Three-dimensional grid model simplification method and device for local and global feature preservation

Abstract

The embodiment of the invention discloses a method and a device for simplifying a three-dimensional grid model with local and global feature preservation. A specific embodiment of the method comprises the following steps: in response to detection that a storage space of target equipment exceeds a preset threshold value, converting a three-dimensional mesh model into a triangular patch mesh list; performing initial feature extraction on the three-dimensional grid model to obtain initial grid features; performing local feature extraction on the initial grid features to obtain grid local features; global average pooling is carried out on the grid local features; performing residual connection on the grid local features and the grid global features to obtain overall shape features; performing significance detection on each triangular patch grid to obtain a grid significance score set; and simplifying the three-dimensional grid model according to the grid significance score set to obtain a simplified three-dimensional grid model for storage. According to the embodiment, the storage space of the three-dimensional grid model can be reduced while the shape representation precision of the simplified model is kept.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the fields of computer vision and virtual reality, and in particular to a method and device for simplifying a three-dimensional mesh model while preserving local and global features. Background Art

[0002] Complex three-dimensional models pose great challenges to computer storage, processing, rendering and transmission. The simplification of three-dimensional models is to appropriately reduce the number of vertices and facet meshes of the model without seriously losing the visual features of the three-dimensional object, thereby reducing the complexity of the three-dimensional model. At present, when simplifying three-dimensional mesh models, the commonly used methods are: geometric element folding method, taking the square of the distance from the vertex to the relevant triangle plane as the error, and determining the coordinates of the vertices to be deleted by minimizing the quadratic error; or based on the mesh convolutional neural network, selectively deleting vertices through adaptive mesh pooling operations.

[0003] However, when the above method is used to simplify the three-dimensional mesh model, the following technical problems often occur:

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

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

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0007] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0008] Some embodiments of the present disclosure propose a method and apparatus for simplifying a three-dimensional mesh model while preserving local and global features, to solve one or more of the technical problems mentioned in the above background technology section.

[0009] In a first aspect, some embodiments of the present disclosure provide a method for simplifying a three-dimensional mesh model while preserving local and global features, the method comprising: in response to detecting that the 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 face mesh list, wherein the three-dimensional mesh model is a stereoscopic model composed of meshes in the form of triangular facets; performing initial feature extraction on the three-dimensional mesh model according to the triangular face 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 face mesh in the triangular face mesh list according to the overall shape features to generate a mesh saliency score to obtain a mesh saliency score set; performing simplification on 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 sending the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display it.

[0010] In a second aspect, some embodiments of the present disclosure provide a three-dimensional mesh model simplification device that preserves local and global features, the device comprising: an acquisition and conversion unit, configured to acquire a three-dimensional mesh model from a target device and convert the three-dimensional mesh model into a triangular face mesh list in response to detecting that the storage space of the target device exceeds a preset threshold, wherein the three-dimensional mesh model is a solid model composed of meshes in the form of triangular faces; an initial feature extraction unit, configured to perform initial feature extraction on the three-dimensional mesh model according to the triangular face 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; and a global average pooling unit. , configured to perform global average pooling on the above-mentioned local features of the mesh to obtain global features of the mesh; a residual connection unit, configured to perform residual connection on the above-mentioned local features of the mesh and the above-mentioned global features of the mesh to obtain overall shape features; a saliency detection unit, configured to perform saliency detection on each triangular face mesh in the triangular face mesh list according to the overall shape features to generate a mesh saliency score to obtain a mesh saliency score set; a simplification unit, configured to simplify the three-dimensional mesh model according to the saliency score of each mesh 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 it.

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

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

[0013] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the local and global feature-preserving three-dimensional mesh model simplification method of some embodiments of the present disclosure, the storage space of the three-dimensional mesh model can be effectively reduced while maintaining the simplified model shape representation accuracy. Specifically, the reason why it is difficult to balance between maintaining the model shape accuracy and the storage space is that the geometric element folding method and the mesh convolutional neural network are difficult to take into account the extraction of local details and global geometric information when capturing the local features of the three-dimensional mesh model, resulting in poor performance when extracting the key features of the complex three-dimensional model. Based on this, the local and global feature-preserving 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, obtain a three-dimensional mesh model from the target device and convert the three-dimensional mesh model into a triangular face mesh list. Wherein, the above-mentioned three-dimensional mesh model is a three-dimensional model composed of meshes in the form of triangular facets. Secondly, according to the above-mentioned triangular face mesh list, the above-mentioned three-dimensional mesh model is subjected to initial feature extraction to obtain initial mesh features. Thus, it is helpful to maintain the global geometric structure of the three-dimensional mesh model. Then, the above-mentioned initial mesh features are subjected to local feature extraction to obtain mesh local features. Local feature extraction enhances the ability to capture local geometric changes by focusing on the details of the mesh. Thus, the details and complex features of the model can be retained during the simplification process. Then, the above-mentioned local features of the mesh are globally averaged and pooled to obtain the global features of the mesh. Thus, the overall geometric information of the model can be extracted. Next, the above-mentioned local features of the mesh and the above-mentioned global features of the mesh are residually connected to obtain the overall shape features. Thus, it can be ensured that the important local features can be retained after the model is simplified, and the global geometric information will not be lost. Afterwards, according to the above-mentioned overall shape features, each triangular face mesh in the triangular face mesh list is subjected to saliency detection to generate a mesh saliency score to obtain a mesh saliency score set. Among them, the mesh saliency score reflects the importance of each face in the overall structure and is used to guide the subsequent mesh simplification process. Finally, according to the saliency score of each mesh 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 is 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 THE DRAWINGS

