Method and apparatus for three-dimensional mesh model simplification with local and global feature preservation

Abstract

Embodiments of the present disclosure disclose a method and device for simplifying a three-dimensional mesh model with local and global feature preservation. A specific implementation of the method comprises: in response to detecting that the storage space of a target device exceeds a preset threshold, converting a three-dimensional mesh model into a triangular mesh list; performing initial feature extraction on the three-dimensional mesh model to obtain initial mesh features; performing local feature extraction on the initial mesh features to obtain mesh local features; performing global average pooling on the mesh local features; performing residual connection on the mesh local features and mesh global features to obtain an overall shape feature; performing saliency detection on each triangular mesh to obtain a set of mesh saliency scores; and simplifying the three-dimensional mesh model according to the set of mesh saliency scores to obtain a simplified three-dimensional mesh model for storage. The implementation can reduce the storage space of the three-dimensional mesh model while maintaining the shape representation accuracy of the simplified model.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the field of computer vision and virtual reality, and in particular, to a method and apparatus for simplifying a three-dimensional mesh model with local and global feature preservation. BACKGROUND

[0002] Complex three-dimensional models bring great challenges to the storage, processing, rendering and transmission of computers. The simplification of three-dimensional models is to appropriately reduce the number of vertex and face mesh of the model under the premise of not losing the visual features of the three-dimensional object, so as to reduce the complexity of the three-dimensional model. At present, when simplifying a three-dimensional mesh model, the commonly used methods are: geometric element folding method, taking the square of the distance from the vertex to the related triangular plane as the error, and determining the vertex coordinates to be deleted by minimizing the quadratic error; or based on mesh convolutional neural network, selectively deleting vertices through adaptive mesh pooling operation.

[0003] However, when simplifying a three-dimensional mesh model by using the above methods, the following technical problems often exist:

[0004] First, the geometric element folding method and the mesh convolutional neural network are difficult to balance the extraction of local details and global geometric information when capturing the local features of the three-dimensional mesh model, resulting in poor performance in extracting key features of complex three-dimensional models, and thus it is difficult to effectively reduce the storage space of the three-dimensional mesh model while maintaining the shape representation accuracy of the simplified model.

[0005] Second, when simplifying a three-dimensional mesh model, it is difficult to preserve important features while improving computational efficiency, which increases the consumption of computing resources.

[0006] The above information disclosed in this BACKGROUND section is only for the purpose of enhancing the understanding of the background of the present disclosure and therefore, it can contain information that does not form the prior art that is already known to those of ordinary skill in the art in the country. SUMMARY

[0007] The summary of the present disclosure is intended to introduce the concepts in a simplified form, which will be described in detail in the specific embodiments section. The summary of the present disclosure is not intended to identify key or essential features of the claimed technology nor is it intended to be used to limit the scope of the claimed technology.

[0008] Some embodiments of the present disclosure propose a method and apparatus for simplifying a three-dimensional mesh model with local and global feature preservation to solve one or more of the technical problems mentioned in the background section.

[0009] In a first aspect, some embodiments of the present disclosure provide a method for simplifying a three-dimensional mesh model with local and global feature preservation, the method comprising: in response to detecting that a storage space of a target device exceeds a preset threshold, obtaining a three-dimensional mesh model from the target device and converting the three-dimensional mesh model into a triangular facet mesh list, wherein the three-dimensional mesh model is a three-dimensional model composed of triangular facet meshes; performing initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list to obtain initial mesh features; performing local feature extraction on the initial mesh features to obtain mesh local features; performing global average pooling on the mesh local features to obtain mesh global features; performing residual connection on the mesh local features and the mesh global features to obtain overall shape features; performing saliency detection on each triangular facet mesh in the triangular facet mesh list according to the overall shape features to generate mesh saliency scores, thereby obtaining a set of mesh saliency scores; simplifying the three-dimensional mesh model according to each mesh saliency score in the set of mesh saliency scores to obtain a simplified three-dimensional mesh model, and sending the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed.

[0010] In a second aspect, some embodiments of the present disclosure provide an apparatus for simplifying a three-dimensional mesh model with local and global feature preservation, the apparatus comprising: an obtaining and converting unit configured to, in response to detecting that a storage space of a target device exceeds a preset threshold, obtain a three-dimensional mesh model from the target device and convert the three-dimensional mesh model into a triangular facet mesh list, wherein the three-dimensional mesh model is a three-dimensional model composed of triangular facet meshes; an initial feature extraction unit configured to perform initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list to obtain initial mesh features; a local feature extraction unit configured to perform local feature extraction on the initial mesh features to obtain mesh local features; a global average pooling unit configured to perform global average pooling on the mesh local features to obtain mesh global features; a residual connection unit configured to perform residual connection on the mesh local features and the mesh global features to obtain overall shape features; a saliency detection unit configured to perform saliency detection on each triangular facet mesh in the triangular facet mesh list according to the overall shape features to generate mesh saliency scores, thereby obtaining a set of mesh saliency scores; a simplifying unit configured to simplify the three-dimensional mesh model according to each mesh saliency score in the set of mesh saliency scores to obtain a simplified three-dimensional mesh model, and send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed.

[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device having stored thereon one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method described in any implementation manner of the first aspect.

[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having stored thereon a computer program, wherein the program, when executed by a processor, implements the method described in any implementation manner of the first aspect.

[0013] The above various embodiments of the present disclosure have the following beneficial effects: the local and global feature preserved three-dimensional mesh model simplification method of some embodiments of the present disclosure can effectively reduce the storage space of the three-dimensional mesh model while maintaining the shape representation accuracy of the simplified model. Specifically, the reason for the difficulty in balancing the model shape accuracy and the storage space is that the geometry element folding method and the mesh convolutional neural network are difficult to balance the extraction of local details and global geometric information when capturing the local features of the three-dimensional mesh model, resulting in poor performance in extracting the key features of complex three-dimensional models. Based on this, the local and global feature preserved three-dimensional mesh model simplification method of some embodiments of the present disclosure first, in response to detecting that the storage space of the target device exceeds a preset threshold, acquires a three-dimensional mesh model from the target device and converts the three-dimensional mesh model into a triangular facet mesh list. Wherein, the three-dimensional mesh model is a three-dimensional model composed of triangular facet meshes. Second, according to the triangular facet mesh list, the initial feature extraction is performed on the three-dimensional mesh model to obtain the initial mesh feature. Thus, it is helpful to maintain the global geometric structure of the three-dimensional mesh model. Then, the local feature extraction is performed on the initial mesh feature to obtain the mesh local feature. The local feature extraction enhances the ability to capture local geometric changes by focusing on the detailed part of the mesh. Thus, the details and complex features of the model can be preserved during the simplification process. Then, the global average pooling is performed on the mesh local feature to obtain the mesh global feature. Thus, the overall geometric information of the model can be extracted. Next, the residual connection is performed on the mesh local feature and the mesh global feature to obtain the overall shape feature. Thus, it can be ensured that the important local features can be preserved after the model is simplified, and the global geometric information will not be lost. After that, according to the overall shape feature, the saliency detection is performed on each triangular facet mesh in the triangular facet mesh list to generate a mesh saliency score and obtain a set of mesh saliency scores. Wherein, the mesh saliency score reflects the importance of each facet in the overall structure, which is used to guide the subsequent mesh simplification process. Finally, according to each mesh saliency score in the set of mesh saliency scores, the three-dimensional mesh model is simplified to obtain a simplified three-dimensional mesh model, and the simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed. Thus, the storage space of the three-dimensional mesh model can be effectively reduced while maintaining the shape representation accuracy of the simplified model. BRIEF DESCRIPTION OF DRAWINGS

[0014] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals can refer to the same or similar elements. It should be understood that the drawings are schematic, and elements and elements are not necessarily drawn to scale.

