Point cloud geometric compression method and system based on interlayer residual error and IRN connection residual error

Abstract

The invention relates to the technical field of point cloud compression, in particular to a point cloud geometric compression method and system based on interlayer residual errors and IRN connection residual errors. The method comprises the steps of performing feature extraction by using a convolutional neural network based on sparse convolution, performing up-sampling after down-sampling based on a D-U residual mechanism, and performing geometric reduction on feature information before up-sampling to obtain an up-sampling residual and a down-sampling residual; calculating a loss function by using the upper and lower sampling residual errors, and optimizing the convolutional neural network by minimizing the loss function; capturing the extracted features based on an IRN connection residual mechanism to obtain global features; coding and decoding the global features based on a codec; and the decoded information is predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud compression, and in particular to a point cloud geometry compression method and system based on inter-layer residuals and IRN connection residuals. Background Art

[0002] Geometry-based point cloud compression methods are applicable to static point clouds and dynamically acquired point clouds. They reduce the amount of data while maintaining the geometric structure by constructing and encoding the geometric relationship between points. However, this method has limited ability to process irregular point clouds. In addition, the most commonly used partitioning method for geometry-based point cloud compression methods is the octree method, which recursively divides the three-dimensional space into eight subspaces to efficiently represent and compress the spatial distribution of the point cloud. This results in a sharp increase in the number of bits required for this method as the depth of the octree increases, and the required storage space and bandwidth during transmission also increase significantly.

[0003] The point cloud geometry compression method based on plane approximation uses the nodes divided by the pruned octree in a larger space to capture the continuous point distribution within the node. This method is suitable for densely sampled and smooth surface point clouds, but due to the inherent error of plane approximation, this method cannot achieve lossless compression.

[0004] Video-based point cloud compression projects a 3D point cloud into a 2D plane and then processes the point cloud using existing high-efficiency video codecs (such as HEVC). This method is suitable for dense and dynamic point clouds, but it cannot avoid the loss of a large number of occluded continuous points during the projection process, resulting in obvious artifacts in the compressed point cloud.

[0005] With the powerful ability of deep learning in feature extraction and data encoding, the learning point cloud compression method has achieved a good balance between compression rate and reconstruction quality. Therefore, the learning point cloud compression method came into being. It can well solve the problems of the above-mentioned traditional point cloud and adapt to different types of point cloud data through training.

[0006] The voxel-based point cloud geometry compression method is a type of learning point cloud compression. It achieves efficient representation and compression by dividing the point cloud data into a three-dimensional grid and processing the voxels. However, for sparse point clouds, most voxels are empty, but 3D convolution is still evenly applied to each voxel. Therefore, this method does not effectively utilize the sparsity between point clouds, resulting in the consumption of a large amount of space and computing resources. In addition, in the end-to-end compression framework, since the downsampling operation will inevitably lead to the loss of part of the point cloud information while extracting the key features of the point cloud, the downsampling accuracy will be reduced and the reconstruction quality will be reduced. When the encoder processes the point cloud attribute information, the entropy model itself cannot directly process continuous floating-point type data. Therefore, there is an urgent need for a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals. Summary of the invention

[0007] In order to solve the above-mentioned problems, the present invention provides a point cloud geometry compression method and system based on inter-layer residuals and IRN connection residuals.

[0008] In a first aspect, the present invention provides a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals, which adopts the following technical solution: A point cloud geometry compression method based on inter-layer residuals and IRN connection residuals, comprising: Get point cloud data; A convolutional neural network based on sparse convolution is used for feature extraction, in which upsampling after downsampling is performed based on the DU residual mechanism, and geometric reduction is performed with the feature information before upsampling to obtain up- and downsampling residuals; the up- and downsampling residuals are used to calculate the loss function, and the convolutional neural network is optimized by minimizing the loss function; The extracted features are captured based on the IRN connection residual mechanism to obtain global features; Encode and decode the global features based on the codec; Make predictions on the decoded information.

[0009] Furthermore, the feature extraction using a convolutional neural network based on sparse convolution includes first performing a sparse convolution operation on the input 3D original point cloud data to expand the input 1-channel data to 16 channels, then activating it with a Relu function, and performing a downsampling operation using sparse convolution to expand the feature data from 16 channels to 32 channels while halving the size.

[0010] Furthermore, the upsampling after downsampling based on the DU residual mechanism is geometrically reduced with the feature information before upsampling to obtain the upsampling and downsampling residuals, including the feature information after downsampling. Perform upsampling operation to , and the feature information before upsampling Perform geometric reduction to obtain the residual , expressed as: in, Represents the coordinate residual result value, and They are respectively represented as the geometric coordinates of the i-th point before upsampling and downsampling, that is, the coordinate difference of each point in the three-dimensional component, expressed as ( ), represents the upsampling operation, Represents a downsampling operation.

[0011] Furthermore, the loss function is calculated by using the up- and down-sampling residuals, and the convolutional neural network is optimized by minimizing the loss function, including using the up- and down-sampling residual result values ​​for calculating the loss function. During the training process, the model is optimized by minimizing the loss function, which is expressed as: in, represents the residual loss value of upsampling and downsampling calculated on the encoder side, is the activation function, M is the number of points, The result value calculated by the formula in claim 3 represents the coordinate residual.

[0012] Furthermore, the extracted features are captured based on the IRN connection residual mechanism to obtain global features, including stacking three IRN residuals using a multi-layer residual connection method, wherein the feature information is first processed by an IRN residual to obtain multi-scale feature information, and then the input feature information is connected with the multi-scale feature information and sent to the next IRN residual processing, and the original feature information of the point cloud data is retained through the residual connection, which is expressed as: in The output of the residual module is connected to IRN, Connect the input of the residual module to IRN, is the input of the IRN module, For connection operation.

[0013] Furthermore, the encoding and decoding of global features based on the codec includes quantizing the global feature information, using an octree codec to process the geometric information in the global features, using an arithmetic codec to process the attribute information of the global features, and introducing a scale hyper-prior c to capture the scale changes of adjacent points in space to model the probability distribution of point cloud features.

[0014] Furthermore, the prediction of the decoded information includes selecting the first k largest elements for processing to reduce data redundancy and improve calculation efficiency. The first k elements are calculated according to the binary cross entropy loss, and the prediction probability is improved and the reconstruction error is reduced by minimizing the binary cross entropy loss, which is expressed as: Among them, N is the number of point clouds that need to be predicted, is the occupancy of the voxel, where 1 indicates occupancy and 0 indicates unoccupied. is the predicted probability value.

[0015] In the second aspect, a point cloud geometry compression system based on inter-layer residuals and IRN connection residuals comprises: The data acquisition module is configured to acquire point cloud data; The DU residual module is configured to perform feature extraction using a convolutional neural network based on sparse convolution, wherein upsampling after downsampling is performed based on the DU residual mechanism, and geometric reduction is performed with feature information before upsampling to obtain up- and downsampling residuals; a loss function is calculated using the up- and downsampling residuals, and the convolutional neural network is optimized by minimizing the loss function; The IRN connection residual module is configured to capture the extracted features based on the IRN connection residual mechanism to obtain global features; A codec module is configured to encode and decode the global feature based on the codec; The prediction module is configured to predict the decoded information.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals.

[0017] In a fourth aspect, the present invention provides a terminal device including a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the point cloud geometry compression method based on inter-layer residuals and IRN connection residuals.

[0018] In summary, the present invention has the following beneficial technical effects: 1. The present invention calculates the up- and down-sampling residuals based on the DU residual mechanism and uses it to optimize the convolutional neural network, which can reduce the information loss and quantization error in the down-sampling process and effectively constrain the down-sampling accuracy. At the same time, the IRN connection residual mechanism captures global features through multi-layer residual connections, retains the original feature information of the point cloud data, and avoids the gradient vanishing problem. In the experiment, compared with G-PCC (octree), the present invention improves the BD-PSNR by 11.70dB and 11.08dB on average in D1 and D2, respectively. At low bit rates, the reconstructed point cloud boundary area performs better, the dataset contour is clearer and more accurate, and the reconstruction effect is better at the foot of the longdress dataset and the edge of the gun of the soldier dataset.

[0019] 2. In the encoding process, the octree codec is used to process geometric information, and only the subspace containing points is encoded. The arithmetic codec is combined to process attribute information, and the scale hyper-prior c is introduced to model the probability distribution of point cloud features. This method improves the data compression efficiency. In the comparative experiments with G-PCC (octree), G-PCC (trisoup) and PCGC v2, the average BD-Rate gain of the present invention on D1 and D2 is 96% and 93% respectively compared with G-PCC (octree), the average BD-Rate gain of the present invention on G-PCC (trisoup) is 74% and 81% respectively, and the average BD-Rate gain of the present invention on PCGC v2 is 6% and 9% respectively, which reflects the advantages of the present invention in compression efficiency.

[0020] 3. The present invention uses sparse convolution for feature extraction, and only extracts the features of occupied coordinates, avoiding invalid calculations for a large number of empty voxels, reducing computational complexity, and reducing the consumption of space and computing resources. At the same time, the first k largest elements are selected for processing in the prediction stage, reducing data redundancy, improving computing efficiency, and further reducing overall resource consumption, so that the model can run efficiently even in a resource-constrained environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to Embodiment 1 of the present invention; Figure 2 is a schematic diagram of an IRN connection residual module according to Embodiment 1 of the present invention; wherein, Figure 2 A in it is the IRN module. Figure 2 B in is the overall IRN connection residual module; Figure 3 is a schematic diagram of an attention module according to Embodiment 1 of the present invention; DETAILED DESCRIPTION

[0022] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0023] Example 1 Reference Figure 1 , a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals in this embodiment includes: Get point cloud data; A convolutional neural network based on sparse convolution is used for feature extraction, in which upsampling after downsampling is performed based on the DU residual mechanism, and geometric reduction is performed with the feature information before upsampling to obtain up- and downsampling residuals; the up- and downsampling residuals are used to calculate the loss function, and the convolutional neural network is optimized by minimizing the loss function; The extracted features are captured based on the IRN connection residual mechanism to obtain global features; Encode and decode the global features based on the codec; Make predictions on the decoded information.

[0024] Specifically: In the first step, the present invention randomly selected 26342 data sets from the ShapeNet data set for training, and selected two different types of data sets, 8iVFB and Owlii, for testing. For the point cloud data in the above data sets, the present invention inputs the data into a convolutional neural network based on sparse convolution, in which the point cloud data is represented as a set of sparse tensors , including geometric coordinates (x, y, z positive occupancy) and feature attributes (color, opacity, reflectivity, etc.), each using a set of coordinates and related attribute characteristics Among them, sparse convolution only extracts the features of occupancy coordinates. For each voxel , whose eigenvector It is obtained by convolution of the input voxels in its neighborhood. The specific calculation process is as follows: (1) in and are the coordinates of the input and output point clouds, respectively. and are the feature vectors of the input and output point clouds at coordinate c, Represents the definition of a 3D convolution kernel centered at offset i. is the value of the convolution kernel at offset i.

[0025] The present invention adopts sparse convolution to reduce complexity and better extract feature information of point cloud. Each convolution operation in the present invention (i.e., feature aggregation in downsampling and point cloud reconstruction in upsampling) is processed according to the above process.

[0026] In other words, for the input 3D original point cloud data, the present invention first performs a sparse convolution operation to represent the point cloud data in the form of a sparse tensor. , and record the coordinates and related eigenvalues ​​of non-empty voxels. In the convolution process, the sparse convolution kernel It acts on the neighborhood of the input voxel, calculates the feature information of each voxel in the neighborhood, and generates the feature information of the output voxel by weighted summation. Initially, the input data is 1 channel. After the sparse convolution operation, the input 1 channel data is expanded to 16 channels through the convolution kernel, and then activated by the Relu function to further increase the nonlinear features to ensure that the model can learn complex feature representations. Subsequently, we use sparse convolution to perform downsampling operations to reduce the spatial resolution of the features, expanding the feature data from 16 channels to 32 channels while halving the size. However, in the downsampling process, due to the halving of the data, some point cloud information will inevitably be lost.

[0027] In the second step, in order to constrain the accuracy of point cloud downsampling and reduce the quantization error as much as possible, the present invention designs a DU residual module. Perform upsampling to obtain , and the feature information before upsampling Perform geometric reduction to obtain the residual In this way, the module can clearly know the point cloud feature information lost during the downsampling process. The module will then constrain the accuracy of the downsampling process by minimizing the residual to reduce the loss of feature information and the residual introduced by the quantization process, thereby improving the reconstruction effect. The specific process is as follows: in, Represents the coordinate residual result value, and They are respectively expressed as the geometric coordinates of the i-th point before upsampling and downsampling, that is, the coordinate difference of each point in the three-dimensional component can be expressed as ( ), represents the upsampling operation, Represents a downsampling operation.

[0028] The present invention records the above residual result value and does not transmit it. It is mainly used to calculate the loss function. During the training process, the above model is optimized by minimizing the loss. The residual loss calculation process is as follows: in, represents the residual loss value of upsampling and downsampling calculated on the encoder side, is the activation function, M is the number of points, The result value calculated by the formula in claim 3 represents the coordinate residual.

[0029] In the third step, the present invention transmits the downsampled feature information to the IRN connection residual module after Relu activation. First, the module introduces the IRN module, and then uses convolution kernels of different sizes in the IRN module to extract feature information of different scales in parallel. Subsequently, the module stacks the three IRN modules by using multiple layers of residual connections. Specifically, the module first processes the input feature information through an IRN module to obtain multi-scale feature information, and then the module connects the multi-scale feature information with the input to ensure the correlation between them. The operation of the subsequent two IRN modules is similar, but after the processing of the last IRN module is completed, the module retains the original feature information of the point cloud through residual connections. This operation not only ensures the correlation between layers, but also makes good use of the spatial correlation between blocks, while also avoiding the problem of gradient disappearance between blocks. The specific operation steps are as follows: in The output of the residual module is connected to IRN, Connect the input of the residual module for IRN. is the input of the IRN module, For connection operation.

[0030] In the fourth step, the following two downsampling operations are performed in the same manner. In the last convolution operation at the encoder end, the present invention converts the number of channels to 8 channels for encoding operation.

[0031] In the fifth step, the present invention uses an octree codec to process geometric information by recursively dividing the 3D space into a plurality of eight subspaces for processing, wherein only the subspace containing the point is coded. During the coding process, the octree codec uses a binary bit string to represent the octree structure, that is, each space uses 8 bits to represent the occupancy of the subspace. If the bit number is 1, it indicates that the subspace has point cloud data. If it is 0, it indicates that the subspace is empty and the storage is skipped. Finally, after coding, a compact bit stream is obtained to represent the geometric information of the stored and transmitted point cloud. Subsequently, during the decoding process, the inverse process of the coding operation is used to restore the geometric structure of the point cloud, and the point cloud is accurately restored by reading the 8-bit information and combining it with the position information of the point.

[0032] The present invention uses arithmetic coding to process attribute information. Before arithmetic coding, the present invention introduces a quantization operation to achieve more efficient data storage and representation by adding uniform noise to the input tensor. The specific expression is as follows: in The uniform noise, and y represent the quantized feature information and the original feature information respectively. Follow Uniform distribution with center .

[0033] Subsequently, the present invention inputs the quantized feature information into the arithmetic codec for processing. The present invention uses a non-parametric fully factorized probability density model to encode the quantized feature information. At the same time, the present invention introduces a scale hyper-prior c to capture the scale changes of adjacent points in space, so as to more accurately model the probability distribution of point cloud features. In short, the present invention uses the Laplace distribution L to approximate The probability density function of , the calculation formula is as follows: in, and are each under the scale hyper-prior c model The mean and mean square error of .

[0034] Specifically, the core idea of ​​arithmetic coding is to use the probability distribution of data to assign shorter bit streams to high-probability events and longer bit streams to low-probability events. , whose probability model is approximated by Laplace distribution. The first step is to set the initial interval , read the quantized eigenvalue of the first symbol The second step is to read the quantitative features from the probability model The cumulative distribution function (CDF) of is: , and then determine the quantitative feature in Subintervals within: ; , so the quantized features Occupies the interval This part. Step 3, for each , repeat the second step above, gradually reduce The final interval after processing represents the complete point cloud feature data. The fourth step is to select any real number in the interval and convert it into a binary bit stream. In the arithmetic decoding process, the first step is to convert the binary bit stream into a real number. , and maps it to the initial interval On. The second step is to reverse the sign (eigenvalue) through the cumulative distribution function (CDF) to find the one that satisfies; of , that is, determine the symbol currently decoded. The third step is to update the interval range according to the interval calculation method in the second step of encoding, and set it as the initial interval of the next symbol to decode the next symbol. Iterate in sequence until all point cloud features are restored.

[0035] Step 6: After the above encoding and decoding operations, the present invention restores the decoded feature information through an upsampling operation, and uses the Rule function for activation and expansion of the feature information. Similarly, in order to extract multi-scale feature information, the present invention also introduces an IRN residual connection module on the decoder side. In order to enhance the decoder's attention to salient features, the present invention also introduces an attention module. This module uses two point-by-point multiplication weighted operations to capture the global dependency range by directly calculating the relationship between the positions of scattered points, thereby further enhancing the interactivity between features. Finally, the module retains the original point cloud information through residual connections while increasing the fluidity of the gradient. The specific operations of this module are expressed as: in is the output of this module, is the input of this module.

[0036] Step 7: Perform a sparse convolution operation on the feature information recovered after the attention module, and convert its output channel to 1, that is, each voxel outputs only one value, indicating the probability that the voxel belongs to a certain category. The present invention prunes voxels according to the occupancy of the voxels. In the voxel pruning stage of hierarchical reconstruction, the present invention sorts the result probabilities, and considers the first k voxels as the most likely occupied voxels, and performs binary classification through sparse convolution with output channel 1. The subsequent two upsampling operations are the same as the above operations, and finally the output is performed to obtain compressed and reconstructed point cloud data.

[0037] However, in order to continuously optimize the model during the training process and reduce the gap between the original point cloud and the reconstructed point cloud, the present invention adopts a binary cross entropy loss function to improve the prediction probability of the occupancy information of the reconstructed point cloud by minimizing the loss function. The prediction probability is to map the feature vector finally output by the decoder end to a probability value through the Relu activation function, which represents the occupancy of the voxel in the space. Then, adaptive threshold classification is performed according to the number of occupied points in the original cube, and it is classified as binary 0 or 1.

[0038] In the prediction process, the present invention selects the first k largest elements for processing to reduce data redundancy and improve calculation efficiency. The first k elements are calculated according to the binary cross entropy loss in the following formula to improve the prediction probability and reduce the reconstruction error by minimizing the binary cross entropy loss:

[0039] Where N is the number of point clouds that need to be predicted, is the occupancy of the voxel, where 1 indicates occupancy and 0 indicates unoccupied. is the predicted probability value.

[0040] Experimental verification: This paper randomly selected 26342 data sets from the ShapeNet data set for training. The Adam optimizer was used in the experiment. The initial learning rate was set to 0.0008. The learning rate was adjusted by halving in each round of training until it was reduced to 0.00001. The batch size was set to 16. According to experimental experience, The value of is between 0.25 and 5. The value of is set to 10. This experiment is run on a computer with an NVIDIA RTX 4080 GPU (16G) and an i7 13790 (32G).

[0041] Objective comparison: This paper selects two different types of data sets, 8iVFB and Owlii, for testing, and uses point-to-point geometric error (D1 PSNR) and point-to-plane geometric error (D2 PSNR) to evaluate the reconstruction quality of point clouds, and uses Bits Per Point (bpp) to measure the storage efficiency after compression. In order to better verify the performance of the model proposed by this paper, this paper compares the proposed method with G-PCC (octree), G-PCC (trisoup) and PCGC v2.

[0042] For fair comparison, the present invention follows the recommendations of MPEG Common Test Condition (CTC) and applies similar bit ranges to G-PCC (octree), G-PCC (trisoup), PCGC v2 and the method proposed by the present invention. For G-PCC, the present invention uses the latest TMC13-v23.0-rc2, and the parameter settings follow CTC. For G-PCC (octree), the present invention sets positionQuantizationScale between 0.15 and 0.5. For G-PCC (trisoup), the present invention sets trisoupNodeSizeLog2 to 2, 3, and 4. Other parameters remain unchanged by default.

[0043] Experimental results show that the average BD-Rate gains of the method of the present invention on D1 and D2 are 96% and 93% respectively compared with G-PCC (octree), 74% and 81% respectively compared with G-PCC (trisoup), and 6% and 9% respectively compared with the most advanced PCGC v2. The method of the present invention also shows better reconstruction quality in the low bit rate range, with an average BD-PSNR improvement of 11.70dB and 11.08dB on D1 and D2 compared with G-PCC (octree), an average BD-PSNR improvement of 3.15dB and 4.66dB on D1 and D2 compared with G-PCC (trisoup), and an average BD-PSNR improvement of 0.22dB and 0.37dB on D1 and D2 compared with the most advanced PCGC v2.

[0044] The present invention visually compares the point cloud reconstructed by the proposed method with the point cloud reconstructed by the G-PCC (octree) and PCGC v2 methods. At a lower bit rate, for the feet of the longdress dataset, the proposed method reconstructs a clearer outline of the left foot and more complete details behind the right foot. For the soldier dataset, the proposed method pays more attention to the edge of the gun. In short, despite the lower bit rate, the proposed method performs better in the boundary area, making the dataset outline clearer and more accurate.

[0045] Example 2 This embodiment provides a point cloud geometry compression system based on inter-layer residuals and IRN connection residuals, including: The data acquisition module is configured as follows: A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals.

[0046] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor for a point cloud geometry compression method based on inter-layer residuals and IRN connection residuals.

[0047] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A point cloud geometry compression method based on inter-layer residuals and IRN connection residuals, characterized in that: include: Get point cloud data; A convolutional neural network based on sparse convolution is used for feature extraction, in which upsampling after downsampling is performed based on the DU residual mechanism, and geometric reduction is performed with the feature information before upsampling to obtain up- and downsampling residuals; the up- and downsampling residuals are used to calculate the loss function, and the convolutional neural network is optimized by minimizing the loss function; The extracted features are captured based on the IRN connection residual mechanism to obtain global features; Encode and decode the global features based on the codec; Make predictions on the decoded information.

2. A point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to claim 1, characterized in that: The feature extraction is performed using a convolutional neural network based on sparse convolution, including first performing a sparse convolution operation on the input 3D original point cloud data to expand the input 1-channel data to 16 channels, then activating it with a Relu function, and performing a downsampling operation using sparse convolution to expand the feature data from 16 channels to 32 channels while halving the size.