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

[0015] Figure 1 is a flowchart of some embodiments of a local and global feature preserving three-dimensional mesh model simplification method according to the present disclosure;

[0016] Figure 2 is a framework diagram of a local feature recognition module according to some embodiments of the local and global feature preserving three-dimensional mesh model simplification method of the present disclosure;

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

[0018] Figure 4 is an effect diagram of a simplified three-dimensional mesh model according to some embodiments of the three-dimensional mesh model simplification method for preserving local and global features of the present disclosure;

[0019] Figure 5 It is a structural schematic diagram of some embodiments of a three-dimensional mesh model simplification device for preserving local and global features according to the present disclosure;

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

[0021] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying 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. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0022] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0023] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0024] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0025] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these 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 The flowchart 100 of some embodiments of the local and global feature preservation 3D mesh model simplification method according to the present disclosure is shown. The local and global feature preservation 3D mesh model simplification method comprises the following steps:

[0028] Step 101 , in response to detecting that the storage space of the 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 face mesh list.

[0029] In some embodiments, the execution subject of the local and global feature preservation 3D mesh model simplification method can obtain a 3D mesh model from the target device and convert the 3D mesh model into a triangular face mesh list in response to detecting that the storage space of the target device exceeds a preset threshold. The target device can be a local device or a remote device. The 3D mesh model is a solid model composed of meshes in the form of triangular facets.

[0030] In some optional implementations of some embodiments, in response to detecting that the storage space of the target device exceeds a preset threshold, the execution subject acquires the three-dimensional mesh model from the target device and converts the three-dimensional mesh model into a triangular patch mesh list, which may include the following steps:

[0031] In the first step, in response to detecting that the storage space of the target device exceeds a preset threshold, a three-dimensional mesh model is obtained from the target device. The three-dimensional mesh model may be from a public dataset. The public dataset may be a Manifold dataset.

[0032] The second step is to use each triangle in the three-dimensional mesh model as a triangle mesh in the triangle mesh list to obtain the triangle mesh list. The index number is assigned according to the reading order of each triangle in the three-dimensional mesh model, and the coordinate information of the three vertices is recorded as the feature of the triangle mesh to generate the triangle mesh list. Here, the triangle mesh is a plane area determined by the three vertices of the triangle.

[0033] The third step is to determine the mesh center coordinates of each triangular mesh in the triangular mesh list according to the three vertex coordinates of each triangular mesh, and obtain a mesh center coordinate set. First, the average coordinates between the three vertex coordinates of the triangular mesh can be determined. Then, the average coordinates can be used as the mesh center coordinates.

[0034] In practice, each grid center point coordinate in the above grid center point coordinate set can be expressed as a 3-dimensional vector.

[0035] The fourth step is to determine the three mesh angle vectors of each triangular mesh in the above triangular mesh list according to the three vertex coordinates and mesh center point coordinates of each triangular mesh to generate a mesh angle vector list, and obtain a mesh angle vector list set.

[0036] In practice, each grid angle vector list in the above grid angle vector list set can be represented as a 9-dimensional vector.

[0037] Step 5, performing a cross product operation on each grid angle vector list in the above grid angle vector list set to generate a grid normal vector, and obtaining a grid normal vector set. Wherein, a cross product operation can be performed on any two angle vectors in the grid angle vector list to generate an initial grid normal vector. Thereafter, the above initial grid normal vector can be normalized to obtain a grid normal vector. Wherein, the above grid normal vector is a unit normal vector of the plane corresponding to the triangular patch mesh.

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

[0039] Step 6: Perform adjacent mesh indexing on each triangular face mesh in the triangular face mesh list to generate an adjacent mesh index number list, and obtain an adjacent mesh index number list set. Among them, the triangular face meshes in the triangular face mesh list that have the same two vertices as the current triangular face mesh can be used as adjacent meshes of the current triangular face mesh, and the index numbers corresponding to the adjacent meshes are used as adjacent mesh index numbers to generate an adjacent mesh index number list. Among them, the adjacent mesh index number list can contain three adjacent mesh index numbers.

[0040] In practice, each adjacent grid index number list in the adjacent grid index number list set may be represented as a 3-dimensional vector.

[0041] Step 102: extract initial features of the three-dimensional mesh model according to the triangular face mesh list to obtain initial mesh features.

[0042] In some embodiments, the execution entity may perform initial feature extraction on the three-dimensional mesh model according to the triangular face mesh list to obtain initial mesh features.