[0015] Figure 1 is a flow chart of some embodiments of the local and global feature preserving 3D mesh model simplification method according to the present disclosure;

[0016] Figure 2 is a framework chart of the local feature identification module of some embodiments of the local and global feature preserving 3D mesh model simplification method according to the present disclosure;

[0017] Figure 3 is an example chart of the vertex merging step of some embodiments of the local and global feature preserving 3D mesh model simplification method according to the present disclosure;

[0018] Figure 4 is an effect chart of the simplified 3D mesh model of some embodiments of the local and global feature preserving 3D mesh model simplification method according to the present disclosure;

[0019] Figure 5 is a structural schematic diagram of some embodiments of the local and global feature preserving 3D mesh model simplification apparatus according to the present disclosure;

[0020] Figure 6 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0021] Embodiments of the present disclosure will be described below in greater detail with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are merely for illustrative purposes and are not intended to limit the scope of protection of the present disclosure.

[0022] It should also be noted that, for the sake of brevity, only the parts of the drawings that are relevant to the present disclosure are shown. The embodiments in the present disclosure and the features in the embodiments can be combined with each other in the case of no conflict.

[0023] It should be noted that the terms “first”, “second”, and the like mentioned in the present disclosure are merely used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0024] It should be noted that the terms “one”, “multiple” mentioned in the present disclosure are illustrative and not restrictive, and those skilled in the art should understand that, unless otherwise explicitly stated in the context, it should be understood as “one or more”.

[0025] Names of messages or information exchanged between multiple devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0026] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0027] Figure 1 A flow 100 of some embodiments of the local and global feature preserving 3D mesh model simplification method according to the present disclosure is shown. The local and global feature preserving 3D mesh model simplification method includes the following steps:

[0028] Step 101, in response to detecting that the storage space of a target device exceeds a preset threshold, obtaining a 3D mesh model from the target device and converting the 3D mesh model into a triangular facet mesh list.

[0029] In some embodiments, the execution subject of the local and global feature preserving 3D mesh model simplification method can obtain a 3D mesh model from a target device and convert the 3D mesh model into a triangular facet mesh list in response to detecting that the storage space of the target device exceeds a preset threshold. Wherein, the above-mentioned target device can be a local device or a remote device. The above-mentioned 3D mesh model is a three-dimensional model composed of a mesh in the form of a triangular facet.

[0030] In some optional implementations of some embodiments, the execution subject of the local and global feature preserving 3D mesh model simplification method, in response to detecting that the storage space of a target device exceeds a preset threshold, obtaining a 3D mesh model from the target device and converting the 3D mesh model into a triangular facet mesh list, can include the following steps:

[0031] First, in response to detecting that the storage space of a target device exceeds a preset threshold, obtaining a 3D mesh model from the target device. Wherein, the above-mentioned 3D mesh model can come from a public dataset. Wherein, the above-mentioned public dataset can be a Manifold dataset.

[0032] Second, each triangular facet in the above-mentioned 3D mesh model is taken as a triangular facet mesh in the triangular facet mesh list to obtain a triangular facet mesh list. Wherein, an index number is assigned according to the reading order of each triangular facet in the above-mentioned 3D mesh model, and the coordinate information of three vertices is recorded as the features of the triangular facet mesh to generate the triangular facet mesh list. Here, the above-mentioned triangular facet mesh is a planar area determined by the three vertices of a triangular facet.

[0033] In the third step, the coordinates of the center point of each triangular facet mesh are determined according to the coordinates of the three vertices of each triangular facet mesh in the triangular facet mesh list, and a set of coordinates of the center point of the mesh is obtained. The average coordinates of the three vertices of the triangular facet mesh can be determined first. Then, the average coordinates can be used as the coordinates of the center point of the mesh.

[0034] In practice, each coordinate of the center point of the mesh in the set of coordinates of the center point of the mesh can be represented as a 3-dimensional vector.

[0035] In the fourth step, three mesh angle vectors of each triangular facet mesh are determined according to the coordinates of the three vertices and the coordinates of the center point of each triangular facet mesh in the triangular facet mesh list, and a list of mesh angle vectors is generated to obtain a set of lists of mesh angle vectors.

[0036] In practice, each list of mesh angle vectors in the set of lists of mesh angle vectors can be represented as a 9-dimensional vector.

[0037] In the fifth step, a cross product operation is performed on each list of mesh angle vectors in the set of lists of mesh angle vectors to generate a mesh normal vector, and a set of mesh normal vectors is obtained. The cross product operation can be performed on any two angle vectors in the list of mesh angle vectors to generate an initial mesh normal vector. Then, the initial mesh normal vector can be normalized to obtain the mesh normal vector. The mesh normal vector is the unit normal vector of the plane corresponding to the triangular facet mesh.

[0038] In practice, each mesh normal vector in the set of mesh normal vectors can be represented as a 3-dimensional vector.

[0039] In the sixth step, the adjacent mesh index number list is generated by indexing the adjacent mesh of each triangular facet mesh in the triangular facet mesh list to obtain a set of adjacent mesh index number lists. The triangular facet mesh in the triangular facet mesh list that has the same two vertices as the current triangular facet mesh is used as the adjacent mesh of the current triangular facet mesh, and the index number corresponding to the adjacent mesh is used as the adjacent mesh index number to generate the adjacent mesh index number list. The adjacent mesh index number list can contain three adjacent mesh index numbers.

[0040] In practice, each adjacent mesh index number list in the set of adjacent mesh index number lists can be represented as a 3-dimensional vector.

[0041] In step 102, the initial feature extraction of the three-dimensional mesh model is performed according to the triangular facet mesh list, and the initial mesh feature is obtained.

[0042] In some embodiments, the execution subject can perform initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list, and obtain the initial mesh feature.

[0043] In some optional implementations of some embodiments, the execution subject performs initial feature extraction on the three-dimensional mesh model according to the triangular mesh list to obtain initial mesh features, which can include the following steps:

[0044] Firstly, spatial features of the mesh center point coordinate set are extracted to obtain mesh spatial features. Specifically, each mesh center point coordinate in the mesh center point coordinate set can be concatenated to obtain a mesh center point tensor. Then, the mesh center point tensor can be nonlinearly mapped by a preset multi-layer perception to obtain the mesh spatial features. The mesh spatial features are used to represent the spatial features of each triangular mesh in the triangular mesh list.

[0045] In practice, the mesh center point tensor is an N x 3 tensor. N is the number of triangular meshes in the triangular mesh list. The multi-layer perception can include two fully connected layers, each of which is connected to a batch normalization layer and a ReLU activation function. The fully connected layer can be a 64-dimensional fully connected layer. The mesh spatial features can be N x 64-dimensional features.

[0046] Secondly, internal structure features of the mesh angle vector list set are extracted to obtain the internal structure features. Specifically, any two mesh angle vectors in the mesh angle vector list can be concatenated to generate a mesh angle vector feature vector to obtain a mesh angle vector feature vector set. The mesh angle vector feature vector set can include three mesh angle vector feature vectors. One mesh angle vector list corresponds to three mesh angle vector feature vectors. Then, the mesh angle vector feature vector sets corresponding to each mesh angle vector list in the mesh angle vector list set are concatenated to obtain a mesh angle vector tensor. The mesh angle vector tensor can be convolved by a preset convolution layer to obtain mesh angle vector convolution features. The mesh angle vector convolution features can be nonlinearly mapped by the multi-layer perception to obtain initial internal structure features. Finally, the initial internal structure features are mean-pooled to obtain the internal structure features.

[0047] In practice, the mesh angle vector feature vector can be a 6-dimensional vector. The mesh angle vector tensor can be an N x (3 x 6) tensor. The convolution kernel size of the convolution layer can be 2, and the number of convolution kernels can be 32. The mesh angle vector convolution features and the initial internal structure features can be N x (3 x 64). The internal structure features can be N x 64.

[0048] In the third step, the external structure feature is obtained by performing external structure feature extraction on the set of mesh normal vectors and the set of lists of adjacent mesh index numbers. Firstly, each mesh normal vector in the set of mesh normal vectors can be spliced to obtain a mesh normal vector tensor. Secondly, each list of adjacent mesh index numbers in the set of lists of adjacent mesh index numbers can be spliced to obtain an adjacent mesh index number tensor. Then, the mesh normal vector tensor and the adjacent mesh index number tensor can be convolved by a preset second convolution layer to obtain an initial external structure feature. Finally, the initial external structure feature can be nonlinearly mapped by the multi-layer perceptron to obtain the external structure feature.

[0049] In practice, the mesh normal vector tensor can be N x 3 dimensional. The adjacent mesh index number tensor can be N x 3 dimensional. The kernel size of the second convolution layer can be 3, and the number of kernels can be 32. The initial external structure feature can be N x 32 dimensional. The external structure feature can be N x 64 dimensional.

[0050] In the fourth step, the mesh structure feature is obtained by splicing the internal structure feature and the external structure feature. Firstly, the initial mesh structure feature can be obtained by splicing the internal structure feature and the external structure feature. Secondly, the initial mesh structure feature can be nonlinearly mapped by a preset second multi-layer perceptron to obtain the mesh structure feature.

[0051] In practice, the initial mesh structure feature can be N x 128 dimensional. The second multi-layer perceptron can include two fully connected layers of 131 dimensions, and each fully connected layer is followed by a batch normalization layer and a ReLU activation function. The mesh structure feature can be N x 131 dimensional.

[0052] In the fifth step, the initial mesh feature is obtained by performing feature fusion on the mesh space feature and the mesh structure feature. Firstly, the mesh space feature and the mesh structure feature can be spliced to obtain a spliced mesh feature. Then, the spliced mesh feature can be nonlinearly mapped by a preset third multi-layer perceptron to obtain a first initial mesh feature. Secondly, the adjacent mesh index number tensor and the mesh structure feature can be spliced to obtain a spliced mesh structure feature. Then, the spliced mesh structure feature can be nonlinearly mapped by the third multi-layer perceptron after being subjected to maximum pooling and average pooling operations to obtain a second initial mesh feature. Thirdly, the first initial mesh feature and the second initial mesh feature can be spliced to obtain a spliced initial mesh feature. Finally, the spliced initial mesh feature can be nonlinearly mapped by the third multi-layer perceptron to obtain the initial mesh feature.

[0053] In practice, the third multi-layer perceptron can include a 256-dimensional fully connected layer, and the fully connected layer is connected with a batch normalization layer and a ReLU activation function. The first initial mesh feature and the second initial mesh feature can both be N x 256-dimensional. The initial mesh feature can be N x 256-dimensional.

[0054] At step 103, local feature extraction is performed on the initial mesh feature to obtain a mesh local feature.

[0055] In some embodiments, the execution subject can perform local feature extraction on the initial mesh feature to obtain a mesh local feature. The local feature extraction on the initial mesh feature can be performed by a preset local feature identification module to obtain the mesh local feature. The local feature identification module can be as shown in FIG. 3. Figure 2

[0056] In some optional implementations of some embodiments, the execution subject performs local feature extraction on the initial mesh feature to obtain a mesh local feature, which can include the following steps:

[0057] Firstly, feature division is performed on the initial mesh feature to generate a sub-initial mesh feature, thereby obtaining a set of sub-initial mesh features. The feature division can be performed on the initial mesh feature along the feature dimension to generate the sub-initial mesh feature, thereby obtaining the set of sub-initial mesh features.

[0058] For example, the initial mesh feature F initial ∈R N×d , and d represents the number of feature channels. The initial mesh feature can be divided into m sub-initial mesh features. wherein m represents the number of sub-initial mesh features. i is between 0 and (m-1). is the i-th sub-initial mesh feature in the set of sub-initial mesh features, Each of the sub-initial mesh features contains features of different triangular mesh.

[0059] Secondly, attention feature extraction is performed on each of the set of sub-initial mesh features to generate a weighted attention feature, thereby obtaining a set of weighted attention features. The attention feature extraction on each of the set of sub-initial mesh features can be performed by a preset attention head to generate the weighted attention feature, thereby obtaining the set of weighted attention features. The attention head can be a Transformer self-attention mechanism. One attention head corresponds to one sub-initial mesh feature.

[0060] ​Optionally, the execution subject performs attention feature extraction on each of the sub-initial mesh features in the sub-initial mesh feature set to generate weighted attention features, to obtain a weighted attention feature set, which can include the following steps:

[0061] In a first sub-step, linear transformation is performed on each of the sub-initial mesh features in the sub-initial mesh feature set to generate transformed sub-initial features, to obtain a transformed sub-initial feature set. The linear transformation on the sub-initial mesh features to generate the transformed sub-initial features can be performed according to the following formula.

