A video compression reconstruction method, device and equipment

Abstract

The application provides a video compression reconstruction method, device and equipment, relates to the technical field of data services, and comprises the following steps: acquiring video data and a first parameter data set of a video service related to the video data, wherein the first parameter data set comprises a key quality indicator (KQI) parameter data set and / or a key performance indicator (KPI) parameter data set; performing importance measurement on various parameters in the first parameter data set; and based on the measurement result, performing compression reconstruction on the video data to obtain a reconstructed image. In this way, the video data is compressed and reconstructed based on the KQI parameter and the KPI parameter, the content of the video image itself is fully considered during compression and reconstruction, the effect of the video image transmitted at a low code rate is ensured, and the phenomenon of blurring is avoided.

Description

Technical Field

[0001] This application relates to the field of data service technology, and in particular to a video compression and reconstruction method, apparatus and equipment. Background Technology

[0002] Traditional deep learning-based image compression methods are mostly lossy compression methods, typically described as a joint rate-distortion optimization problem involving the learning encoder, quantizer, and decoder, such as JPEG and JPEG2000. While these traditional compression methods can guarantee high-quality transmission, they neglect the image content itself, resulting in poor image quality at low bitrates and the appearance of blurry images. Summary of the Invention

[0003] This application provides a video compression and reconstruction method, apparatus, and device, which solves the problem that existing compression methods, by ignoring the content of the image itself, result in poor image quality or even blurring when transmitting at low bit rates.

[0004] Firstly, in order to achieve the above objectives, embodiments of this application provide a video compression and reconstruction method, comprising:

[0005] Acquire video data and a first parameter dataset of video services related to the video data, wherein the first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset;

[0006] The importance of each type of parameter in the first parameter dataset is measured.

[0007] Based on the measurement results, the video data is compressed and reconstructed to obtain a reconstructed image.

[0008] Optionally, the step of measuring the importance of various parameters in the first parameter dataset includes:

[0009] The first parameter dataset is input into the random forest classification module to obtain the importance order of each type of parameter in the first parameter dataset.

[0010] Optionally, the random forest classification module is used for:

[0011] In each base learner of the random forest classification module, a set of target nodes using target class parameters is determined, wherein the target class parameters are any class parameters in the first parameter dataset;

[0012] Measure the importance of the target class parameter to each node in the target node set;

[0013] Based on the importance of the target class parameters to each of the nodes, determine the importance of the target class parameters to the base learner in which the target node set is located;

[0014] The importance of the target class parameters to the random forest classification module is determined based on the importance of the target class parameters to each of the base learners.

[0015] The importance of each parameter to the random forest classification module is sorted, and the order of importance of each parameter in the first parameter data is output.

[0016] Optionally, the step of reconstructing the video data based on the measurement results to obtain a reconstructed image includes:

[0017] The video data, the first parameter dataset, and the metric results are input into a mask compression reconstruction network to obtain the reconstructed image; wherein, the metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

[0018] Optionally, the mask compression reconstruction network includes an encoder, a mask compression module, and a decoder; wherein:

[0019] The encoder is used to: encode the video data according to the first parameter dataset and the measurement result, and output the latent representation of the video data;

[0020] The mask compression module is used to: transform and compress the latent representation using a mask; wherein the mask is related to channel correlation;

[0021] The decoder is used to: reconstruct the image based on the transformed and compressed latent representation to obtain the reconstructed image.

[0022] Optionally, the mask compression reconstruction network further includes a classifier module for outputting classification labels.

[0023] Optionally, the method further includes:

[0024] When training the mask compression reconstruction network, a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image are obtained.

[0025] The loss function is determined based on the various parameters in the second parameter dataset, the various parameters in the third parameter dataset, the order of importance of the various parameters in the second parameter dataset, and the various parameters and their order of importance in the third parameter dataset.

[0026] Backpropagation is performed based on the loss function to adjust the parameters of the mask compression reconstruction network.

[0027] Secondly, in order to achieve the above objectives, embodiments of this application also provide a video compression and reconstruction apparatus, comprising:

[0028] The first acquisition module is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset.

[0029] The metric module is used to measure the importance of various parameters in the first parameter dataset;

[0030] The reconstruction module is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image.

[0031] Optionally, the metric module is used to: input the first parameter dataset into the random forest classification module to obtain the importance order of various parameters in the first parameter dataset.

[0032] Optionally, the reconstruction module is used for:

[0033] The video data, the first parameter dataset, and the metric results are input into a mask compression reconstruction network to obtain the reconstructed image; wherein, the metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

[0034] Optionally, the device further includes:

[0035] The second acquisition module is used to acquire, during the training of the mask compression reconstruction network, a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image.

[0036] The determination module is used to determine the loss function based on the various parameters in the second parameter dataset, the various parameters in the third parameter dataset, the importance order of the various parameters in the second parameter dataset, and the various parameters and importance order in the third parameter dataset.

[0037] An adjustment module is used to perform backpropagation based on the loss function and adjust the parameters of the mask compression reconstruction network.

[0038] Thirdly, in order to achieve the above objectives, embodiments of this application also provide a video compression and reconstruction device, including a transceiver and a processor;

[0039] The transceiver is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset.

[0040] The processor is used to measure the importance of various parameters in the first parameter dataset;

[0041] The processor is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image.

[0042] Fourthly, in order to achieve the above objectives, embodiments of this application also provide a video compression and reconstruction device, including: a transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; when the processor executes the program or instructions, it implements the video compression and reconstruction method as described in the first aspect.

[0043] Fifthly, in order to achieve the above-mentioned objectives, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the video compression and reconstruction method as described in the first aspect.

[0044] The beneficial effects of the above technical solution in this application are as follows:

[0045] The video compression and reconstruction method of this application, according to embodiments, firstly acquires video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset. Secondly, it measures the importance of various parameters in the first parameter dataset. Thirdly, based on the measurement results, it compresses and reconstructs the video data to obtain a reconstructed image. This achieves video data compression and reconstruction based on the relevant content of the KQI parameter data and / or KPI parameter data, ensuring that the content of the video image itself is fully considered during video compression and reconstruction, thereby ensuring the quality of video images transmitted at low bitrates and avoiding blurring. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the video compression and reconstruction method according to an embodiment of this application.

[0047] Figure 2 This is a schematic diagram of the structure of the video compression and reconstruction apparatus according to an embodiment of this application;

[0048] Figure 3 This is one of the structural schematic diagrams of the video compression and reconstruction device according to an embodiment of this application;

[0049] Figure 4 This is a second schematic diagram of the structure of the video compression and reconstruction device according to an embodiment of this application. Detailed Implementation

[0050] To make the technical problems, technical solutions and advantages of this application clearer, a detailed description will be provided below in conjunction with the accompanying drawings and specific embodiments.

[0051] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0052] In the various embodiments of this application, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0053] In addition, the terms "system" and "network" are often used interchangeably in this article.

[0054] In the embodiments provided in this application, it should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0055] like Figure 1 As shown, this application provides a video compression and reconstruction method, including:

[0056] Step 101: Obtain video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicators (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset.

[0057] In this step, the KQI parameter dataset and KPI parameter dataset can be a collection of existing parameters, such as video bitrate, buffer latency, peak video buffer bitrate, number of video rejections, total video rejection time, average video rate during rejection duration, etc.

[0058] Step 102: Measure the importance of each type of parameter in the first parameter dataset;

[0059] In this step, when the first parameter dataset includes the KQI parameter dataset, the various parameters are different types of parameters in the KQI parameter dataset; when the first parameter dataset only includes the KPI parameter dataset, the various parameters are different types of parameters in the KPI parameter dataset; and when the first parameter dataset includes both the KQI parameter dataset and the KPI parameter dataset, the various parameters are different types of parameters in the KQI parameter dataset and the KPI parameter dataset, respectively.

[0060] Step 103: Based on the measurement results, compress and reconstruct the video data to obtain the reconstructed image.

[0061] It should be noted that in actual user experience quality monitoring scenarios, the KPI parameter dataset and KQI parameter dataset of video services can truly reflect the user experience quality. Therefore, operators hope to evaluate the importance of different KPI parameter data and / or KQI parameter data in order to compress and reconstruct video data based on the evaluation results, so that the reconstructed image can reflect the content of the video itself and improve the quality of the reconstructed video image.

[0062] The video compression and reconstruction method of this application embodiment first acquires video data and a first parameter dataset of video services related to the video data. This first parameter dataset includes a KQI parameter dataset and / or a KPI parameter dataset. Second, it measures the importance of various parameters in the first parameter dataset. Third, based on the measurement results, it compresses and reconstructs the video data to obtain a reconstructed image. In this way, it achieves video data compression and reconstruction based on the importance measurement results of KQI parameter data and / or KPI parameter data. This ensures that the content of the video image itself is fully considered during video compression and reconstruction, thereby ensuring the quality of the video image transmitted at low bitrates, improving the quality of the reconstructed video image, and avoiding blurring.

[0063] As an optional implementation, step 102 involves measuring the importance of various parameters in the first parameter dataset, including:

[0064] The first parameter dataset is input into the random forest classification module to obtain the importance order of each type of parameter in the first parameter dataset.

[0065] In other words, in this embodiment, the importance of various parameters in the first parameter dataset can be measured using a random forest classification module. Random forest is an algorithm that integrates multiple trees using the idea of ​​ensemble learning. The basic unit of random forest is a decision tree, and random forest essentially belongs to a major branch of machine learning (ensemble learning methods). The random forest classification module uses the Variable Importance Measurement (VIM) algorithm to measure the importance of different parameters for model prediction at different resolution levels.

[0066] Here, it should be noted that a common method for interpreting variable importance in random forests is to use the number of times a variable is used in each base learner as a measure of its importance. Another method for measuring variable importance is the change in the splitting metric used by each variable in relation to the branches of the base learners. This method first determines the type of base learner, such as ID3, C4.5, or CART decision tree. Then, based on the splitting metric used in the decision tree, such as information gain, information gain ratio, or Gini coefficient, it uses the mean or weighted sum of the changes in the splitting metric across all base learners as a measure of variable importance. This method can be integrated into the modeling process and can relatively accurately quantify the importance of different variables in random forest modeling. Based on this, the random forest classification module in this embodiment is specifically used for:

[0067] (1) In each base learner of the random forest classification module, determine the target node set using the target class parameter. The target class parameter is any class parameter in the first parameter dataset. As mentioned above, a base learner is a decision tree, and a decision tree has multiple nodes.

[0068] (2) Measure the importance of the target class parameters to each node in the target node set;

[0069] (3) Determine the importance of the target class parameters to the base learner of the target node set based on the importance of the target class parameters to each node;

[0070] (4) Determine the importance of the target class parameters to the random forest classification module based on the importance of the target class parameters to each base learner;

[0071] (5) Sort the importance of each type of parameter to the random forest classification module and output the order of importance of each type of parameter in the first parameter data.

[0072] In other words, in this embodiment of the application, when using the random forest classification module to measure the importance of various parameters, the importance of each parameter to each node can be determined first. Then, based on the importance to each node, the importance to each base learner can be determined. After that, based on the importance to each base learner, the importance to the random forest can be determined. Finally, the importance of each parameter to the random forest can be ranked to obtain the final importance order. The implementation process of the random forest classification module is described below:

[0073] A) Define the input parameters for the random forest classification module:

[0074] Let the first parameter dataset be defined as: X1, X2, ..., X C Assuming the random forest has n base learners, and the base learner i uses parameter X... j The set of nodes is M;

[0075] B) The importance of each parameter in the first parameter dataset output by the random forest classification module:

[0076] X j The importance of a parameter in node m is denoted as: It can be calculated using the following formula:

[0077]

[0078] Among them, Gini m It is the Gini coefficient of node m. r and Gini l These are the Gini coefficients after branching at node m;

[0079] For the base learner i, the parameter X j Its importance in the modeling process can be calculated using the following formula:

[0080]

[0081] Parameter X j In random forests, importance is denoted as VIM. j VIM j It can be calculated using the following formula:

[0082]

[0083] Parameter X j Importance is denoted as: IR j IR j It can be calculated using the following formula:

[0084]

[0085] In other words, in this implementation, when sorting the importance of various parameters, the sorting can be based on the proportion of importance of each parameter.

[0086] It's important to note that in real-world user experience quality monitoring scenarios, operators not only aim to accurately perceive experience quality through model building, but also to assess the importance of different video KQI and network KPI parameters. This allows for adaptive allocation of network bandwidth at the sending end, rapid identification of the cause of experience quality degradation, and targeted application of network optimization strategies to improve the user's video experience and prevent user churn. Therefore, this optional implementation provides a method for evaluating the importance of KQI and / or KPI parameters to address the aforementioned issues.

[0087] The method for measuring the importance of various parameters in this optional implementation avoids the problem that existing importance measures require A / B testing with a corresponding platform to achieve control and experimental groups. At the same time, since A / B testing requires that the user attributes of the control and experimental groups be basically similar, it is necessary to use a user profiling system to screen users to determine the user composition of the control and experimental groups. However, the collection of user profiling data is extremely difficult and it is hard to meet the experimental conditions required for A / B testing.

[0088] As an optional implementation, step 103 involves reconstructing the video data based on the measurement results to obtain a reconstructed image, including:

[0089] The video data, the first parameter dataset, and the metric results are input into the mask compression reconstruction network to obtain the reconstructed image. The metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

[0090] In this optional implementation, the mask compression and reconstruction network first determines the weights of various parameter data based on the measurement results. Among them, the parameters with higher importance ranking have higher weights. Then, the video data is compressed and reconstructed based on the weights of various parameter data, realizing the compression and reconstruction of video data based on the user's subjective experience. In this way, the reconstructed video image fully considers the content of the video data itself, avoiding the problem of poor image quality or even blurring at low bitrates.

[0091] As a specific implementation, the mask compression reconstruction network includes an encoder, a mask compression module, and a decoder; where:

[0092] The encoder is used to: encode video data based on the first parameter dataset and the measurement results, and output the latent representation of the video data; specifically, the encoder module configures weights for various parameters in the first parameter dataset based on the measurement results, and then encodes the video data based on the weights of various parameters to output the latent representation W of the video data, where the latent representation can also be called the feature vector W of the video data.

[0093] The mask compression module is used to: transform and compress latent representations using masks; wherein, the mask is related to channel correlation; specifically, the mask compression module maps the feature vector W of each channel of the video data to a real number ρW through a continuous function ρ(·), and then uses binary segmentation with a single threshold on ρW to obtain the corresponding mask Ω. ij Finally, the encoded feature vector is further compressed based on the mask.

[0094] Here, it should be noted that channel correlation is used to measure ρ(W), where, as mentioned earlier, ρ(·) is a continuous function of the encoded latent representation W. The channel correlation measurement is performed individually at each location, specifically by mapping the C channels to a real number at each location, ultimately representing the channel correlation as ρW. Then, the corresponding mask Ω can be obtained by binary segmentation of ρW using a single threshold. ij The embodiments in this application employ variance, which is calculated across channels. Based on this, the latent representation can be further transformed into:

[0095]

[0096] Among them, (w C ) ij The latent representation is the feature vector after channel C is compressed. The variance effectively estimates the redundancy and reconstruction loss. Using this method, mask compression tends to compress locations containing a large amount of redundancy while causing almost no reconstruction loss. Thus, it surpasses typical video coding at significantly low bit rates, resulting in better image transmission at low bit rates.

[0097] The decoder is used to: reconstruct the image based on the transformed and compressed latent representation to obtain the reconstructed image. Specifically, the decoder is used to generate realistic samples using the latent representation so that the reconstructed image is a high-resolution image that is subjectively visually appealing.

[0098] Furthermore, the mask compression reconstruction network also includes a classifier module, which outputs classification labels. In other words, the classifier module predicts the image classification probability.

[0099] It should be noted that traditional compression schemes mainly consist of an encoder and a decoder. Based on deep learning methods, Generative Adversarial Networks (GANs) mainly consist of a generator and a discriminator. In the embodiments of this application, the encoder, mask compression module, decoder, and classifier constitute the generator. That is, the generator is a convolutional neural network that compresses the video through latent representations and reconstructs it using latent representations.

[0100] Furthermore, as an optional implementation, the method also includes:

[0101] (1) When training the mask compression reconstruction network, a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image are obtained.

[0102] It should be noted that the second and third parameter datasets have the same types of parameters as the first parameter dataset. That is, during model training, firstly, the video training data and the second parameter dataset related to the video training data are obtained. Secondly, the importance of each type of parameter data in the second parameter dataset is measured using the random forest classification module. Thirdly, the mask compression and reconstruction network to be trained compresses and reconstructs the video training data based on the importance ranking of the second parameter dataset to obtain the reconstructed training image, and further obtains the third parameter dataset related to the reconstructed training image.

[0103] (2) Determine the loss function based on the various parameters in the second parameter dataset, the various parameters in the third parameter dataset, the order of importance of the various parameters in the second parameter dataset, and the various parameters and their order of importance in the third parameter dataset;

[0104] For example, this step specifically involves: calculating the difference between similar data in the second parameter dataset and the third dataset, assigning weights to the differences based on the importance of each parameter, and using these weights as the loss function. The higher the importance of a parameter, the higher the weight of its difference.

[0105] (3) Perform backpropagation based on the loss function and adjust the parameters of the mask compression reconstruction network.

[0106] In this optional implementation, the reconstruction quality results are analyzed based on the user's subjective experience, and the parameters of the mask compression reconstruction network are optimized based on the analysis results, thereby effectively reducing the amount of data transmitted without compromising the user experience.

[0107] In this embodiment, a mask compression reconstruction network is used for mask compression suitable for semantic tasks. Specifically, the channel redundancy of the latent representation generated by the encoder can be effectively compressed by mask compression. Mask compression quantifies the channel variance of the latent identifier instead of training multiple models. This achieves collaborative compression and classification through end-to-end learning and effectively saves bandwidth resources. In addition, the mask compression reconstruction method suitable for semantic tasks is used for image reconstruction and then for reverse verification of network parameters and input quality. The optimization problem can be solved by using the Lagrangian function and iteratively accelerated gradient descent algorithm based on the reverse verification results.

[0108] like Figure 2 As shown in the illustration, this application also provides a video compression and reconstruction apparatus, comprising:

[0109] The first acquisition module 201 is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a key quality indicator (KQI) parameter dataset and / or a key performance indicator (KPI) parameter dataset.

[0110] The measurement module 202 is used to measure the importance of various parameters in the first parameter dataset;

[0111] The reconstruction module 203 is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image.

[0112] Optionally, the metric module 202 is used to: input the first parameter dataset into the random forest classification module to obtain the importance order of various parameters in the first parameter dataset.

[0113] Optionally, the random forest classification module is used for:

[0114] In each base learner of the random forest classification module, a set of target nodes using target class parameters is determined, wherein the target class parameters are any class parameters in the first parameter dataset;

[0115] Measure the importance of the target class parameter to each node in the target node set;

[0116] Based on the importance of the target class parameters to each of the nodes, determine the importance of the target class parameters to the base learner in which the target node set is located;

[0117] The importance of the target class parameters to the random forest classification module is determined based on the importance of the target class parameters to each of the base learners.