[0043] In some optional implementations of some embodiments, the execution subject extracts initial features from the three-dimensional mesh model according to the triangular face mesh list to obtain initial mesh features, which may include the following steps:

[0044] The first step is to extract spatial features from the above grid center point coordinate set to obtain grid spatial features. First, each grid center point coordinate in the above grid center point coordinate set can be spliced ​​to obtain a grid center point tensor. Finally, the above grid center point tensor can be nonlinearly mapped through a preset multi-layer perceptron to obtain grid spatial features. The above grid spatial features are used to characterize the spatial features of each triangular patch mesh in the above triangular patch mesh list.

[0045] In practice, the mesh center point tensor is an N×3 dimensional tensor. The N is the number of triangle meshes in the triangle mesh list. The multilayer perceptron may include two fully connected layers, each of which is followed by a batch normalization layer and a ReLU activation function. The fully connected layer may be a 64-dimensional fully connected layer. The mesh spatial feature may be an N×64 dimensional feature.

[0046] The second step is to extract the internal structure features of the above-mentioned grid angle vector list set to obtain the internal structure features. Among them, first, any two grid angle vectors in the grid angle vector list can be spliced ​​to generate a grid angle vector feature vector to obtain a grid angle vector feature vector set. Here, the above-mentioned grid angle vector feature vector set can contain three grid angle vector feature vectors. Among them, one grid angle vector list corresponds to three grid angle vector feature vectors. Secondly, the grid angle vector feature vector sets corresponding to each grid angle vector list in the above-mentioned grid angle vector list set can be spliced ​​to obtain a grid angle vector tensor. After that, the above-mentioned grid angle vector tensor can be convolved through a preset convolution layer to obtain a grid angle vector convolution feature. Then, the above-mentioned grid angle vector convolution feature can be nonlinearly mapped through the above-mentioned multi-layer perceptron to obtain the initial internal structure feature. Finally, the above-mentioned initial internal structure feature is mean pooled to obtain the internal structure feature.

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

[0048] The third step is to extract external structural features from the above-mentioned grid normal vector set and the above-mentioned adjacent grid index number list set to obtain external structural features. Among them, first, each grid normal vector in the above-mentioned grid normal vector set can be spliced ​​to obtain a grid normal vector tensor. Secondly, each adjacent grid index number list in the above-mentioned adjacent grid index number list set can be spliced ​​to obtain an adjacent grid index number tensor. Then, the above-mentioned grid normal vector tensor and the above-mentioned adjacent grid index number tensor can be convolved through a preset second convolution layer to obtain initial external structural features. Finally, the above-mentioned initial external structural features can be nonlinearly mapped through the above-mentioned multi-layer perceptron to obtain external structural features.

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

[0050] The fourth step is to splice the internal structure features and the external structure features to obtain the grid structure features. First, the internal structure features and the external structure features can be spliced ​​to obtain the initial grid structure features. Secondly, the initial grid structure features can be nonlinearly mapped through a preset second multi-layer perceptron to obtain the grid structure features.

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

[0052] The fifth step is to fuse the above-mentioned grid space features and the above-mentioned grid structure features to obtain the initial grid features. Among them, the grid space features and the above-mentioned grid structure features can be firstly spliced ​​to obtain the spliced ​​grid features. Afterwards, the above-mentioned spliced ​​grid features can be nonlinearly mapped by the preset third multi-layer perceptron to obtain the first initial grid features. Secondly, the adjacent grid index number tensor and the above-mentioned grid structure features can be spliced ​​to obtain the spliced ​​grid structure features. Afterwards, the above-mentioned spliced ​​grid structure features are subjected to maximum pooling and average pooling operations, and then nonlinearly mapped by the above-mentioned third multi-layer perceptron to obtain the second initial grid features. Next, the above-mentioned first initial grid features and the above-mentioned second initial grid features are spliced ​​to obtain the spliced ​​initial grid features. Finally, the above-mentioned spliced ​​initial grid features are nonlinearly mapped by the above-mentioned third multi-layer perceptron to obtain the initial grid features.

[0053] In practice, the third multi-layer perceptron may include a 256-dimensional fully connected layer, followed by a batch normalization layer and a ReLU activation function. The first initial grid feature and the second initial grid feature may both be N×256 dimensional. The initial grid feature may be N×256 dimensional.

[0054] Step 103, extracting local features from the initial mesh features to obtain mesh local features.

[0055] In some embodiments, the execution subject may extract local features from the initial mesh features to obtain mesh local features. The local features may be extracted from the initial mesh features by a preset local feature recognition module to obtain mesh local features. The local feature recognition module may be configured as follows: Figure 2 shown.

[0056] In some optional implementations of some embodiments, the execution subject extracts local features from the initial mesh features to obtain mesh local features, which may include the following steps:

[0057] In the first step, the initial grid features are divided into sub-initial grid features to obtain a sub-initial grid feature set. The initial grid features can be divided into sub-initial grid features along a feature dimension to obtain a sub-initial grid feature set.

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

[0059] In the second step, attention features are extracted from each sub-initial grid feature in the above sub-initial grid feature set to generate weighted attention features, thereby obtaining a weighted attention feature set. In this case, attention features can be extracted from each sub-initial grid feature in the above sub-initial grid feature set through a preset attention head to generate weighted attention features, thereby obtaining a weighted attention feature set. Here, the above attention head can be a Transformer self-attention mechanism. In this case, one attention head corresponds to one sub-initial grid feature.

[0060] Optionally, the execution subject extracts attention features from each sub-initial grid feature in the sub-initial grid feature set to generate a weighted attention feature to obtain a weighted attention feature set, which may include the following steps:

[0061] In the first sub-step, each sub-initial mesh feature in the sub-initial mesh feature set is linearly transformed to generate a transformed sub-initial feature, thereby obtaining a transformed sub-initial feature set. The sub-initial mesh features can be linearly transformed to generate the transformed sub-initial features by the following formula.

[0062]

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

[0064] The second sub-step is to 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. The query vector, key vector and value vector of each transformed sub-initial feature in the transformed sub-initial feature set can be determined by the following formula.

[0065]

[0066] Among them, Q i K represents the query vector. i V represents the key vector. i Represents a vector of values. is a weight matrix independent of each attention head.

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

[0068]

[0069] Among them, Attention i Represents the attention weight. softmax() represents the Softmax function.

[0070] is the dimension of the key vector.

[0071] In the fourth sub-step, each attention weight in the attention weight set is dot-producted with each value vector in the value vector set to generate a weighted attention feature, thereby obtaining a weighted attention feature set. The weighted attention feature can be generated by the following formula.

[0072] O i =Attention i ·V i .

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

[0074] The third step is to concatenate each weighted attention feature in the above weighted attention feature set to obtain the weighted overall feature. Figure 2 As shown in C (Concat).

[0075] The fourth step is to add the weighted global features to the initial grid features element by element to obtain the grid local features. The weighted global features can be mapped through a linear layer to obtain the mapped weighted global features of the same dimension as the initial grid features. After that, the mapped weighted global features can be added to the initial grid features element by element to obtain the grid local features.

[0076] In practice, the above step 103 divides the initial mesh feature into multiple sub-features along the feature dimension, and then uses a multi-head attention mechanism to process each sub-feature separately. In this way, the key local features in the mesh model can be effectively captured, the core area can be highlighted, and the overall efficiency of simplifying the three-dimensional mesh model can be improved.

[0077] Step 104, performing global average pooling on the local features of the grid to obtain the global features of the grid.

[0078] In some embodiments, the execution subject may perform global average pooling on the local features of the mesh to obtain the global features of the mesh. The global features of the mesh may be obtained by performing global average pooling on the local features of the mesh using the following formula.

[0079]

[0080] in, Represents the value of the jth triangular patch in the mesh global feature. local The number of feature channels representing the local features of the mesh. Represents the local feature of the grid F local The value of the cth feature channel of the jth triangular patch mesh in F. globalRepresents the global characteristics of the grid.

[0081] In practice, the above-mentioned global mesh features can reflect the geometric structure of the entire 3D mesh model.

[0082] Step 105, performing residual connection on the local features of the mesh and the global features of the mesh to obtain the overall shape features.

[0083] In some embodiments, the execution subject may perform residual connection on the local features of the mesh and the global features of the mesh to obtain the overall shape features, as shown in the following formula:

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

[0085] Among them, F entire represents the overall shape feature. λ represents the learnable scaling factor.

[0086] In practice, a learnable scaling factor λ is set to dynamically balance the weights of local features and global features, ensuring that the fused features can retain local details while reflecting the overall shape structure. The above overall shape features effectively integrate local details and global information, enhancing the ability to describe complex three-dimensional shapes.

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

[0088] In some embodiments, the execution entity may perform a saliency test on each triangular face mesh in the triangular face mesh list according to the overall shape feature to generate a mesh saliency score, thereby obtaining a mesh saliency score set.

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

[0090] In the first step, the overall shape feature is divided into grid dimensions to generate grid shape features, thereby obtaining a grid shape feature set, wherein the grid shape feature may be a vector consisting of a row of values ​​in the overall shape feature.

[0091] For example, F entire ={f1, f2, f i ..., f N}, Among them, f irepresents the mesh shape feature in the mesh shape feature set, d e The number of feature channels representing the overall shape characteristics.

[0092] The second step is to determine the neighborhood grid feature difference matrix between each grid shape feature in the grid shape feature set and each adjacent grid shape feature according to the adjacent grid index number list set, and obtain the neighborhood grid feature difference matrix set. The neighborhood grid feature difference matrix can be determined by the following formula.

[0093]

[0094] Where, ΔF i represents the neighborhood mesh feature difference matrix of triangular patch mesh i, P is the neighborhood size of the triangle mesh i, and the value of p is between 1 and 3. pc is the value of the pth row and cth column in the neighborhood grid feature difference matrix. i (c) represents f i The c-th eigenvalue of . Represents the cth eigenvalue of the mesh shape features of the pth neighboring mesh of triangular patch mesh i.

[0095] The third step is to concatenate the neighborhood grid feature difference matrices in the above neighborhood grid feature difference matrix set 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] The fourth step is to perform significant attention extraction on the above neighborhood feature difference matrix to obtain a significant attention matrix. The significant attention extraction can be performed on the above neighborhood feature difference matrix to obtain a significant attention matrix by the following formula.

[0097]

[0098] Where X represents the salient attention matrix, X∈R N×d Atten() is a function that obtains the attention distribution from the neighborhood feature difference matrix. I2 and W I1 They are 1×1 convolution kernel, ReLU(W I2 (ReLU(W i1 The output dimension of ΔF) can be N×P×d. Sum the output in P dimension to get the channel attention map A channel ∈R N×d; The output is average pooled in P dimension to obtain the spatial attention map A spatial ∈R N×d Where X = A channel +A spatial .

[0099] The fifth step is to process the above-mentioned salient attention matrix with the maximum value of the salient score to obtain the initial salient score vector. The maximum value of each row in the above-mentioned salient attention matrix can be used as the value of the corresponding row of the initial salient score vector to obtain the initial salient score vector. The dimension of the initial salient score vector is N.

[0100] Step 6: According to the initial saliency score vector, the initial mesh saliency score of each triangular mesh in the triangular mesh list is determined to obtain an initial mesh saliency score set, wherein each element in the initial saliency score vector corresponds to each triangular mesh in the triangular mesh list.

[0101] In the seventh step, each initial grid saliency score in the initial grid saliency score set is subjected to reverse interpolation processing to generate a grid saliency score, thereby obtaining a grid saliency score set. The initial grid saliency score can be subjected to reverse interpolation processing to generate a grid saliency score by the following formula.

[0102]

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

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

[0105] In practice, the above-mentioned salient 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 above-mentioned channel attention map and the above-mentioned spatial attention map in turn to obtain the weighted overall shape feature. Afterwards, the high-dimensional weighted overall shape feature is mapped to the classification space through the fully connected layer. Then, the output is converted into category probability using the Softmax function. Subsequently, the cross entropy loss function is used to determine the loss value between the above-mentioned category probability and the true label. This loss value can be used to guide the back propagation update parameters to improve the accuracy of shape classification. Among them, the above-mentioned parameters may include the parameters of the above-mentioned each convolution kernel, the parameters of the fully connected layer, and the above-mentioned weight matrix W. i , Bias term b i , a learnable scaling factor λ.

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

[0107] Step 107, simplifying the three-dimensional mesh model according to each mesh significance score in the mesh significance score set 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 store and display it.

[0108] In some embodiments, the above-mentioned execution entity can simplify the three-dimensional mesh model according to each mesh significance score in the mesh significance 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 for storage and display.

[0109] In some optional implementations of some embodiments, the execution subject simplifies the three-dimensional mesh model according to each mesh significance score in the mesh significance 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 it, which may include the following steps:

[0110] In the first step, according to the mesh saliency score set, each triangular face mesh in the triangular face mesh list is divided into regions to generate saliency mesh partitions, and a saliency mesh partition set is obtained. Among them, the saliency mesh partition contains multiple adjacent triangular face meshes sharing the same edge. First, each triangular face mesh in the triangular face mesh list can be evenly divided into regions in the order of mesh saliency scores from low to high to generate saliency mesh partitions, and a saliency mesh partition set is obtained.

[0111] In the second step, each significant grid partition in the significant grid partition set is partition-simplified according to a preset order to obtain a simplified grid partition set. In particular, each significant grid partition in the significant grid partition set is partition-simplified according to the order of grid significance scores from low to high to obtain a simplified grid partition set.

[0112] Optionally, the execution subject may perform the following partition simplification steps on each significant grid partition in the significant grid partition set according to a preset order to generate a simplified grid partition, thereby obtaining a simplified grid partition set:

[0113] In the first sub-step, each mesh vertex and each mesh edge in the significant mesh partition are determined according to the vertex coordinates of each triangular face mesh in the significant mesh partition to obtain 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.

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

[0115]

[0116] Among them, e ij Represents the mesh edge saliency score of the mesh edge between triangular mesh i and triangular mesh j. Then, the mesh edge priority of each mesh edge in the mesh edge set is determined in descending order of the mesh edge saliency score, and a mesh edge priority list is obtained. Among them, the mesh edge with a low mesh edge saliency score has a high mesh edge priority.

[0117] The third sub-step is to merge the mesh vertices in the mesh vertex set according to the mesh edge priority list and a preset geometric error threshold to obtain a merged vertex set.

[0118] First, in descending order of mesh edge priority, two mesh vertices corresponding to the mesh edge with the highest mesh edge priority are merged to obtain a merged vertex.

[0119] Here, the above merged vertices are determined by the following formula, such as Figure 3 shown.

[0120]

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

[0122] Afterwards, each mesh vertex that has not been merged in the mesh vertex set and the merged vertex are used 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] Where E represents the maximum distance error. A represents the initial merged vertex set. B represents the mesh vertex set. a and b are any points in the A and B point sets, respectively. D(a, b) is the Euclidean distance between the two points.

[0125] Then, if the maximum distance error is less than the preset geometric error threshold, continue to perform the vertex merging step. If the maximum distance error is greater than or equal to the preset geometric error threshold, the 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 significant mesh partition and the simplified mesh partition to ensure that the maximum distance error allowed during the vertex merging process does not exceed the preset tolerance. Here, each significant mesh partition in the significant mesh partition set corresponds to a different geometric error threshold.

[0126] The fourth sub-step is to determine the merged edge corresponding to every two merged vertices in the merged vertex set to obtain a merged edge set.

[0127] The fifth sub-step is to determine the merged triangle facets composed of every three merged edges in the merged edge set to obtain a merged triangle facet set.

[0128] The sixth sub-step is to determine the simplified mesh partition corresponding to the above-mentioned merged triangular face set.

[0129] The third step is to determine the global geometric error between the above-mentioned significant grid partition set and the above-mentioned simplified grid partition set. The global geometric error between the above-mentioned significant grid partition set and the above-mentioned simplified grid partition set can be obtained by the above-mentioned maximum distance error determination formula based on the coordinates of all grid vertices in the above-mentioned significant grid partition set and the coordinates of all merged vertices in the above-mentioned simplified grid partition set. No further details will be given 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, the simplified mesh partitions in the simplified mesh partition set are 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 to ensure that the maximum distance error allowed during 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 lowered and the second to third steps are re-executed. If the global geometric error is less than or equal to the preset global geometric error threshold, the simplified mesh partitions in the simplified mesh partition set are merged to generate a simplified three-dimensional mesh model. The effect diagram of the simplified three-dimensional mesh model is shown in the figure below. Figure 4 shown.

[0131] The fifth step is to send the simplified three-dimensional mesh model to the target device to replace the three-dimensional mesh model before simplification and store and display it. The simplified three-dimensional mesh model can be sent to the target device to replace the three-dimensional mesh model before simplification and store and display it through a wired or wireless transmission method.

[0132] The above steps 106-107 and their related contents, as an inventive point of an embodiment of the present disclosure, solve the technical problem "increased consumption of computing resources" mentioned in the background technology. 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 computing efficiency while retaining important features. If the above factors are solved, the situation of excessive consumption of computing resources can be reduced. In order to achieve this effect, first, the saliency of the three-dimensional mesh model is characterized by the feature difference between each triangular face mesh and its adjacent meshes. Thus, the saliency of each triangular face mesh is represented as a quantifiable neighborhood feature difference matrix, which is convenient for subsequent simplification operations. Secondly, through collaborative training, the three-dimensional mesh model can be simultaneously classified and scored for shape, which improves the accuracy of the saliency score and reduces the consumption of computing resources. Among them, the saliency score reflects the contribution of each triangular face in the overall structure and is used to guide the priority sorting of subsequent mesh simplification. Thus, the core feature area can be highlighted, the retention effect of the saliency area can be improved, and a more accurate basis is provided for mesh simplification, which optimizes the geometric quality and computing efficiency of the overall model during the simplification process. At the same time, the preset geometric error threshold and global geometric error threshold are used to avoid deleting highly significant meshes during the simplification process, and the important features of the simplified model are retained. This improves the computational efficiency and ensures that the simplified 3D model still has a high-precision shape representation, which is suitable for complex 3D data processing scenarios.

[0133] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: through the local and global feature-preserving three-dimensional mesh model simplification method of some embodiments of the present disclosure, the storage space of the three-dimensional mesh model can be effectively reduced while maintaining the simplified model shape representation accuracy. Specifically, the reason why it is difficult to balance between maintaining the model shape accuracy and the storage space is that the geometric element folding method and the mesh convolutional neural network are difficult to take into account the extraction of local details and global geometric information when capturing the local features of the three-dimensional mesh model, resulting in poor performance when extracting the key features of the complex three-dimensional model. Based on this, the local and global feature-preserving 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, obtain a three-dimensional mesh model from the target device and convert the three-dimensional mesh model into a triangular face mesh list. Wherein, the above-mentioned three-dimensional mesh model is a three-dimensional model composed of meshes in the form of triangular facets. Secondly, according to the above-mentioned triangular face mesh list, the above-mentioned three-dimensional mesh model is subjected to initial feature extraction to obtain initial mesh features. Thus, it is helpful to maintain the global geometric structure of the three-dimensional mesh model. Then, the above-mentioned initial mesh features are subjected to local feature extraction to obtain mesh local features. Local feature extraction enhances the ability to capture local geometric changes by focusing on the details of the mesh. Thus, the details and complex features of the model can be retained during the simplification process. Then, the above-mentioned local features of the mesh are globally averaged and pooled to obtain the global features of the mesh. Thus, the overall geometric information of the model can be extracted. Next, the above-mentioned local features of the mesh and the above-mentioned global features of the mesh are residually connected to obtain the overall shape features. Thus, it can be ensured that the important local features can be retained after the model is simplified, and the global geometric information will not be lost. Afterwards, according to the above-mentioned overall shape features, each triangular face mesh in the triangular face mesh list is subjected to saliency detection to generate a mesh saliency score to obtain a mesh saliency score set. Among them, the mesh saliency score reflects the importance of each face in the overall structure and is used to guide the subsequent mesh simplification process. Finally, according to the saliency score of each mesh 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 is 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 references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a three-dimensional mesh model simplification device that preserves local and global features. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0135] like Figure 5 As shown, in some embodiments, a 3D mesh model simplification device 500 that preserves local and global features 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, and a simplification unit 507. The acquisition and conversion unit 501 is configured to acquire a 3D mesh model from the target device and convert the 3D mesh model into a triangular patch mesh list in response to detecting that the storage space of the target device exceeds a preset threshold, wherein the 3D mesh model is a stereoscopic model composed of meshes in the form of triangular patches; the initial feature extraction unit 502 is configured to perform initial feature extraction on the 3D mesh model according to the triangular patch 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 local features; The mesh global features are obtained by pooling; the residual connection unit 505 is configured to perform residual connection on the above-mentioned mesh local features and the above-mentioned mesh global features to obtain the overall shape features; the saliency detection unit 506 is configured to perform saliency detection on each triangular face mesh in the triangular face mesh list according to the above-mentioned overall shape features to generate a mesh saliency score and obtain a mesh saliency score set; the simplification unit 507 is configured to simplify the three-dimensional mesh model according to the saliency score of each mesh 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 it.

[0136] It is understood that the units described in the device 500 are similar to those described in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device 500 and the units included therein, and will not be described in detail here.

[0137] Reference below Figure 6 , which shows a structural schematic diagram of an electronic device (eg, a computing device) 600 suitable for implementing some embodiments of the present disclosure. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0138] like Figure 6As shown, the electronic device 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory 602 or a program 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 via a bus 604. An input / output interface 605 is also connected to the bus 604.

[0139] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 6 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0140] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from a network through a communication device 609, or installed from a storage device 608, or installed from a read-only memory 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0141] It should be noted that the computer-readable medium recorded in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

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

[0143] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: in response to detecting that the storage space of the target device exceeds a preset threshold, obtains a three-dimensional mesh model from the target device and converts the three-dimensional mesh model into a triangular mesh list, wherein the three-dimensional mesh model is a solid model composed of meshes in the form of triangular meshes; according to the triangular mesh list, performs initial feature extraction on the three-dimensional mesh model to obtain initial mesh features; performs local feature extraction on the initial mesh features to obtain mesh local features; performs global average pooling on the mesh local features to obtain mesh global features; performs residual connection on the mesh local features and the mesh global features to obtain overall shape features; according to the overall shape features, performs saliency detection on each triangular mesh in the triangular mesh list to generate a mesh saliency score to obtain a mesh saliency score set; according to the saliency score of each mesh in the mesh saliency score set, simplifies the three-dimensional mesh model 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 it.

[0144] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0145] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0146] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be provided in a processor, for example, may be described as: a processor including an acquisition and conversion 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, in some cases, constitute limitations on the units themselves, for example, a simplification unit may also be described as a "unit for simplifying a three-dimensional mesh model".

[0147] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0148] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A method for simplifying a three-dimensional mesh model while preserving local and global features, comprising: In response to detecting that the storage space of the target device exceeds a preset threshold, acquiring a three-dimensional mesh model from the target device and converting the three-dimensional mesh model into a triangular face mesh list, wherein the three-dimensional mesh model is a solid model composed of meshes in the form of triangular facets; According to the triangular face mesh list, initial feature extraction is performed on the three-dimensional mesh model to obtain initial mesh features; Performing local feature extraction on the initial grid features to obtain grid local features; Performing global average pooling on the local features of the grid to obtain global features of the grid; Performing residual connection on the local features of the grid and the global features of the grid to obtain overall shape features; According to the overall shape feature, performing saliency detection on each triangular face mesh in the triangular face mesh list to generate a mesh saliency score, and obtaining a mesh saliency score set; According to each mesh significance score in the mesh significance 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 is stored and displayed.

2. The method according to claim 1, wherein: In response to detecting that the storage space of the target device exceeds a preset threshold, acquiring a three-dimensional mesh model from the target device and converting the three-dimensional mesh model into a triangular face mesh list includes: In response to detecting that the storage space of the target device exceeds a preset threshold, acquiring a three-dimensional mesh model from the target device; Taking each triangular face in the three-dimensional mesh model as a triangular face mesh in a triangular face mesh list, to obtain a triangular face mesh list, wherein the triangular face mesh is a plane area determined by three vertices of the triangular face; Determine the coordinates of the center point of each triangular patch mesh according to the three vertex coordinates of each triangular patch mesh in the triangular patch mesh list to obtain a set of the center point coordinates of each triangular patch mesh; According to the three vertex coordinates and the coordinates of the mesh center point of each triangular face mesh in the triangular face mesh list, three mesh angle vectors of each triangular face mesh are determined to generate a mesh angle vector list, thereby obtaining a mesh angle vector list set; Performing a cross product operation on each grid angle vector list in the grid angle vector list set to generate a grid normal vector, thereby obtaining a grid normal vector set; An adjacent mesh index is performed on each triangular patch mesh in the triangular patch mesh list to generate an adjacent mesh index number list, thereby obtaining an adjacent mesh index number list set.

3. The method according to claim 2, wherein: The step of extracting initial features from the three-dimensional mesh model according to the triangular face mesh list to obtain initial mesh features includes: Extracting spatial features from the grid center point coordinate set to obtain grid spatial features; Extracting internal structural features from the grid angle vector list set to obtain internal structural features; Extracting external structural features from the grid normal vector set and the adjacent grid index number list set to obtain external structural features; Splicing the internal structural features with the external structural features to obtain a grid structural feature; The grid space features and the grid structure features are fused to obtain initial grid features.

4. The method according to claim 1, wherein: The extracting local features of the initial grid features to obtain grid local features includes: Performing feature division on the initial grid features to generate sub-initial grid features, thereby obtaining a sub-initial grid feature set; Performing attention feature extraction on each sub-initial grid feature in the sub-initial grid feature set to generate a weighted attention feature, thereby obtaining a weighted attention feature set; Concatenating each weighted attention feature in the weighted attention feature set to obtain a weighted overall feature; The weighted global feature is added to the initial grid feature element by element to obtain a grid local feature.

5. The method according to claim 4, wherein: The step of extracting attention features from each sub-initial grid feature in the sub-initial grid feature set to generate a weighted attention feature to obtain a weighted attention feature set includes: Performing a linear transformation on each sub-initial grid feature in the sub-initial grid feature set to generate a transformed sub-initial feature, thereby obtaining a transformed sub-initial feature set; Determine a query vector, a key vector and a 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; Determine a corresponding attention weight according to a dot product between each query vector in the query vector set and each key vector in the key vector set to obtain an attention weight set; Each attention weight in the attention weight set is dot-producted with each value vector in the value vector set to generate a weighted attention feature, thereby obtaining a weighted attention feature set.

6. The method according to claim 1, wherein: The method of simplifying the three-dimensional mesh model according to each mesh significance score in the mesh significance score set 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 store and display the model, comprises: According to the mesh saliency score set, each triangular face mesh in the triangular face mesh list is divided into regions to generate saliency mesh partitions, thereby obtaining a saliency mesh partition set, wherein the saliency mesh partitions contain a plurality of adjacent triangular face meshes that share the same edge; According to a preset order, partition simplification is performed on each significant grid partition in the significant grid partition set to obtain a simplified grid partition set; determining a global geometric error between the set of significant grid partitions and the set of simplified grid partitions; 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; The simplified three-dimensional mesh model is sent to the target device to replace the three-dimensional mesh model before simplification and is stored and displayed.

7. The method according to claim 6, wherein: The step of performing partition simplification on each significant grid partition in the significant grid partition set according to a preset order to obtain a simplified grid partition set includes: According to a preset order, the following partition simplification steps are performed on each salient grid partition in the salient grid partition set to generate a simplified grid partition, thereby obtaining a simplified grid partition set: According to the vertex coordinates of each triangular patch mesh in the significant mesh partition, each mesh vertex and each mesh edge in the significant mesh partition is determined to obtain 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; Determine the mesh edge priority of each mesh edge according to the mesh significance scores of two triangular patch meshes adjacent to each mesh edge in the mesh edge set, and obtain a mesh edge priority list; Merging each mesh vertex in the mesh vertex set according to the mesh edge priority list and a preset geometric error threshold to obtain a merged vertex set; Determine the merged edge corresponding to every two merged vertices in the merged vertex set to obtain a merged edge set; Determine a merged triangle face composed of every three merged edges in the merged edge set to obtain a merged triangle face set; Determine a simplified mesh partition corresponding to the merged triangular face set.

8. A three-dimensional mesh model simplification device that preserves local and global features, comprising: an acquisition and conversion unit, configured to acquire a three-dimensional mesh model from the target device and convert the three-dimensional mesh model into a triangular face mesh list in response to detecting that the storage space of the target device exceeds a preset threshold, wherein the three-dimensional mesh model is a solid model composed of meshes in the form of triangular facets; An initial feature extraction unit is configured to perform initial feature extraction on the three-dimensional mesh model according to the triangular face mesh list to obtain initial mesh features; A local feature extraction unit is configured to perform local feature extraction on the initial grid feature to obtain a grid local feature; A global average pooling unit is configured to perform global average pooling on the local features of the grid to obtain a global feature of the grid; A residual connection unit is configured to perform residual connection on the local features of the grid and the global features of the grid to obtain an overall shape feature; a saliency detection unit configured to perform saliency detection on each triangular face mesh in the triangular face mesh list according to the overall shape feature to generate a mesh saliency score, thereby obtaining a mesh saliency score set; The simplification unit is configured to simplify the three-dimensional mesh model according to each mesh significance score in the mesh significance 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 it.

9. 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 according to any one of claims 1 to 7.

10. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images