[0062]

[0063] wherein, represents the i-th transformed sub-initial feature in the transformed sub-initial feature set. W i is a weight matrix of the linear transformation. b i is a bias term.

[0064] In a second sub-step, a query vector, a key vector and a value vector are determined for each of the transformed sub-initial features in the transformed sub-initial feature set, to obtain a query vector set, a key vector set and a value vector set. The query vector, the key vector and the value vector for each of the transformed sub-initial features in the transformed sub-initial feature set can be determined according to the following formula.

[0065]

[0066] wherein, Q i represents the query vector. K i represents the key vector. V i represents the value vector. is a weight matrix of each attention head.

[0067] In a third sub-step, an attention weight is determined according to the dot product between each of the query vectors in the query vector set and each of the key vectors in the key vector set, to obtain an attention weight set. The corresponding attention weight can be determined according to the following formula.

[0068]

[0069] wherein, Attention i represents the attention weight. softmax() represents a Softmax function.

[0070] is the dimension of the key vector.

[0071] Fourth sub-step, dot product processing is performed between each attention weight in the above attention weight set and each value vector in the above value vector set to generate a weighted attention feature, obtaining a weighted attention feature set. Wherein, the weighted attention feature can be generated by the following formula.

[0072] O i =Attention i ·V i .

[0073] Wherein, O i represents the i-th weighted attention feature in the weighted attention feature set.

[0074] Third step, each weighted attention feature in the above weighted attention feature set is spliced to obtain a weighted overall feature. As shown in C (Concat, Splice) in Figure 2 .

[0075] Fourth step, element-wise addition is performed between the above weighted overall feature and the initial grid feature to obtain a grid local feature. Wherein, the above weighted overall feature can be mapped by a linear layer (Linear Layer) to obtain a mapped weighted overall feature with the same dimension as the initial grid feature. Then, element-wise addition can be performed between the above mapped weighted overall feature and the initial grid feature to obtain the grid local feature.

[0076] In practice, the above step 103 divides the initial grid feature into a plurality of sub-features along the feature dimension, and then processes each sub-feature using a multi-head attention mechanism. In this way, key local features in the grid model can be effectively captured, core areas can be highlighted, and the overall efficiency of three-dimensional grid model simplification can be improved.

[0077] Step 104, global average pooling is performed on the grid local feature to obtain a grid global feature.

[0078] In some embodiments, the above execution subject can perform global average pooling on the above grid local feature to obtain a grid global feature. Wherein, the above grid local feature can be globally averaged pooled to obtain a grid global feature by the following formula.

[0079]

[0080] Wherein, represents the value of the j-th triangular mesh in the grid global feature. d local represents the number of feature channels of the grid local feature. represents the value of the c-th feature channel of the j-th triangular mesh in the grid local feature F local . globalrepresenting the global feature of the mesh.

[0081] In practice, the global feature of the mesh can reflect the geometry of the entire three-dimensional mesh model.

[0082] Step 105, residual connection is performed on the local feature of the mesh and the global feature of the mesh to obtain an overall shape feature.

[0083] In some embodiments, the execution subject can perform residual connection on the local feature of the mesh and the global feature of the mesh to obtain an overall shape feature. As shown in the following formula:

[0084] F entire = F local + λ · F global .

[0085] Wherein, F entire represents the overall shape feature. λ represents a learnable scale factor.

[0086] In practice, the learnable scale factor λ is set to dynamically balance the weights of local features and global features, ensuring that the fused features can not only retain local details but also reflect overall shape structures. The overall shape feature effectively fuses local details and global information, enhancing the description ability of complex three-dimensional shapes.

[0087] Step 106, according to the overall shape feature, performing saliency detection on each triangular mesh in the triangular mesh list to generate a mesh saliency score, obtaining a mesh saliency score set.

[0088] In some embodiments, the execution subject can perform saliency detection on each triangular mesh in the triangular mesh list according to the overall shape feature to generate a mesh saliency score, obtaining a mesh saliency score set.

[0089] In some optional implementations of some embodiments, the execution subject performs saliency detection on each triangular mesh in the triangular mesh list according to the overall shape feature to generate a mesh saliency score, obtaining a mesh saliency score set, which can include the following steps:

[0090] First, the overall shape feature is divided into mesh dimensions to generate mesh shape features, obtaining a mesh shape feature set. Wherein, the mesh shape feature can be a vector composed of a row of numerical values in the overall shape feature.

[0091] For example, F entire = {f1, f2, f i ..., f N}, Wherein, f ia mesh shape feature in the mesh shape feature set, d e a feature channel number representing the overall shape feature.

[0092] In the second step, according to the above-mentioned adjacent mesh index number list set, the neighborhood mesh feature difference matrix between each mesh shape feature in the above-mentioned mesh shape feature set and each adjacent mesh shape feature is determined, and a neighborhood mesh feature difference matrix set is obtained. The neighborhood mesh feature difference matrix can be determined by the following formula.

[0093]

[0094] where ΔF i represents the neighborhood mesh feature difference matrix of the triangular facet mesh i, P is the neighborhood size of the triangular facet mesh i, and p takes a value between 1 and 3. x pc is the value in the pth row and cth column of the neighborhood mesh feature difference matrix. f i (c) represents the cth eigenvalue of f i . represents the cth eigenvalue of the mesh shape feature of the pth adjacent mesh of the triangular facet mesh i.

[0095] In the third step, each neighborhood mesh feature difference matrix in the above-mentioned neighborhood mesh feature difference matrix set is spliced to obtain a neighborhood feature difference matrix. The neighborhood feature difference matrix can be ΔF=[ΔF1, ΔF2, ΔF i ,..., ΔF N ], ΔF∈R N×P×d .

[0096] In the fourth step, the above-mentioned neighborhood feature difference matrix is subjected to significant attention extraction to obtain a significant attention matrix. The significant attention matrix can be obtained by significant attention extraction of the above-mentioned neighborhood feature difference matrix by the following formula.

[0097]

[0098] where X represents the significant attention matrix, X∈R N×d . Atten() is a function for obtaining attention distribution from the neighborhood feature difference matrix. W I2 and W I1 are 1×1 convolution kernels, and the output dimension of ReLU(W I2 (ReLU(W i1 ΔF)) can be N×P×d. The output is summed in P dimensions to obtain a channel attention map A channel ∈R N×d; the output is spatial attention map A through average pooling operation in P dimensions spatial ∈R N×d . Wherein, X=A channel +A spatial .

[0099] Step 5, the maximum significant score of the above significant attention matrix is processed to obtain the initial saliency score vector. Wherein, the maximum value of each row in the above significant attention matrix can be taken as the numerical value of the corresponding row of the initial saliency score vector to obtain the initial saliency score vector. Wherein, the dimension of the initial saliency score vector is N.

[0100] Step 6, according to the initial saliency score vector, determine the initial mesh saliency score of each triangular mesh in the triangular mesh list, to obtain the initial mesh saliency score set. Wherein, each element in the initial saliency score vector corresponds to each triangular mesh in the triangular mesh list one by one.

[0101] Step 7, the initial mesh saliency score in the initial mesh saliency score set is inversely interpolated to generate the mesh saliency score, to obtain the mesh saliency score set. Wherein, the initial mesh saliency score can be inversely interpolated to generate the mesh saliency score by the following formula.

[0102]

[0103] Wherein, S represents the mesh saliency score of triangular mesh i at the qth layer of the network. q (i) represents the initial mesh saliency score of triangular mesh i at the qth layer of the network. ip D represents the Euclidean distance between the center point of triangular mesh i and its adjacent mesh p. q represents the number of layers of the network.

[0104] Wherein, the network can be a three-dimensional mesh model simplification network. The three-dimensional mesh model simplification network can execute the above steps 101 to 105. For example, step 101 can correspond to the input layer, which is the first layer of the network, and step 102 can correspond to the initial mesh feature extraction layer, which is the second layer of the network.

[0105] In practice, the above significant attention extraction step can be trained in conjunction with the shape classification task. Specifically, the overall shape feature can be element-wise multiplied with the channel attention map and the spatial attention map in sequence to obtain a weighted overall shape feature. Then, the high-dimensional weighted overall shape feature is mapped to a classification space through a fully connected layer. Next, the output is converted to a class probability using a Softmax function. Subsequently, the loss value between the class probability and the true label is determined using a cross-entropy loss function. The loss value can be used to guide the backpropagation to update the parameters and improve the accuracy of shape classification. The parameters can include the parameters of each convolution kernel, the parameters of the fully connected layer, and the weight matrix W i 、 bias term b i a learnable scale factor λ.

[0106] During the co-training process, the optimization of the saliency score is performed together with the shape classification task. Through the backpropagation of the saliency score, the network simultaneously updates the feature extraction module for shape classification and the calculation module of the saliency score. This co-training method ensures that the network not only learns how to classify the shape of the mesh, but also identifies the triangular patches that have a key impact on the overall structure during the training process, improving the accuracy of saliency detection.

[0107] At step 107, according to each mesh saliency score in the mesh saliency score set, the three-dimensional mesh model is simplified to obtain a simplified three-dimensional mesh model, and the simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and stored and displayed.

[0108] In some embodiments, the execution subject can simplify the three-dimensional mesh model according to each mesh saliency score in the mesh saliency score set to obtain a simplified three-dimensional mesh model, and send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display.

[0109] In some optional implementations of some embodiments, the execution subject simplifies the three-dimensional mesh model according to each mesh saliency score in the mesh saliency score set to obtain a simplified three-dimensional mesh model, and sends the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display, which can include the following steps:

[0110] In a first step, each triangular mesh in the triangular mesh list is regionally partitioned according to the mesh saliency score set to generate a saliency mesh partition, and a saliency mesh partition set is obtained. The saliency mesh partition includes a plurality of adjacent triangular meshes sharing the same edge. First, each triangular mesh in the triangular mesh list can be regionally partitioned in a low-to-high order of the mesh saliency score to generate a saliency mesh partition, and a saliency mesh partition set is obtained.

[0111] In a second step, each saliency mesh partition in the saliency mesh partition set is partition simplified according to a preset order, and a simplified mesh partition set is obtained. The saliency mesh partition in the saliency mesh partition set can be partition simplified in a low-to-high order of the mesh saliency score, and a simplified mesh partition set is obtained.

[0112] Optionally, the execution subject can perform the following partition simplification step on each saliency mesh partition in the saliency mesh partition set according to a preset order to generate a simplified mesh partition, and a simplified mesh partition set is obtained:

[0113] In a first sub-step, each mesh vertex and each mesh edge in the saliency mesh partition is determined according to the vertex coordinates of each triangular mesh in the saliency mesh partition, and a mesh vertex set and a mesh edge set are obtained. The mesh edge is a line segment composed of two mesh vertices in the mesh vertex set.

[0114] In a second sub-step, the mesh edge priority of each mesh edge in the mesh edge set is determined according to the mesh saliency scores of the two triangular meshes adjacent to the mesh edge, and a mesh edge priority list is obtained. The mesh edge saliency score of each mesh edge can be determined by the following formula.

[0115]

[0116] wherein, e ij represents the mesh edge saliency score of the mesh edge between the triangular mesh i and the triangular mesh j. Then, the mesh edge priority of each mesh edge in the mesh edge set is determined in a low-to-high order of the mesh edge saliency score, and a mesh edge priority list is obtained. The mesh edge with a low mesh edge saliency score corresponds to a high mesh edge priority.

[0117] In a third sub-step, each mesh vertex in the mesh vertex set is vertex-merged according to the mesh edge priority list and a preset geometric error threshold, and a merged vertex set is obtained.

[0118] Firstly, the two mesh vertices corresponding to the mesh edge with the highest mesh edge priority are merged according to the mesh edge priority from high to low, to obtain a merged vertex.

[0119] Here, the merged vertex is determined by the following formula, as shown in Figure 3

[0120]

[0121] wherein, V new represents the merged vertex. V1 and V2 are the coordinates of the two mesh vertices corresponding to the mesh edge.

[0122] Then, each mesh vertex in the mesh vertex set and the merged vertex are taken as an initial merged vertex set, and the maximum distance error between the initial merged vertex set and the mesh vertex set is determined by the following formula.

[0123]

[0124] wherein, E represents the maximum distance error. A represents the initial merged vertex set. B represents the mesh vertex set. a and b are any point in the point set A and B respectively. D(a, b) is the Euclidean distance between two points.

[0125] Next, if the maximum distance error is less than a preset geometric error threshold, the vertex merging step is continued. If the maximum distance error is greater than or equal to the preset geometric error threshold, the obtained initial merged vertex set is determined as the merged vertex set. The geometric error threshold can be used to control the maximum distance difference range between the significance mesh partition and the simplified mesh partition, to ensure that the maximum distance error allowed in the vertex merging process does not exceed the preset tolerance. Here, each significance mesh partition in the significance mesh partition set corresponds to a different geometric error threshold.

[0126] Fourth sub-step, determine the merged edge corresponding to each two merged vertices in the merged vertex set, to obtain a merged edge set.

[0127] Fifth sub-step, determine the merged triangle patch composed of each three merged edges in the merged edge set, to obtain a merged triangle patch set.

[0128] Sixth sub-step, determine the simplified mesh partition corresponding to the merged triangle patch set.

[0129] ​In the third step, a global geometric error between the set of significant mesh partitions and the set of simplified mesh partitions is determined. The global geometric error between the set of significant mesh partitions and the set of simplified mesh partitions can be determined according to the coordinates of all mesh vertices in the set of significant mesh partitions and the coordinates of all merged vertices in the set of simplified mesh partitions by using the maximum distance error determination formula. Details are not repeated here.

[0130] In the fourth step, in response to the global geometric error being less than or equal to a preset global geometric error threshold, each simplified mesh partition in the set of simplified mesh partitions is merged to generate a simplified three-dimensional mesh model. The global geometric error threshold can be used to control the maximum distance difference range between the original three-dimensional mesh model and the simplified three-dimensional mesh model, and to ensure that the maximum distance error allowed in the simplification process does not exceed the preset tolerance. If the global geometric error is greater than the preset global geometric error threshold, each geometric error threshold is adjusted to be lower, and the second step to the third step is re-executed. If the global geometric error is less than or equal to the preset global geometric error threshold, each simplified mesh partition in the set of simplified mesh partitions is merged to generate a simplified three-dimensional mesh model. The effect diagram of the simplified three-dimensional mesh model is shown in FIG. 6. Figure 4

[0131] In the fifth step, the simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed. The simplified three-dimensional mesh model can be sent to the target device by wired or wireless transmission to replace the three-dimensional mesh model before simplification and to be stored and displayed.

[0132] ​The steps 106-107 and the related content thereof are an inventive point of the embodiment of the present disclosure, and solve the technical problem of "increased consumption of computing resources" mentioned in the background. The factors that lead to excessive consumption of computing resources are often as follows: when simplifying a three-dimensional mesh model, it is difficult to improve computational efficiency while preserving important features. If the above factors are addressed, the situation of excessive consumption of computing resources can be reduced. To achieve this effect, first, the saliency of the three-dimensional mesh model is characterized by the feature difference between each triangular mesh and its adjacent mesh. Thus, the saliency of each triangular mesh is represented as a quantifiable neighborhood feature difference matrix, facilitating subsequent simplification operations. Second, through co-training, the shape classification and saliency scoring of the three-dimensional mesh model can be performed simultaneously, improving the accuracy of saliency scoring and reducing the consumption of computing resources. Among them, the saliency score reflects the contribution of each triangular facet in the overall structure, and is used to guide the priority ordering of subsequent mesh simplification. Thus, the core feature area can be highlighted, the preservation effect of the salient area can be improved, and a more accurate basis for mesh simplification can be provided, optimizing the geometric quality and computational efficiency of the overall model during the simplification process. At the same time, by pre-setting the geometric error threshold and the global geometric error threshold, the deletion of meshes with high saliency during the simplification process is avoided, and important features of the simplified model are preserved. Thus, the computational efficiency is improved, while ensuring that the simplified three-dimensional model still has a high-precision shape representation, suitable for complex three-dimensional data processing scenarios.

[0133] The above various embodiments of the present disclosure have the following beneficial effects: the local and global feature preserved three-dimensional mesh model simplification method of some embodiments of the present disclosure can effectively reduce the storage space of the three-dimensional mesh model while maintaining the shape representation accuracy of the simplified model. Specifically, the reason for the difficulty in balancing the model shape accuracy and storage space is that the geometry folding method and the mesh convolutional neural network are difficult to balance the extraction of local details and global geometric information when capturing the local features of the three-dimensional mesh model, resulting in poor performance in extracting key features of complex three-dimensional models. Based on this, the local and global feature preserved three-dimensional mesh model simplification method of some embodiments of the present disclosure first, in response to detecting that the storage space of the target device exceeds the preset threshold, acquires a three-dimensional mesh model from the target device and converts the three-dimensional mesh model into a triangular facet mesh list. Wherein, the three-dimensional mesh model is a three-dimensional model composed of triangular facet meshes. Second, according to the triangular facet mesh list, the initial feature extraction is performed on the three-dimensional mesh model to obtain the initial mesh feature. Thus, it is helpful to maintain the global geometric structure of the three-dimensional mesh model. Then, the local feature extraction is performed on the initial mesh feature to obtain the mesh local feature. The local feature extraction enhances the ability to capture local geometric changes by focusing on the detailed part of the mesh. Thus, the details and complex features of the model can be preserved during the simplification process. Then, the global average pooling is performed on the mesh local feature to obtain the mesh global feature. Thus, the overall geometric information of the model can be extracted. Next, the residual connection is performed on the mesh local feature and the mesh global feature to obtain the overall shape feature. Thus, it can be ensured that important local features can be preserved after model simplification, and global geometric information will not be lost. Then, according to the overall shape feature, the saliency detection is performed on each triangular facet mesh in the triangular facet mesh list to generate a mesh saliency score, and a set of mesh saliency scores is obtained. Wherein, the mesh saliency score reflects the importance of each facet in the overall structure, which is used to guide the subsequent mesh simplification process. Finally, according to each mesh saliency score in the set of mesh saliency scores, the three-dimensional mesh model is simplified to obtain a simplified three-dimensional mesh model, and the simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed. Thus, the storage space of the three-dimensional mesh model can be effectively reduced while maintaining the shape representation accuracy of the simplified model.

[0134] Further reference Figure 5 As an implementation of the method shown in the above figures, the present disclosure provides some embodiments of a local and global feature preserved three-dimensional mesh model simplification device, which device embodiments correspond to the method embodiments shown in Figure 1 The device can be applied in various electronic devices.

[0135] As shown in Figure 5 the three-dimensional mesh model simplification device 500 of some embodiments includes: an acquisition and conversion unit 501, an initial feature extraction unit 502, a local feature extraction unit 503, a global average pooling unit 504, a residual connection unit 505, a saliency detection unit 506, a simplification unit 507. Among them, the acquisition and conversion unit 501 is configured to acquire a three-dimensional mesh model from a target device in response to detecting that the storage space of the target device exceeds a preset threshold, and convert the three-dimensional mesh model into a triangular facet mesh list, wherein the three-dimensional mesh model is a three-dimensional model composed of a mesh in the form of a triangular facet; the initial feature extraction unit 502 is configured to perform initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list to obtain initial mesh features; the local feature extraction unit 503 is configured to perform local feature extraction on the initial mesh features to obtain mesh local features; the global average pooling unit 504 is configured to perform global average pooling on the mesh local features to obtain mesh global features; the residual connection unit 505 is configured to perform residual connection on the mesh local features and the mesh global features to obtain overall shape features; the saliency detection unit 506 is configured to perform saliency detection on each triangular facet mesh in the triangular facet mesh list according to the overall shape features to generate mesh saliency scores, thereby obtaining a set of mesh saliency scores; the simplification unit 507 is configured to simplify the three-dimensional mesh model according to each mesh saliency score in the set of mesh saliency scores to obtain a simplified three-dimensional mesh model, and send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display.

[0136] It can be understood that the units described in the device 500 correspond to the respective steps in the method described with reference to Figure 1 Therefore, the operations, features and beneficial effects described above for the method also apply to the device 500 and the units contained therein, which will not be described here.

