any one of claims 1 to 7, characterized in that:
in the course of the training process,
the feature extraction module extracts a sample feature vector of each patch of the input sample three-dimensional grid model and outputs the sample feature vector to the specific task module;
the specific task module processes the sample feature vector by a specific rule to obtain a prediction result;
constructing a loss function by using a preset learning target based on the prediction result and the patch label construction; the preset learning target is grid classification or triangular grid model object type identification.
10. A three-dimensional model feature extraction device based on a mesh subdivision structure is characterized by comprising:
an input unit for determining a mesh representation of a three-dimensional model of features to be extracted;
the output unit is used for inputting the grids of the three-dimensional model into the feature extraction module, outputting the target grid representation after feature extraction and the target feature vector of each patch in the target grid representation;
the grid representation is composed of a plurality of triangular patches or a plurality of four-corner patches and is used for describing the three-dimensional shape of a real object or a scene, and the feature extraction module comprises a re-meshing module and a grid upper and lower sampling module which are sequentially connected.
11. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method for extracting features of a three-dimensional model based on a mesh subdivision structure according to any of claims 1 to 7 when executing the program.
12. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for three-dimensional model feature extraction based on mesh subdivision of any of claims 1 to 7.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110620148 | 2021-06-03 | ||
| CN2021106201485 | 2021-06-03 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113593033A true CN113593033A (en) | 2021-11-02 |
Family
ID=78243657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110649502.7A Pending CN113593033A (en) | 2021-06-03 | 2021-06-10 | Three-dimensional model feature extraction method based on grid subdivision structure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113593033A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115661407A (en) * | 2022-12-29 | 2023-01-31 | 如你所视(北京)科技有限公司 | Multi-detail-level model generation method and device, electronic equipment and storage medium |
| CN116778027A (en) * | 2023-08-22 | 2023-09-19 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
| CN117315194A (en) * | 2023-09-27 | 2023-12-29 | 南京航空航天大学 | Triangular mesh representation learning method for large aircraft appearance |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063719A (en) * | 2010-12-23 | 2011-05-18 | 西北工业大学 | Local matching method of three-dimensional model |
| US20130024545A1 (en) * | 2010-03-10 | 2013-01-24 | Tangentix Limited | Multimedia content delivery system |
| CN108830812A (en) * | 2018-06-12 | 2018-11-16 | 福建帝视信息科技有限公司 | A kind of high frame per second of video based on network deep learning remakes method |
| US20200012940A1 (en) * | 2017-03-17 | 2020-01-09 | Portland State University | Frame interpolation via adaptive convolution and adaptive separable convolution |
| CN111311751A (en) * | 2020-02-12 | 2020-06-19 | 叠境数字科技（上海）有限公司 | Three-dimensional clothes model reconstruction method based on deep neural network |
| CN112017196A (en) * | 2020-08-27 | 2020-12-01 | 重庆邮电大学 | Three-dimensional tooth model mesh segmentation method based on local attention mechanism |
| CN112200819A (en) * | 2020-10-16 | 2021-01-08 | 河海大学 | Human body three-dimensional skeleton model marking method based on data driving and template guiding |
-
2021
- 2021-06-10 CN CN202110649502.7A patent/CN113593033A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130024545A1 (en) * | 2010-03-10 | 2013-01-24 | Tangentix Limited | Multimedia content delivery system |
| CN102063719A (en) * | 2010-12-23 | 2011-05-18 | 西北工业大学 | Local matching method of three-dimensional model |
| US20200012940A1 (en) * | 2017-03-17 | 2020-01-09 | Portland State University | Frame interpolation via adaptive convolution and adaptive separable convolution |
| CN108830812A (en) * | 2018-06-12 | 2018-11-16 | 福建帝视信息科技有限公司 | A kind of high frame per second of video based on network deep learning remakes method |
| CN111311751A (en) * | 2020-02-12 | 2020-06-19 | 叠境数字科技（上海）有限公司 | Three-dimensional clothes model reconstruction method based on deep neural network |
| CN112017196A (en) * | 2020-08-27 | 2020-12-01 | 重庆邮电大学 | Three-dimensional tooth model mesh segmentation method based on local attention mechanism |
| CN112200819A (en) * | 2020-10-16 | 2021-01-08 | 河海大学 | Human body three-dimensional skeleton model marking method based on data driving and template guiding |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115661407A (en) * | 2022-12-29 | 2023-01-31 | 如你所视(北京)科技有限公司 | Multi-detail-level model generation method and device, electronic equipment and storage medium |
| CN116778027A (en) * | 2023-08-22 | 2023-09-19 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
| CN116778027B (en) * | 2023-08-22 | 2023-11-07 | 中国空气动力研究与发展中心计算空气动力研究所 | Curved surface parameterization method and device based on neural network |
| CN117315194A (en) * | 2023-09-27 | 2023-12-29 | 南京航空航天大学 | Triangular mesh representation learning method for large aircraft appearance |
| CN117315194B (en) * | 2023-09-27 | 2024-05-28 | 南京航空航天大学 | Triangular mesh representation learning method for large aircraft appearance |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113593033A (en) | Three-dimensional model feature extraction method based on grid subdivision structure | |
| CN116109798B (en) | Image data processing method, device, equipment and medium | |
| CN111028335B (en) | Point cloud data block surface patch reconstruction method based on deep learning | |
| CN112102480B (en) | Image data processing method, apparatus, device and medium | |
| US11600044B2 (en) | Rendering textures utilizing sharpness maps | |
| Denninger et al. | 3d scene reconstruction from a single viewport | |
| CN110368694A (en) | Data processing method, device, equipment and the readable storage medium storing program for executing of scene of game | |
| CN113313828B (en) | Three-dimensional reconstruction method and system based on single-picture intrinsic image decomposition | |
| CN115115805A (en) | Training method, device and equipment for three-dimensional reconstruction model and storage medium | |
| CN109447897A (en) | A kind of real scene image composition method and system | |
| CN115731365A (en) | Grid model reconstruction method, system, device and medium based on two-dimensional image | |
| CN116797768A (en) | Method and device for reducing reality of panoramic image | |
| CN108986210B (en) | Method and device for reconstructing three-dimensional scene | |
| CN114494395A (en) | Depth map generation method, device and equipment based on plane prior and storage medium | |
| KR20210058638A (en) | Apparatus and method for image processing | |
| CN115375847B (en) | Material recovery method, three-dimensional model generation method and model training method | |
| Davis et al. | 3d modeling of cities for virtual environments | |
| Baldacci et al. | GPU-based approaches for shape diameter function computation and its applications focused on skeleton extraction | |
| CN114742950A (en) | Ship shape 3D digital reconstruction method and device, storage medium and electronic equipment | |
| Zhang et al. | Fast Mesh Reconstruction from Single View Based on GCN and Topology Modification. | |
| CN116778065B (en) | Image processing method, device, computer and storage medium | |
| Udayan | An analysis of reconstruction algorithms applied to 3d building modeling | |
| CN115984583B (en) | Data processing method, apparatus, computer device, storage medium, and program product | |
| Ji et al. | REFRAME: Reflective Surface Real-Time Rendering for Mobile Devices | |
| Jiang et al. | PMPI: Patch-Based Multiplane Images for Real-Time Rendering of Neural Radiance Fields |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |