Quality evaluation method and device, storage medium and computer program product

Abstract

The embodiment of the invention provides a quality evaluation method and device, a storage medium and a computer program product. The method comprises the steps of obtaining first data; wherein the first data comprises point cloud data corresponding to one or more Three Dimensional (3D) models, and the point cloud data comprises point cloud data corresponding to one or more Three Dimensional (3D) models; generating a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model comprises a first module, and the first module is at least used for converting the point cloud data into a first representation form and a second representation form; wherein the first representation form comprises a voxel grid form, and the second representation form comprises a two-dimensional (Two Dimensional, 2D) image form, that is, according to the embodiment of the invention, the quality of the 3D model can be evaluated based on the voxel grid data and the 2D image data at the same time, so that the advantages of the multi-modal data can be fully utilized, and the accuracy of the quality evaluation of the 3D model is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a quality evaluation method, apparatus, storage medium, and computer program product. Background Technology

[0002] With the development of multimedia technology, the number of 3D models has grown rapidly. 3D models have wide applications in entertainment, education, and healthcare, such as immersive experiences in Virtual Reality (VR) / Augmented Reality (AR), realistic scenes and characters in game development, and surgical simulations in the medical field. The application of 3D models not only enhances visual effects and user experience but also optimizes existing industrial design and production processes, improving efficiency and quality. With the widespread application of 3D models in virtual reality, autonomous driving, and industrial inspection, the number of 3D models is showing a rapid growth trend, making the efficient evaluation of 3D model quality crucial.

[0003] However, current methods for evaluating the quality of 3D models only utilize data from a single modality. Single-modal data has limitations, resulting in insufficient accuracy in evaluating model quality. Summary of the Invention

[0004] This application provides a quality evaluation method, apparatus, storage medium, and computer program product that can improve the accuracy of model quality evaluation.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] In a first aspect, embodiments of this application provide a quality evaluation method, the method comprising:

[0007] Acquire first data; wherein, the first data includes point cloud data corresponding to one or more 3D models;

[0008] Based on the point cloud data, a first model generates a quality evaluation result corresponding to the 3D model; wherein, the first model includes a first module, which is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

[0009] Secondly, embodiments of this application provide a quality evaluation device, which includes: an acquisition unit and a generation unit; wherein,

[0010] The acquisition unit is used to acquire first data; wherein, the first data includes point cloud data corresponding to one or more 3D models;

[0011] The generation unit is used to generate a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein, the first model includes a first module, and the first module is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

[0012] Thirdly, embodiments of this application provide a quality evaluation device, which includes: a processor and a memory; wherein,

[0013] The memory is used to store computer programs that can run on the processor;

[0014] The processor is configured to execute the quality evaluation method described above when running the computer program.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program code, which, when executed by a computer, implements the quality evaluation method described above.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the quality evaluation method described above.

[0017] This application provides a quality assessment method, apparatus, storage medium, and computer program product. The method includes: acquiring first data; wherein the first data includes point cloud data corresponding to one or more 3D models; generating a quality assessment result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model includes a first module, the first module being at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form. Therefore, after acquiring the first data, a quality assessment result corresponding to the 3D model can be generated based on the point cloud data corresponding to the 3D model in the first data and the first model. The first model may include a first module, the first module being at least used to convert the point cloud data into a voxel mesh form and a 2D image form respectively. That is, this application embodiment can simultaneously evaluate the quality of the 3D model based on voxel mesh data and 2D image data, thereby fully utilizing the advantages of multimodal data and improving the accuracy of 3D model quality assessment. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the quality evaluation method proposed in the embodiments of this application;

[0019] Figure 2 This is a schematic diagram of the structure of the first model proposed in the embodiments of this application;

[0020] Figure 3 This is a schematic diagram of the network structure of the fourth module proposed in the embodiments of this application;

[0021] Figure 4 This is a schematic diagram of the network structure of the fifth module proposed in the embodiments of this application;

[0022] Figure 5 This is a schematic diagram of the network structure of the third module proposed in an embodiment of this application;

[0023] Figure 6 This is a schematic diagram of the quality evaluation system proposed in the embodiments of this application;

[0024] Figure 7 This is a schematic diagram of the composition and structure of the quality evaluation device proposed in the embodiments of this application. Figure 1 ;

[0025] Figure 8 This is a schematic diagram of the composition and structure of the quality evaluation device proposed in the embodiments of this application. Figure 2 . Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the relevant application and not for limiting the application. Furthermore, it should be noted that, for ease of description, only the parts related to the relevant application are shown in the accompanying drawings.

[0027] With the development of multimedia technology, the number of 3D models has grown rapidly. 3D models have wide applications in entertainment, education, and healthcare, such as immersive experiences in VR / AR, realistic scenes and characters in game development, and surgical simulations in the medical field. The application of 3D models not only enhances visual effects and user experience but also optimizes existing industrial design and production processes, improving efficiency and quality. With the widespread application of 3D models in virtual reality, autonomous driving, and industrial inspection, the number of 3D models is showing a rapid growth trend, making the efficient evaluation of 3D model quality crucial.

[0028] However, current methods for evaluating the quality of 3D models only utilize data from a single modality. Single-modal data has limitations, resulting in insufficient accuracy in evaluating model quality.

[0029] To address the current problem of insufficient accuracy in model quality evaluation, this application provides a quality evaluation method, apparatus, storage medium, and computer program product. The method includes: acquiring first data; wherein the first data includes point cloud data corresponding to one or more 3D models; generating a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model includes a first module, which is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form. Therefore, after acquiring the first data, a quality evaluation result corresponding to the 3D model can be generated based on the point cloud data corresponding to the 3D model in the first data and the first model itself. The first model may include a first module, which is at least used to convert the point cloud data into voxel mesh form and 2D image form respectively. That is, this application embodiment can simultaneously evaluate the quality of the 3D model based on voxel mesh data and 2D image data, thereby fully utilizing the advantages of multimodal data and improving the accuracy of 3D model quality evaluation.