3. The point cloud geometry compression method based on inter-layer residual and IRN connection residual according to claim 2, characterized in that: The upsampling after downsampling based on the DU residual mechanism is geometrically reduced with the feature information before upsampling to obtain the upsampling and downsampling residuals, including the downsampling feature information. Perform upsampling to obtain , and the feature information before upsampling Perform geometric reduction to obtain the residual , expressed as: in, Represents the coordinate residual result value, and They are respectively represented as the geometric coordinates of the i-th point before upsampling and downsampling, that is, the coordinate difference of each point in the three-dimensional component, expressed as ( ), represents the upsampling operation, Represents a downsampling operation.

4. The point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to claim 3, characterized in that: The above method uses the up- and down-sampling residuals to calculate the loss function, and optimizes the convolutional neural network by minimizing the loss function, including using the up- and down-sampling residual result values ​​to calculate the loss function. During the training process, the model is optimized by minimizing the loss function, which is expressed as: in, represents the residual loss value of upsampling and downsampling calculated on the encoder side, is the activation function, M is the number of points, The result value calculated by the formula in claim 3 represents the coordinate residual.

5. The point cloud geometry compression method based on inter-layer residual and IRN connection residual according to claim 4, characterized in that: The extracted features are captured based on the IRN connection residual mechanism to obtain global features, including stacking three IRN residuals using a multi-layer residual connection method, wherein the feature information is first processed by an IRN residual to obtain multi-scale feature information, and then the input feature information is connected with the multi-scale feature information and sent to the next IRN residual processing, and the original feature information of the point cloud data is retained through the residual connection, which is expressed as: in The output of the residual module is connected to IRN, Connect the input of the residual module to IRN, is the input of the IRN module, For connection operation.

6. The point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to claim 5, characterized in that: The encoding and decoding of global features based on the codec includes quantizing the global feature information, using the octree codec to process the geometric information in the global features, using the arithmetic codec to process the attribute information of the global features, and introducing the scale hyper-prior c to capture the scale changes of adjacent points in space to model the probability distribution of point cloud features.

7. The point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to claim 6, characterized in that: The encoding and decoding of global features based on the encoder and decoder also includes using two point-by-point multiplication weighted operations based on the attention mechanism to capture the global dependency range by directly calculating the relationship between the positions of scattered points, thereby enhancing the interactivity between features. Finally, the residual connection is used to retain the original point cloud information while increasing the fluidity of the gradient, which is expressed as: in is the output of this module, is the input of this module.

8. The point cloud geometry compression method based on inter-layer residuals and IRN connection residuals according to claim 7, characterized in that: The predicting of the decoded information includes performing a sparse convolution operation on the decoded feature information. The predicted value of the voxel category is output by convolution based on the feature map of the previous layer, including the occupied and unoccupied conditions of the voxel; then the top k voxels with the highest scores are retained to ensure the performance of the model.

9. The point cloud geometry compression method based on inter-layer residual and IRN connection residual according to claim 7, characterized in that: The prediction of the decoded information also includes selecting the first k largest elements for processing to reduce data redundancy and improve calculation efficiency. The first k elements are calculated according to the binary cross entropy loss. The prediction probability is improved and the reconstruction error is reduced by minimizing the binary cross entropy loss, which is expressed as: Among them, N is the number of point clouds that need to be predicted, is the occupancy of the voxel, where 1 indicates occupancy and 0 indicates unoccupied. is the predicted probability value.

10. A point cloud geometry compression system based on inter-layer residuals and IRN connection residuals, characterized in that: include: The data acquisition module is configured to acquire point cloud data; The DU residual module is configured to perform feature extraction using a convolutional neural network based on sparse convolution, wherein upsampling after downsampling is performed based on the DU residual mechanism, and geometric reduction is performed with feature information before upsampling to obtain up- and downsampling residuals; a loss function is calculated using the up- and downsampling residuals, and the convolutional neural network is optimized by minimizing the loss function; The IRN connection residual module is configured to capture the extracted features based on the IRN connection residual mechanism to obtain global features; A codec module is configured to encode and decode the global feature based on the codec; The prediction module is configured to predict the decoded information.

Images