[0137] Reference is made below to Figure 6 which shows a structural schematic diagram of an electronic device (e.g., a computing device) 600 suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the function and scope of use of embodiments of the present disclosure.

[0138] As shown in Figure 6As shown, the electronic device 600 can include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601 that can perform various appropriate actions and processes according to programs stored in a read-only memory 602 or loaded from a storage device 608 into a random access memory 603. Various programs and data required for the operation of the electronic device 600 are also stored in the random access memory 603. The processing device 601, the read-only memory 602, and the random access memory 603 are connected to each other through a bus 604. An input / output interface 605 is also connected to the bus 604.

[0139] Generally, the following devices can be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 609. The communication devices 609 can allow the electronic device 600 to communicate wirelessly or wired with other devices to exchange data. Although Figure 6 The electronic device 600 is shown with various devices, but it should be understood that all of the illustrated devices are not required, and more or fewer devices can alternatively be implemented. Figure 6 Each block shown in the flowcharts can represent a device, or a plurality of devices, as needed.

[0140] In particular, processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product including a computer program carried on a computer readable medium, the computer program containing program codes for performing the methods shown in the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network through the communication devices 609, or installed from the storage devices 608, or installed from the read-only memory 602. When the computer program is executed by the processing device 601, the above-described functions defined in the methods of some embodiments of the present disclosure are performed.

[0141] Note that the computer readable medium in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program used by an instruction execution system, apparatus or device, or that can be used by or in connection with an instruction execution system, apparatus or device. In some embodiments of the present disclosure, the computer readable signal medium can include a computer readable program code propagated in or on a carrier medium, in which the computer readable program code is embodied. Such propagated computer readable program code can take many forms, including but not limited to, an electromagnetic signal, an optical signal or any suitable combination of the foregoing. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus or device. Program code embodied on a computer readable medium can be transmitted using any suitable medium, including but not limited to, wire, cable, wireless, RF, infrared or any suitable combination of the foregoing.

[0142] In some embodiments, the client, server, or both can communicate using any current known or future developed network protocol, such as HTTP (Hyper Text Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current known or future developed networks.

[0143] The computer readable medium can be included in the electronic device, or can exist separately from the electronic device. The computer readable medium bears one or more programs, which, when executed by the electronic device, cause the electronic device to: in response to detecting that the storage space of the target device exceeds a preset threshold, acquire a three-dimensional mesh model from the target device, and convert the three-dimensional mesh model into a triangular facet mesh list, where the three-dimensional mesh model is a three-dimensional model composed of triangular facet meshes; perform initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list to obtain initial mesh features; perform local feature extraction on the initial mesh features to obtain mesh local features; perform global average pooling on the mesh local features to obtain mesh global features; perform residual connection on the mesh local features and the mesh global features to obtain an overall shape feature; perform saliency detection on each triangular facet mesh in the triangular facet mesh list according to the overall shape feature to generate a mesh saliency score, to obtain a set of mesh saliency scores; simplify the three-dimensional mesh model according to each mesh saliency score in the set of mesh saliency scores to obtain a simplified three-dimensional mesh model, and send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to store and display.

[0144] Computer program code for carrying out operations of some embodiments of the disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0146] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition and transformation unit, an initial feature extraction unit, a local feature extraction unit, a global average pooling unit, a residual connection unit, a saliency detection unit, and a simplification unit. The names of these units do not necessarily limit the unit itself; for example, a simplification unit may also be described as "a unit for simplifying a 3D mesh model."

[0147] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0148] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for simplifying a three-dimensional mesh model with local and global feature preservation, comprising: in response to detecting that a storage space of a target device exceeds a preset threshold, obtaining a three-dimensional mesh model from the target device and converting the three-dimensional mesh model into a triangular facet mesh list, wherein the three-dimensional mesh model is a three-dimensional model composed of a mesh in the form of a triangular facet; performing initial feature extraction on the three-dimensional mesh model according to the triangular facet mesh list to obtain initial mesh features; performing local feature extraction on the initial mesh features to obtain mesh local features; performing global average pooling on the mesh local features to obtain mesh global features; performing residual connection on the mesh local features and the mesh global features to obtain an overall shape feature; performing saliency detection on each triangular facet mesh in the triangular facet mesh list according to the overall shape feature to generate a mesh saliency score, thereby obtaining a set of mesh saliency scores; simplifying the three-dimensional mesh model according to each mesh saliency score in the set of mesh saliency scores to obtain a simplified three-dimensional mesh model, and sending the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to store and display the simplified three-dimensional mesh model; wherein the response to detecting that the storage space of the target device exceeds the preset threshold, obtaining the three-dimensional mesh model from the target device and converting the three-dimensional mesh model into the triangular facet mesh list comprises: in response to detecting that the storage space of the target device exceeds the preset threshold, obtaining the three-dimensional mesh model from the target device; treating each triangular facet in the three-dimensional mesh model as a triangular facet mesh in the triangular facet mesh list to obtain the triangular facet mesh list, wherein the triangular facet mesh is a planar region determined by three vertices of a triangular facet; determining a mesh center point coordinate of each triangular facet mesh according to three vertex coordinates of each triangular facet mesh in the triangular facet mesh list to obtain a set of mesh center point coordinates; determining three mesh angle vectors of each triangular facet mesh according to the three vertex coordinates and the mesh center point coordinate of each triangular facet mesh in the triangular facet mesh list to generate a mesh angle vector list, thereby obtaining a set of mesh angle vector lists; performing cross product operation on each mesh angle vector list in the set of mesh angle vector lists to generate a mesh normal vector, thereby obtaining a set of mesh normal vectors; performing adjacent mesh indexing on each triangular facet mesh in the triangular facet mesh list to generate a list of adjacent mesh index numbers, thereby obtaining a set of lists of adjacent mesh index numbers; wherein the local feature extraction on the initial mesh features to obtain mesh local features comprises: performing feature division on the initial mesh features to generate sub-initial mesh features, thereby obtaining a set of sub-initial mesh features; performing attention feature extraction on each sub-initial mesh feature in the set of sub-initial mesh features to generate a weighted attention feature, thereby obtaining a set of weighted attention features; performing splicing on each weighted attention feature in the set of weighted attention features to obtain a weighted overall feature; adding the weighted global feature and the initial mesh feature element by element to obtain a mesh local feature; According to each mesh saliency score in the mesh saliency score set, the three-dimensional mesh model is simplified to obtain a simplified three-dimensional mesh model, and the simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed, comprising: According to the mesh saliency score set, each triangular facet mesh in the triangular facet mesh list is regionally divided to generate a saliency mesh partition, to obtain a saliency mesh partition set, wherein the saliency mesh partition contains a plurality of adjacent triangular facet meshes sharing the same edge; According to a preset order, each saliency mesh partition in the saliency mesh partition set is partition simplified to obtain a simplified mesh partition set; Determine the global geometric error between the saliency mesh partition set and the simplified mesh partition set; In response to the global geometric error being less than or equal to a preset global geometric error threshold, each simplified mesh partition in the simplified mesh partition set is merged to generate a simplified three-dimensional mesh model; The simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and to be stored and displayed.