[0030] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0031] This application provides a quality evaluation method. Figure 1 This is a schematic diagram of the quality evaluation method proposed in the embodiments of this application, such as... Figure 1 As shown, a quality assessment method may include the following steps:

[0032] Step 101: Obtain the first data; wherein, the first data includes point cloud data corresponding to one or more 3D models.

[0033] In embodiments of this application, the quality evaluation device can acquire first data.

[0034] It should be noted that, in the embodiments of this application, the quality evaluation device can be any terminal or device with storage and communication functions. For example, the quality evaluation device can be a personal computer (PC). This application does not specifically limit the type of quality evaluation device.

[0035] It should be noted that, in the embodiments of this application, the first data may include point cloud data corresponding to one or more 3D models, and this application does not specifically limit the number of 3D models included in the first data.

[0036] Step 102: Generate quality evaluation results for the 3D model based on point cloud data and the first model; wherein, the first model includes a first module, which is used to convert point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form and the second representation form includes a 2D image form.

[0037] In the embodiments of this application, after acquiring the first data, the quality evaluation device can generate a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model.

[0038] It should be noted that, in the embodiments of this application, the first model can be a quality evaluation model, which can be used to evaluate the quality of the 3D model to obtain the quality evaluation result of the 3D model. This application does not specifically limit the structural type of the quality evaluation model.

[0039] It should be noted that, in the embodiments of this application, the first model can be obtained by training an initial first model based on a first loss function and a training dataset.

[0040] It should be noted that, in the embodiments of this application, the initial first model may include an initial first module, an initial second module, an initial third module, an initial fourth module, and an initial fifth module. This application does not specifically limit the number and type of modules included in the initial first model.

[0041] It should be noted that, in the embodiments of this application, the initial first module may be an initial data preprocessing module. The initial first module is at least used to convert the point cloud data corresponding to the 3D model in the training dataset into a first representation form and a second representation form, respectively; wherein, the first representation form includes a voxel grid form and the second representation form includes a 2D image form.

[0042] It should be noted that, in the embodiments of this application, the initial second module may be an initial feature fusion module. The initial second module is at least used to fuse the 3D feature vector corresponding to the voxel grid and the 2D feature vector corresponding to the 2D image to obtain the fused feature vector.

[0043] It should be noted that, in the embodiments of this application, the initial third module may be an initial quality scoring module, and the initial third module is at least used to perform quality evaluation processing on the fused feature vector.

[0044] It should be noted that, in the embodiments of this application, the initial fourth module may be an initial 3D feature extraction module, and the initial fourth module is at least used to obtain the 3D feature vector corresponding to the voxel grid.

[0045] It should be noted that, in the embodiments of this application, the initial fifth module may be an initial 2D feature extraction module, and the initial fifth module is at least used to obtain the 2D feature vector corresponding to the 2D image.

[0046] It should be noted that, in the embodiments of this application, the training dataset may include original sample data, positive sample data, and negative sample data, and this application does not specifically limit the number of data included in the training dataset.

[0047] It should be noted that, in the embodiments of this application, the original sample data may include one or more first 3D models, and this application does not specifically limit the number of 3D models included in the original sample data.

[0048] It should be noted that, in the embodiments of this application, the positive sample data may include one or more second 3D models. The second 3D model may be obtained by transforming the first 3D model, for example, by performing rigid transformations such as rotation and translation on the first 3D model.

[0049] It should be noted that, in the embodiments of this application, the negative sample data includes one or more third 3D models, which are obtained by adding random noise to the point cloud data of the first 3D model.

[0050] In other words, in the embodiments of this application, the quality evaluation device can perform data augmentation on the training dataset, so that the training dataset can include positive sample data and negative sample data. The positive sample data can include 3D models after rigid transformation processing such as rotation and translation on the 3D models included in the original sample data, and the negative sample data can include 3D models obtained by adding random noise to the point cloud data corresponding to the 3D models in the original sample data. This makes the augmented training dataset contain diverse data types, thereby making the generalization ability and robustness of the first model trained based on the augmented training dataset better.

[0051] It should be noted that, in the embodiments of this application, the first loss function may include a second loss function, a third loss function, a fourth loss function, and a fifth loss function. This application does not specifically limit the type and number of loss functions included in the first loss function.

[0052] For example, in an embodiment of this application, the first loss function may be as shown in the following formula (1).

[0053]

[0054] in, Denotes the first loss function. This represents the second loss function. This represents the third loss function. This represents the fourth loss function. Let λ1, λ2, λ3, and λ4 represent the fifth loss function, and let λ1, λ2, λ3, and λ4 represent the weight values.

[0055] For example, in an embodiment of this application, the second loss function may be a cosine similarity loss function, which can be represented by the following formula (2). The second loss function can be used to measure the similarity between the fused feature vector corresponding to the first 3D model and the third feature vector, where the third feature vector contains the fused feature vector corresponding to the second 3D model.

[0056]

[0057] in, Let F represent the second loss function, and let F represent the fused feature vector corresponding to the first 3D model. gt represents the fused feature vector corresponding to the second 3D model, · represents the dot product, and ∥∥ represents the vector norm.

[0058] For example, in the embodiments of this application, the third loss function may be a triplet loss function, which can be used to minimize the distance between the fused feature vector corresponding to the first 3D model and the fused feature vector corresponding to the second 3D model, and maximize the distance between the fused feature vector corresponding to the first 3D model and the fused feature vector corresponding to the third 3D model. This loss function can be expressed by the following formula (3).

[0059]

[0060] in, Denotes the third loss function, D W F represents the distance between two feature vectors, and F represents the fused feature vector corresponding to the first 3D model. + F represents the fused feature vector corresponding to the second 3D model. - This represents the fused feature vector corresponding to the third 3D model, and m represents the preset threshold.

[0061] It should be noted that, in the embodiments of this application, the third loss function can optimize the initial first model by minimizing the distance between the feature vector corresponding to the original sample data and the feature vector corresponding to the positive sample data, while maximizing the distance between the feature vector corresponding to the original sample data and the feature vector corresponding to the negative sample data.

[0062] For example, in an embodiment of this application, the fourth loss function may be a L2 loss function, which can be used to regularize the offset of the output of the deformable convolutional layer in the initial fourth module. It mainly regularizes the offset of the output of the 3D deformable convolutional layer to prevent overfitting and maintain the stability of the convolutional kernel.

[0063] It should be noted that, in the embodiments of this application, the initial fourth module may include an initial 3D convolutional neural network, which may include M deformable convolutional layers and L normalization layers. The deformable convolutional layers can adaptively adjust the position of the convolutional kernels, thereby better capturing the local features of the 3D model. The size relationship between M and L may include one or more of the following relationships: M = L; M = L + 1. This application does not specifically limit the size of M and L.

[0064] For example, in an embodiment of this application, the fourth loss function may be as shown in the following formula (4).

[0065]

[0066] in, Let λ represent the fourth loss function, λ represent the regularization parameter, and Δp represent the second loss function. i This represents the offset of the i-th position of the convolution kernel.

[0067] It should be noted that, in the embodiments of this application, the fifth loss function can be the mean squared error loss function, which can be used to minimize the distance between the predicted quality score corresponding to the first 3D model and the true score corresponding to the first 3D model, thereby improving the accuracy of the initial first model in scoring the 3D model.

[0068] For example, in an embodiment of this application, the fifth loss function may be as shown in the following formula (5).

[0069]

[0070] in, This represents the fifth loss function, where N represents the number of samples, and y i This represents the true quality score of the i-th sample. This represents the prediction quality score for the i-th sample.

[0071] In other words, in the embodiments of this application, the quality assessment device can perform data augmentation on the training dataset so that the training dataset can include positive sample data and negative sample data, thereby training the initial first model based on the augmented training dataset and the first loss function to obtain the trained first model, thereby improving the generalization ability and robustness of the first model, and improving the accuracy of the first model in predicting the quality score of the 3D model.

[0072] Furthermore, in the embodiments of this application, after the quality evaluation device trains the initial first model based on the first loss function and the training dataset to obtain the first model, it can generate the quality evaluation result corresponding to the 3D model based on the first model.

[0073] It should be noted that, in the embodiments of this application, the first model may include a first module, a second module, and a third module. This application does not specifically limit the number and type of modules included in the first model.

[0074] It should be noted that, in the embodiments of this application, the first module may be a data preprocessing module, which is at least used to convert point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form and the second representation form includes a 2D image form. This application does not specifically limit the function of the first module.

[0075] It should be noted that, in the embodiments of this application, the second module may be a feature fusion module, which is at least used to fuse the first feature vector corresponding to the voxel mesh and the second feature vector corresponding to the 2D image. This application does not specifically limit the function of the second module.

[0076] It should be noted that, in the embodiments of this application, the third module may be a quality scoring module, which is at least used to output the quality score of the 3D model. This application does not specifically limit the function of the third module.

[0077] Optionally, in the embodiments of this application, when the quality evaluation device generates the quality evaluation result corresponding to the 3D model based on point cloud data and the first model, it can first convert the point cloud data into a voxel grid form through the first module, and project the point cloud data into N 2D images according to a preset viewing direction, where N is a positive integer; then, it can fuse the first feature vector corresponding to the voxel grid and the second feature vector corresponding to the N 2D images through the second module to obtain the fused feature vector; wherein, the first feature vector contains a 3D feature vector and the second feature vector contains a 2D feature vector; then, it can perform quality evaluation processing on the fused feature vector based on the third module to obtain the quality evaluation result of the 3D model.

[0078] It should be noted that, in the embodiments of this application, the preset viewing direction may include six viewing directions: front, back, left, right, up, and down. This application does not specifically limit the number and type of preset viewing directions.

[0079] For example, in an embodiment of this application, the quality evaluation device can project point cloud data into 6 2D images according to a preset viewing direction through the first module. Assuming that a coordinate system is established for a 3D model, which is located between (0, 0, 0) and (x0, y0, z0), then the corresponding six projection planes are {x = 0, y = 0, z = 0, x = x0, y = y0, z = z0}. By projecting the data onto the above planes through projection transformation, 6 2D images can be obtained.

[0080] In other words, in the embodiments of this application, after the quality evaluation device converts the point cloud data into a voxel grid and projects the point cloud data into N 2D images according to a preset viewing direction, it can then use the second module to fuse the 3D feature vectors corresponding to the voxel grid and the 2D feature vectors corresponding to the N 2D images to obtain the fused feature vector. The third module can then perform quality evaluation processing on the fused feature vector to obtain the quality evaluation result of the 3D model. That is, the embodiments of this application can perform quality evaluation processing on the fused feature vector based on the third module. This fused feature vector is obtained by fusing the 3D feature vectors corresponding to the voxel grid and the 2D feature vectors corresponding to the N 2D images. This allows the advantages of multimodal data to be utilized when performing quality evaluation processing on the 3D model, fully considering the information of different modal data, thereby greatly improving the accuracy of the 3D model quality evaluation.

[0081] It should be noted that, in the embodiments of this application, Figure 2 This is a schematic diagram of the structure of the first model proposed in the embodiments of this application, as shown below. Figure 2 As shown, in addition to the first module, the second module and the third module, the first model may also include a fourth module and a fifth module. This application does not make specific limitations on the number and type of modules included in the first model.

[0082] It should be noted that, in the embodiments of this application, the fourth module may be a 3D feature extraction module, which is at least used to extract the 3D features corresponding to the voxel mesh. This application does not specifically limit the function of the fourth module.

[0083] It should be noted that, in the embodiments of this application, the fifth module may be a 2D feature extraction module, which is used to extract 2D features corresponding to at least N 2D images. This application does not specifically limit the function of the fifth module.

[0084] Optionally, in the embodiments of this application, after the quality evaluation device converts the point cloud data into a voxel grid through the first module and projects the point cloud data into N 2D images according to a preset viewing direction, it can perform feature extraction processing on the voxel grid through the fourth module to obtain a first feature vector; it can also perform feature extraction processing on the N 2D images through the fifth module to obtain a second feature vector.

[0085] It should be noted that, in the embodiments of this application, the fourth module may include a 3D convolutional neural network, which includes M deformable convolutional layers and L normalization layers, where M and L are positive integers. This application does not specifically limit the type of network structure included in the fourth module.

[0086] For example, in the embodiments of this application, the 3D convolutional neural network includes M deformable convolutional layers and L normalized layers; the size relationship between M and L may include one or more of the following relationships: M = L; M = L + 1. This application does not specifically limit the size of M and L.

[0087] For example, in the embodiments of this application, Figure 3 This is a schematic diagram of the network structure of the fourth module proposed in the embodiments of this application, such as... Figure 3 As shown, the fourth module may include a 3D convolutional neural network, which may include four deformable convolutional layers and four normalization layers.

[0088] Optionally, in embodiments of this application, such as Figure 3 As shown, when the quality evaluation device performs feature extraction processing on the voxel grid through the fourth module to obtain the first feature vector, it can input the voxel grid into M deformable convolutional layers and L normalization layers to obtain the first feature vector; wherein, the deformable convolutional layers are used to adjust the position of the convolution kernel.

[0089] It should be noted that, in the embodiments of this application, as shown in the following formula (6), the sampling position of the deformable convolutional layer is variable.

[0090]

[0091] Among them, w(p) n ) represents the weight at the corresponding position of the convolution kernel, x(p0+p n ) represents p0+p on the input feature map n The element value at position p0, y(p0), represents the element value at position p0 in the output feature map, obtained by convolving the input feature map with the convolution kernel. Variable convolution introduces an offset Δp at each point compared to traditional position-invariant convolution. nThis offset can be learned through an additional convolutional layer, such as a k×k×k convolutional kernel. After passing through the above convolutional layer, a 3k offset will be obtained. 3 The offset is 3, where 3 represents the offset in the x, y, and z directions. Finally, the deformable convolutional layer is obtained through the above process.

[0092] It should be noted that, in the embodiments of this application, the fifth module may include N 2D convolutional neural networks, each of which includes P convolutional layers, S normalization layers and Q max pooling layers, where P, S and Q are positive integers. This application does not specifically limit the network structure type included in the fifth module.

[0093] For example, in the embodiments of this application, each 2D convolutional neural network may include P convolutional layers, S normalization layers and Q max pooling layers; wherein, the sizes of P, S and Q can be equal, that is, P = S = Q, and this application does not specifically limit the size of P, S and Q.

[0094] It should be noted that, in the embodiments of this application, Figure 4 This is a schematic diagram of the network structure of the fifth module proposed in the embodiments of this application, as shown below. Figure 4 As shown, the fifth module may include six 2D convolutional neural networks, each of which may include three convolutional layers, three normalization layers, and three max pooling layers.

[0095] Optionally, in embodiments of this application, such as Figure 4 As shown, when the quality evaluation device performs feature extraction processing on N 2D images through the fifth module to obtain the second feature vector, it can input 6 2D images into 6 2D convolutional neural networks respectively to obtain 6 2D feature vectors; then the 6 2D feature vectors can be concatenated to obtain the second feature vector.

[0096] It should be noted that, in the embodiments of this application, after the quality evaluation device performs feature extraction processing on the voxel grid through the fourth module to obtain the first feature vector and performs feature extraction processing on the N 2D images through the fifth module to obtain the second feature vector, the first feature vector corresponding to the voxel grid and the second feature vector corresponding to the N 2D images can be fused through the second module to obtain the fused feature vector.

[0097] It should be noted that, in the embodiments of this application, the second module may include one or more fully connected layers.

[0098] Optionally, in embodiments of this application, when the quality evaluation device fuses the first feature vector corresponding to the voxel grid and the second feature vector corresponding to N 2D images through the second module to obtain the fused feature vector, it can align the dimensions of the first feature vector and the second feature vector through one or more fully connected layers to obtain the aligned first feature vector and the aligned second feature vector; then the aligned first feature vector and the aligned second feature vector can be weighted and fused to obtain the fused feature vector.

[0099] For example, in the embodiments of this application, when the quality evaluation device aligns the dimensions of the first feature vector and the second feature vector through one or more fully connected layers, it can map the dimensions of the first feature vector and the second feature vector to the same dimension through one or more fully connected layers, for example, mapping to a 128-dimensional vector, thereby obtaining the aligned first feature vector and the aligned second feature vector.

[0100] Furthermore, in the embodiments of this application, after obtaining the aligned first feature vector and the aligned second feature vector, the quality evaluation device can perform a weighted fusion process on the aligned first feature vector and the aligned second feature vector. For example, it can be calculated by the following formula (7) to obtain the fused feature vector.

[0101] F = α 3D F 3D +α 2D F 2D (7)

[0102] Where F represents the fused feature vector, F 3D F represents the first eigenvector after alignment. 2D Let α represent the aligned second eigenvector. 3D and α 2D This represents the weight value.

[0103] For example, in an embodiment of this application, α 3D and α 2D This can represent a weight value, which can be preset, for example, α. 3D =0.6, α 2D =0.4, or the corresponding weighted weight can be obtained through neural network learning. This application does not specify the size of the weight value.

[0104] In other words, in the embodiments of this application, the quality evaluation device can perform weighted fusion of 3D feature vectors and 2D feature vectors to obtain fused feature vectors, thereby enabling subsequent quality evaluation processing of the fused feature vectors to obtain the quality evaluation result of the 3D model. That is, the embodiments of this application can perform quality evaluation processing based on multimodal data types, thereby improving the accuracy of the quality evaluation of the 3D model.

[0105] It should be noted that, in the embodiments of this application, the third module may include I fully connected layers, nonlinear layers and normalization layers, where I is a positive integer, and this application does not specifically limit the number of fully connected layers.

[0106] For example, in the embodiments of this application, Figure 5 This is a schematic diagram of the network structure of the third module proposed in the embodiments of this application, such as... Figure 5 As shown, the third module may include three fully connected layers, a nonlinear layer, and a normalization layer. This application does not specifically limit the structural type of the third module.

[0107] Optionally, in embodiments of this application, such as Figure 5 As shown, when the quality evaluation device performs quality evaluation processing on the fused feature vector based on the third module to obtain the quality evaluation result of the 3D model, it can perform regression processing on the fused feature vector based on three fully connected layers, nonlinear layers and normalization layers to obtain the quality evaluation result of the 3D model; wherein, the quality evaluation result includes the quality score of the 3D model.

[0108] For example, in the embodiments of this application, such as Figure 5 As shown, the quality assessment device can input the fused feature vector into three fully connected layers, a nonlinear layer, and a normalization layer, and then output the quality score of the 3D model.

[0109] It should be noted that, in the embodiments of this application, the quality evaluation device can perform quality evaluation processing on the fused feature vector, thereby obtaining the quality evaluation result of the 3D model. Since the fused feature combines 3D features and 2D features, the accuracy of the quality evaluation of the 3D model is higher.

[0110] In summary, the quality assessment device can first augment the training dataset to include both positive and negative samples. Based on this augmented dataset and the first loss function, the initial first model can be trained to obtain a trained first model, thereby improving its generalization ability and robustness. Then, the quality assessment device can use the first module in the first model to convert the point cloud data into a voxel grid and project it into N 2D images (N being a positive integer) according to a preset viewing direction. Finally, the second module in the first model can fuse the first feature vector corresponding to the voxel grid with the second feature vector corresponding to the N 2D images to obtain a fused model. The fused feature vector is obtained by fusing the 3D feature vector into a first feature vector and the 2D feature vector into a second feature vector. The quality evaluation of the fused feature vector can then be performed based on the third module in the first model to obtain the quality evaluation result of the 3D model. Specifically, this embodiment can perform quality evaluation on the fused feature vector based on the third module. The fused feature vector is obtained by fusing the 3D feature vector corresponding to the voxel mesh and the 2D feature vectors corresponding to N 2D images. This allows the advantages of multimodal data to be utilized when performing quality evaluation on the 3D model, fully considering the information from different modal data, thereby greatly improving the accuracy of the quality evaluation of the 3D model.

[0111] This application provides a quality assessment method, which includes: acquiring first data; wherein the first data includes point cloud data corresponding to one or more 3D models; generating a quality assessment result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model includes a first module, the first module being at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form. Therefore, after acquiring the first data, a quality assessment result corresponding to the 3D model can be generated based on the point cloud data corresponding to the 3D model in the first data and the first model. The first model may include a first module, the first module being at least used to convert the point cloud data into a voxel mesh form and a 2D image form respectively. That is, this application embodiment can simultaneously evaluate the quality of the 3D model based on voxel mesh data and 2D image data, thereby fully utilizing the advantages of multimodal data and improving the accuracy of 3D model quality assessment.

[0112] Based on the above embodiments, another embodiment of this application provides a quality assessment method. When assessing the quality of a 3D model, this method can combine multimodal inputs of point cloud (i.e., voxel mesh data) and projected image data (i.e., 2D image data), and use a neural network to assess the quality of the 3D model. At the same time, it proposes a training method that combines data augmentation and multiple loss functions, which improves the accuracy and robustness of the model (i.e., the first model) quality assessment.

[0113] It should be noted that, in the embodiments of this application, Figure 6 This is a schematic diagram of the quality evaluation system proposed in the embodiments of this application, such as... Figure 6 As shown, the quality evaluation system mainly includes a data preprocessing module (i.e., the first module), a 3D feature extraction module (i.e., the fourth module), a 2D feature extraction module (i.e., the fifth module), a feature fusion module (i.e., the second module), and a quality scoring module (i.e., the third module). Next, each module in the quality evaluation system will be explained in detail.

[0114] It should be noted that, in the embodiments of this application, the data preprocessing module (i.e., the first module) can convert the input point cloud data into a voxel mesh representation (i.e., the first representation), and project it into six 2D images (i.e., N 2D images) along six fixed orthogonal viewpoints (i.e., preset viewpoint directions).

[0115] In other words, in the embodiments of this application, the data preprocessing module can be used to convert point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form and the second representation form includes a 2D image form. This application does not specifically limit the function of the first module.

[0116] It should be noted that, in the embodiments of this application, the 3D feature extraction module (i.e., the fourth module) can input voxel grid data into a 3D convolutional neural network for feature extraction to obtain 3D features (i.e., the first feature vector).

[0117] It should be noted that, in the embodiments of this application, the 2D feature extraction module (i.e., the fifth module) can input six two-dimensional images (i.e., N 2D images) into a 2D convolutional neural network for feature extraction to obtain 2D features (i.e., the second feature vector).

[0118] It should be noted that, in the embodiments of this application, the feature fusion module (i.e., the second module) can fuse 2D features (i.e., the second feature vector) and 3D features (i.e., the first feature vector), map them to a low-dimensional vector through a fully connected layer, and fuse them through specific weights.

[0119] It should be noted that, in the embodiments of this application, the quality scoring module (i.e., the third module) can process the fused feature vector through several fully connected layers to finally output the quality score of the 3D model.

[0120] It should be noted that, in the embodiments of this application, the quality evaluation device can convert the point cloud into a voxel mesh representation (i.e., the first representation form) through the data preprocessing module; at the same time, the point cloud is projected into 6 two-dimensional images according to the orthogonal viewpoint (i.e., the preset viewpoint direction). These six viewpoints include the six directions of front, back, left, right, up, and down. For example, a coordinate system can be established for a 3D model, which is located between (0, 0, 0) and (x0, y0, z0). Then the corresponding six projection planes are {x = 0, y = 0, z = 0, x = x0, y = y0, z = z0}. By projecting onto the above planes through projection transformation, 6 2D images can be obtained.

[0121] It should be noted that, in the embodiments of this application, as... Figure 3 As shown, the 3D feature extraction module (i.e., the fourth module) may include a 3D convolutional neural network, which may include four deformable convolutional layers and four normalization layers. The quality evaluation device may take the voxel grid output by the data preprocessing module as input to the 3D feature extraction module for feature extraction. Among them, the deformable convolutional layer can adaptively adjust the position of the convolutional kernel to better capture the local features of the 3D model. Finally, the network is shaped to output a 3D feature vector (i.e., the first feature vector).

[0122] It should be noted that, in the embodiments of this application, the 2D feature extraction module (i.e. the fifth module) may include N 2D convolutional neural networks, each of which includes P convolutional layers, S normalization layers and Q max pooling layers, where P, S and Q are positive integers.

[0123] For example, in the embodiments of this application, each 2D convolutional neural network may include P convolutional layers, S normalization layers and Q max pooling layers; wherein, the sizes of P, S and Q can be equal, that is, P = S = Q, and this application does not specifically limit the size of P, S and Q.

[0124] It should be noted that, in the embodiments of this application, as... Figure 4 As shown, the 2D feature extraction module (i.e. the fifth module) can include 6 2D convolutional neural networks, each of which can include 3 convolutional layers, 3 normalization layers and 3 max pooling layers.

[0125] It should be noted that, in the embodiments of this application, the quality evaluation device can input the six two-dimensional images (i.e., N 2D images) obtained by the data preprocessing module (i.e., the first module) into the 2D feature extraction module (i.e., the fifth module) for feature extraction. Its structure includes six 2D convolutional neural networks, each corresponding to one of the six images. Each 2D convolutional neural network has the same structure, including multiple convolutional layers, normalization layers, and max pooling layers, which can capture the local features of the image. Finally, the features output from the six channels are shaped to obtain feature vectors (i.e., 2D feature vectors), and then the features are concatenated to obtain the final 2D feature (i.e., the second feature vector) output.

[0126] It should be noted that, in the embodiments of this application, after the quality evaluation device performs feature extraction processing on the voxel grid through the 3D feature extraction module (i.e., the fourth module) to obtain the first feature vector, and performs feature extraction processing on N 2D images through the 2D feature extraction module (i.e., the fifth module) to obtain the second feature vector, the first feature vector corresponding to the voxel grid and the second feature vector corresponding to the N 2D images can be fused through the feature fusion module (i.e., the second module) to obtain the fused feature vector.

[0127] It should be noted that, in the embodiments of this application, the feature fusion module (i.e., the second module) may include one or more fully connected layers.

[0128] Optionally, in embodiments of this application, when the quality evaluation device fuses the first feature vector corresponding to the voxel grid and the second feature vector corresponding to N 2D images through the second module to obtain the fused feature vector, it can align the dimensions of the first feature vector and the second feature vector through one or more fully connected layers to obtain the aligned first feature vector and the aligned second feature vector; then the aligned first feature vector and the aligned second feature vector can be weighted and fused to obtain the fused feature vector.

[0129] For example, in the embodiments of this application, due to differences in input and network structure, the dimensions of 3D feature vector (i.e., the first feature vector) and 2D feature vector (i.e., the second feature vector) may be inconsistent; therefore, firstly, through one or more fully connected layers, the 3D feature vector and 2D feature vector are mapped to the same dimension (e.g., 128-dimensional vector). After the alignment of 3D and 2D features is completed, the 3D feature vector and 2D feature vector can be weighted and fused to obtain the fused feature vector. The weighting formula is shown in the above formula (7).

[0130] In other words, in the embodiments of this application, the quality evaluation device can perform weighted fusion of 3D feature vectors and 2D feature vectors to obtain fused feature vectors, thereby enabling subsequent quality evaluation processing of the fused feature vectors to obtain the quality evaluation result of the 3D model. That is, the embodiments of this application can perform quality evaluation processing based on multimodal data types, thereby improving the accuracy of the quality evaluation of the 3D model.

[0131] It should be noted that, in the embodiments of this application, the quality scoring module (i.e., the third module) may include I fully connected layers, nonlinear layers and normalization layers, where I is a positive integer. This application does not specifically limit the number of fully connected layers.

[0132] For example, in the embodiments of this application, such as Figure 5 As shown, the third module may include three fully connected layers, a nonlinear layer, and a normalization layer. This application does not specifically limit the structural type of the third module.

[0133] It should be noted that, in the embodiments of this application, as... Figure 5 As shown, after obtaining the fused feature vector, the quality assessment device can perform regression on the fused feature vector through three fully connected layers to finally output the quality score of the 3D model. Alternatively, other similar network structures can be used to perform logistic regression to obtain the quality score.

[0134] It should be noted that, in the embodiments of this application, the quality evaluation device can perform quality evaluation processing on the fused feature vector, thereby obtaining the quality evaluation result of the 3D model. Since the fused feature combines 3D features and 2D features, the accuracy of the quality evaluation of the 3D model is higher.

[0135] In other words, in the embodiments of this application, after the quality evaluation device converts the point cloud data into a voxel grid and projects the point cloud data into N 2D images according to a preset viewing direction, it can then use the second module to fuse the 3D feature vectors corresponding to the voxel grid and the 2D feature vectors corresponding to the N 2D images to obtain the fused feature vector. The third module can then perform quality evaluation processing on the fused feature vector to obtain the quality evaluation result of the 3D model. That is, the embodiments of this application can perform quality evaluation processing on the fused feature vector based on the third module. This fused feature vector is obtained by fusing the 3D feature vectors corresponding to the voxel grid and the 2D feature vectors corresponding to the N 2D images. This allows the advantages of multimodal data to be utilized when performing quality evaluation processing on the 3D model, fully considering the information of different modal data, thereby greatly improving the accuracy of the 3D model quality evaluation.

[0136] In summary, the quality assessment device can first augment the training dataset to include both positive and negative samples. Based on this augmented dataset and the first loss function, the initial first model can be trained to obtain a trained first model, thereby improving its generalization ability and robustness. Then, the quality assessment device can use the first module in the first model to convert the point cloud data into a voxel grid and project it into N 2D images (N being a positive integer) according to a preset viewing direction. Finally, the second module in the first model can fuse the first feature vector corresponding to the voxel grid with the second feature vector corresponding to the N 2D images to obtain a fused model. The fused feature vector is obtained by fusing the 3D feature vector into a first feature vector and the 2D feature vector into a second feature vector. The quality evaluation of the fused feature vector can then be performed based on the third module in the first model to obtain the quality evaluation result of the 3D model. Specifically, this embodiment can perform quality evaluation on the fused feature vector based on the third module. The fused feature vector is obtained by fusing the 3D feature vector corresponding to the voxel mesh and the 2D feature vectors corresponding to N 2D images. This allows the advantages of multimodal data to be utilized when performing quality evaluation on the 3D model, fully considering the information from different modal data, thereby greatly improving the accuracy of the quality evaluation of the 3D model.

[0137] This application provides a quality assessment method, which includes: acquiring first data; wherein the first data includes point cloud data corresponding to one or more 3D models; generating a quality assessment result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model includes a first module, the first module being at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form. Therefore, after acquiring the first data, a quality assessment result corresponding to the 3D model can be generated based on the point cloud data corresponding to the 3D model in the first data and the first model. The first model may include a first module, the first module being at least used to convert the point cloud data into a voxel mesh form and a 2D image form respectively. That is, this application embodiment can simultaneously evaluate the quality of the 3D model based on voxel mesh data and 2D image data, thereby fully utilizing the advantages of multimodal data and improving the accuracy of 3D model quality assessment.

[0138] Based on the above embodiments, this application provides a quality evaluation device. Figure 7 Schematic diagram of the composition and structure of the quality evaluation device Figure 1 ,like Figure 7 As shown, the device 10 includes: an acquisition unit 11 and a generation unit 12; wherein,

[0139] The acquisition unit 11 is used to acquire first data; wherein, the first data includes point cloud data corresponding to one or more 3D models;

[0140] The generation unit 12 is used to generate a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein, the first model includes a first module, and the first module is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

[0141] In the embodiments of this application, further, Figure 8 Schematic diagram of the composition and structure of the quality evaluation device Figure 2 ,like Figure 8 As shown, the quality evaluation device 10 proposed in this application embodiment may further include a processor 13, a memory 14 storing instructions executable by the processor 13, and further, the device 10 may also include a communication interface 15 and a bus 16 for connecting the processor 13, the memory 14 and the communication interface 15.

[0142] In the embodiments of this application, the processor 13 can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other types, and this application embodiment does not specifically limit this. The device 10 may also include a memory 14, which can be connected to the processor 13. The memory 14 is used to store executable program code, which includes computer operation instructions. The memory 14 may include high-speed RAM memory and may also include non-volatile memory, such as at least two disk drives.

[0143] In embodiments of this application, bus 16 is used to connect communication interface 15, processor 13 and memory 14 and the mutual communication between these devices.

[0144] In embodiments of this application, memory 14 is used to store instructions and data.

[0145] Further, in an embodiment of this application, the processor 13 is configured to acquire first data; wherein the first data includes point cloud data corresponding to one or more 3D models; and generate a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein the first model includes a first module, the first module being configured to at least convert the point cloud data into a first representation form and a second representation form respectively; wherein the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

[0146] In practical applications, the aforementioned memory 14 can be volatile memory, such as random-access memory (RAM); or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 13.

[0147] This application provides a quality evaluation device that acquires first data, including point cloud data corresponding to one or more 3D models. Based on the point cloud data and the first model, a quality evaluation result corresponding to the 3D model is generated. The first model includes a first module, which is at least used to convert the point cloud data into a first representation form and a second representation form. The first representation form includes a voxel mesh, and the second representation form includes a 2D image. Therefore, after acquiring the first data, a quality evaluation result corresponding to the 3D model can be generated based on the point cloud data and the first model within the first data. The first model can include a first module, which is at least used to convert the point cloud data into both voxel mesh and 2D image forms. This application can simultaneously evaluate the quality of the 3D model based on both voxel mesh data and 2D image data, thus fully utilizing the advantages of multimodal data and improving the accuracy of 3D model quality evaluation.

[0148] This application provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the quality evaluation method described above.

[0149] Specifically, the program instructions corresponding to a quality evaluation method in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to a quality evaluation method in the storage media are read or executed by an electronic device, the following steps are included:

[0150] Acquire first data; wherein, the first data includes point cloud data corresponding to one or more 3D models;

[0151] Based on the point cloud data, a first model generates a quality evaluation result corresponding to the 3D model; wherein, the first model includes a first module, which is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

[0152] This application also provides a computer program product, including a computer program that can be executed by the processor 13 of the quality evaluation device 10 to complete the steps described in any of the foregoing methods.

[0153] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0154] This application is described with reference to schematic and / or block diagrams of implementations of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the schematic and / or block diagrams can be implemented by computer program instructions, and combinations of blocks in the schematic and / or block diagrams can be implemented. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the schematic and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0155] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in the implementation flow diagram. Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0156] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0157] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A quality evaluation method, characterized in that, The method includes: Acquire first data; wherein, the first data includes point cloud data corresponding to one or more three-dimensional 3D models; Based on the point cloud data, a first model generates a quality evaluation result corresponding to the 3D model; wherein, the first model includes a first module, which is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a two-dimensional 2D image form.

2. The method according to claim 1, characterized in that, The first model further includes a second module and a third module. The step of generating the quality evaluation result corresponding to the 3D model based on the point cloud data and the first model includes: The first module converts the point cloud data into the voxel grid form and projects the point cloud data into N 2D images according to a preset viewing direction, where N is a positive integer. The second module fuses the first feature vector corresponding to the voxel grid and the second feature vector corresponding to the N 2D images to obtain a fused feature vector; wherein, the first feature vector contains a 3D feature vector and the second feature vector contains a 2D feature vector. The quality evaluation results of the 3D model are obtained by performing quality evaluation processing on the fused feature vector based on the third module.

3. The method according to claim 2, characterized in that, The first model further includes a fourth module and a fifth module. After the first module converts the point cloud data into the voxel mesh form and projects the point cloud data into N 2D images according to a preset viewing direction, the method further includes: The first feature vector is obtained by extracting the features corresponding to the voxel grid through the fourth module. The second feature vector is obtained by extracting features from the N 2D images through the fifth module.

4. The method according to claim 3, characterized in that, The fourth module includes a 3D convolutional neural network, which comprises M deformable convolutional layers and L normalized layers, where M and L are positive integers. The extraction of features corresponding to the voxel grid through the fourth module to obtain the first feature vector includes: The voxel grid is input into the M deformable convolutional layers and the L normalization layers to obtain the first feature vector; wherein, the deformable convolutional layers are used to adjust the position of the convolutional kernel.

5. The method according to claim 3, characterized in that, The fifth module includes N 2D convolutional neural networks, each of which includes P convolutional layers, S normalization layers, and Q max pooling layers, where P, S, and Q are positive integers. The fifth module extracts features from the N 2D images to obtain the second feature vector, including: The N 2D images are respectively input into the N 2D convolutional neural networks to obtain N 2D feature vectors; The N 2D feature vectors are concatenated to obtain the second feature vector.

6. The method according to claim 2, characterized in that, The second module includes one or more fully connected layers. The process of fusing the first feature vector corresponding to the voxel grid and the second feature vector corresponding to the N 2D images through the second module to obtain the fused feature vector includes: The dimensions of the first feature vector and the second feature vector are aligned using one or more fully connected layers to obtain aligned first feature vector and aligned second feature vector. The aligned first feature vector and the aligned second feature vector are weighted and fused to obtain the fused feature vector.

7. The method according to claim 2, characterized in that, The third module includes I fully connected layers, a nonlinear layer, and a normalization layer, where I is a positive integer. The quality evaluation processing of the fused feature vector based on the third module yields the quality evaluation result of the 3D model, including: The fused feature vector is regressed based on the I fully connected layers, the nonlinear layer, and the normalization layer to obtain the quality evaluation result of the 3D model; wherein, the quality evaluation result includes the quality score of the 3D model.

8. The method according to claim 1, characterized in that, The first model is obtained by training an initial first model based on a first loss function and a training dataset; wherein, the training dataset includes original sample data, positive sample data, and negative sample data; The original sample data includes one or more first 3D models, the positive sample data includes one or more second 3D models, the second 3D models are obtained by transforming the first 3D models, and the negative sample data includes one or more third 3D models, the third 3D models are obtained by adding random noise to the point cloud data of the first 3D models. The first loss function includes a second loss function, a third loss function, a fourth loss function, and a fifth loss function; the second loss function is used to measure the similarity between the fused feature vector corresponding to the first 3D model and the third feature vector, wherein the third feature vector contains the fused feature vector corresponding to the second 3D model; The third loss function is used to minimize the distance between the fused feature vector corresponding to the first 3D model and the fused feature vector corresponding to the second 3D model, and to maximize the distance between the fused feature vector corresponding to the first 3D model and the fused feature vector corresponding to the third 3D model; the fourth loss function is used to regularize the offset output of the deformable convolutional layer; and the fifth loss function is used to minimize the distance between the quality score corresponding to the first 3D model and the true score corresponding to the first 3D model.

9. The method according to claim 8, characterized in that, The initial first model includes an initial first module, an initial second module, an initial third module, an initial fourth module, and an initial fifth module; wherein, The initial first module is at least used to convert the point cloud data corresponding to the 3D model in the training dataset into a first representation form and a second representation form, respectively; wherein, the first representation form includes a voxel grid form, and the second representation form includes a 2D image form; The initial second module is at least used to fuse the 3D feature vector corresponding to the voxel grid and the 2D feature vector corresponding to the 2D image to obtain the fused feature vector. The initial third module is at least used to perform quality evaluation processing on the fused feature vector; The initial fourth module is at least used to obtain the 3D feature vector corresponding to the voxel grid; The initial fifth module is used at least to obtain the 2D feature vector corresponding to the 2D image.

10. A quality evaluation device, characterized in that, The quality evaluation device includes: an acquisition unit and a generation unit; wherein... The acquisition unit is used to acquire first data; wherein, the first data includes point cloud data corresponding to one or more 3D models; The generation unit is used to generate a quality evaluation result corresponding to the 3D model based on the point cloud data and the first model; wherein, the first model includes a first module, and the first module is at least used to convert the point cloud data into a first representation form and a second representation form respectively; wherein, the first representation form includes a voxel mesh form, and the second representation form includes a 2D image form.

11. A quality evaluation device, characterized in that, The quality evaluation device includes: a processor and a memory; wherein... The memory is used to store computer programs that can run on the processor; The processor is configured to perform the method as described in any one of claims 1-9 when running the computer program.

12. A computer-readable storage medium, characterized in that, The storage medium stores computer program code, which, when executed by a computer, performs the method described in any one of claims 1-9.

13. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the method according to any one of claims 1-9.

Images