2. The method of claim 1, wherein, The initial feature extraction of the three-dimensional mesh model according to the triangular facet mesh list comprises: The spatial feature extraction of the mesh center point coordinate set is performed to obtain a mesh spatial feature; The internal structure feature extraction of the mesh angle vector list set is performed to obtain an internal structure feature; The external structure feature extraction of the mesh normal vector set and the adjacent mesh index number list set is performed to obtain an external structure feature; The internal structure feature and the external structure feature are spliced to obtain a mesh structure feature; The feature fusion of the mesh spatial feature and the mesh structure feature is performed to obtain an initial mesh feature.

3. The method of claim 2, wherein, The attention feature extraction of each sub-initial mesh feature in the sub-initial mesh feature set to generate a weighted attention feature comprises: The linear transformation of each sub-initial mesh feature in the sub-initial mesh feature set is performed to generate a transformed sub-initial feature to obtain a transformed sub-initial feature set; Determine the query vector, key vector and value vector of each transformed sub-initial feature in the transformed sub-initial feature set to obtain a query vector set, a key vector set and a value vector set; According to the dot product between each query vector in the query vector set and each key vector in the key vector set, determine the corresponding attention weight to obtain an attention weight set; The dot product processing of each attention weight in the attention weight set and each value vector in the value vector set is performed to generate a weighted attention feature to obtain a weighted attention feature set.

4. The method of claim 3, wherein, The partition simplification of each saliency mesh partition in the saliency mesh partition set according to a preset order to obtain a simplified mesh partition set comprises: According to a preset order, each of the saliency mesh partitions in the saliency mesh partition set is subjected to the following partition simplification steps to generate a simplified mesh partition, thereby obtaining a simplified mesh partition set: According to the vertex coordinates of each triangular mesh in the saliency mesh partition, each mesh vertex and each mesh edge in the saliency mesh partition are determined, thereby obtaining a mesh vertex set and a mesh edge set, wherein a mesh edge is a line segment composed of two mesh vertices in the mesh vertex set; According to the mesh saliency scores of the two triangular meshes adjacent to each mesh edge in the mesh edge set, a mesh edge priority of each mesh edge is determined, thereby obtaining a mesh edge priority list; According to the mesh edge priority list and a preset geometric error threshold, vertex merging is performed on each mesh vertex in the mesh vertex set, thereby obtaining a merged vertex set; Each merged edge corresponding to each pair of merged vertices in the merged vertex set is determined, thereby obtaining a merged edge set; Each merged triangular mesh composed of three merged edges in the merged edge set is determined, thereby obtaining a merged triangular mesh set; The simplified mesh partition corresponding to the merged triangular mesh set is determined.

5. A device for simplifying a three-dimensional mesh model with local and global feature preservation, comprising: an acquisition and conversion unit configured to, in response to detecting that a storage space of a target device exceeds a preset threshold, acquire a three-dimensional mesh model from the target device and convert the three-dimensional mesh model into a triangular mesh list, wherein the three-dimensional mesh model is a three-dimensional model composed of triangular meshes; an initial feature extraction unit configured to perform initial feature extraction on the three-dimensional mesh model according to the triangular mesh list, thereby obtaining initial mesh features; a local feature extraction unit configured to perform local feature extraction on the initial mesh features, thereby obtaining mesh local features; a global average pooling unit configured to perform global average pooling on the mesh local features, thereby obtaining mesh global features; a residual connection unit configured to perform residual connection on the mesh local features and the mesh global features, thereby obtaining overall shape features; a saliency detection unit configured to perform saliency detection on each triangular mesh in the triangular mesh list according to the overall shape features to generate mesh saliency scores, thereby obtaining a mesh saliency score set; a simplification unit configured to simplify the three-dimensional mesh model according to each mesh saliency score in the mesh saliency score set, thereby obtaining a simplified three-dimensional mesh model, and send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display the simplified three-dimensional mesh model; wherein the response to detecting that the storage space of the target device exceeds the preset threshold, acquiring the three-dimensional mesh model from the target device and converting the three-dimensional mesh model into the triangular mesh list comprises: in response to detecting that the storage space of the target device exceeds the preset threshold, acquiring the three-dimensional mesh model from the target device; obtaining a triangle mesh list, where each triangle mesh in the triangle mesh list is a planar area determined by three vertices of a triangle mesh; determining a mesh center point coordinate of each triangle mesh in the triangle mesh list according to three vertex coordinates of each triangle mesh in the triangle mesh list, to obtain a mesh center point coordinate set; determining three mesh angle vectors of each triangle mesh in the triangle mesh list according to three vertex coordinates and the mesh center point coordinate of each triangle mesh in the triangle mesh list, to generate a mesh angle vector list, to obtain a mesh angle vector list set; performing a cross product operation on each mesh angle vector list in the mesh angle vector list set to generate a mesh normal vector, to obtain a mesh normal vector set; performing adjacent mesh indexing on each triangle mesh in the triangle mesh list to generate an adjacent mesh index number list, to obtain an adjacent mesh index number list set; wherein the local feature extraction on the initial mesh feature to obtain a mesh local feature comprises: performing feature division on the initial mesh feature to generate a sub-initial mesh feature, to obtain a sub-initial mesh feature set; performing attention feature extraction on each sub-initial mesh feature in the sub-initial mesh feature set to generate a weighted attention feature, to obtain a weighted attention feature set; splicing each weighted attention feature in the weighted attention feature set to obtain a weighted overall feature; performing element-by-element addition on the weighted overall feature and the initial mesh feature to obtain a mesh local feature; wherein the simplification of the three-dimensional mesh model according to each mesh saliency score in the mesh saliency score set to obtain a simplified three-dimensional mesh model, and the sending of the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to store and display, comprises: performing region division on each triangle mesh in the triangle mesh list according to the mesh saliency score set to generate a saliency mesh partition, to obtain a saliency mesh partition set, where the saliency mesh partition contains multiple adjacent triangle meshes sharing the same edge; performing partition simplification on each saliency mesh partition in the saliency mesh partition set according to a preset order, to obtain a simplified mesh partition set; determining a global geometric error between the saliency mesh partition set and the simplified mesh partition set; in response to the global geometric error being less than or equal to a preset global geometric error threshold, merging each simplified mesh partition in the simplified mesh partition set to generate a simplified three-dimensional mesh model; sending the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and to store and display. 6.An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-4.

7. A computer readable medium having stored thereon a computer program, wherein, The program, when executed by the processor, implements the method of any one of claims 1 to 4.

Images