[0118] The importance of each parameter to the random forest classification module is sorted, and the order of importance of each parameter in the first parameter data is output.

[0119] Optionally, the reconstruction module 203 is used for:

[0120] The video data, the first parameter dataset, and the metric results are input into a mask compression reconstruction network to obtain the reconstructed image; wherein, the metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

[0121] Optionally, the mask compression reconstruction network includes an encoder, a mask compression module, and a decoder; wherein:

[0122] The encoder is used to: encode the video data according to the first parameter dataset and the measurement result, and output the latent representation of the video data;

[0123] The mask compression module is used to: transform and compress the latent representation using a mask; wherein the mask is related to channel correlation;

[0124] The decoder is used to: reconstruct the image based on the transformed and compressed latent representation to obtain the reconstructed image.

[0125] Optionally, the mask compression reconstruction network further includes a classifier module for outputting classification labels.

[0126] Optionally, the device further includes:

[0127] The second acquisition module is used to acquire, during the training of the mask compression reconstruction network, a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image.

[0128] The determination module is used to determine the loss function based on the various parameters in the second parameter dataset, the various parameters in the third parameter dataset, the importance order of the various parameters in the second parameter dataset, and the various parameters and importance order in the third parameter dataset.

[0129] An adjustment module is used to perform backpropagation based on the loss function and adjust the parameters of the mask compression reconstruction network.

[0130] It should be noted that the video compression and reconstruction apparatus provided in this application embodiment can implement all the method steps implemented in the above video compression and reconstruction method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0131] like Figure 3 As shown, this application embodiment also provides a video compression and reconstruction device, including a transceiver 310 and a processor 300;

[0132] The transceiver 310 is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset.

[0133] The processor 300 is used to measure the importance of various parameters in the first parameter dataset;

[0134] The processor 300 is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image.

[0135] Optionally, when the processor 300 performs importance measurement on various parameters in the first parameter dataset, it is specifically used for:

[0136] The first parameter dataset is input into the random forest classification module to obtain the importance order of each type of parameter in the first parameter dataset.

[0137] Optionally, the random forest classification module is used for:

[0138] In each base learner of the random forest classification module, a set of target nodes using target class parameters is determined, wherein the target class parameters are any class parameters in the first parameter dataset;

[0139] Measure the importance of the target class parameter to each node in the target node set;

[0140] Based on the importance of the target class parameters to each of the nodes, determine the importance of the target class parameters to the base learner in which the target node set is located;

[0141] The importance of the target class parameters to the random forest classification module is determined based on the importance of the target class parameters to each of the base learners.

[0142] The importance of each parameter to the random forest classification module is sorted, and the order of importance of each parameter in the first parameter data is output.

[0143] Optionally, when the processor 300 reconstructs the video data based on the measurement results to obtain a reconstructed image, it specifically performs the following:

[0144] The video data, the first parameter dataset, and the metric results are input into a mask compression reconstruction network to obtain the reconstructed image; wherein, the metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

[0145] Optionally, the mask compression reconstruction network includes an encoder, a mask compression module, and a decoder; wherein:

[0146] The encoder is used to: encode the video data according to the first parameter dataset and the measurement result, and output the latent representation of the video data;

[0147] The mask compression module is used to: transform and compress the latent representation using a mask; wherein the mask is related to channel correlation;

[0148] The decoder is used to: reconstruct the image based on the transformed and compressed latent representation to obtain the reconstructed image.

[0149] Optionally, the mask compression reconstruction network further includes a classifier module for outputting classification labels.

[0150] Optionally, the transceiver 310 is further configured to, when training the mask compression reconstruction network, acquire a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image.

[0151] The processor 300 is further configured to: determine a loss function based on various parameters in the second parameter dataset, various parameters in the third parameter dataset, the order of importance of various parameters in the second parameter dataset, and various parameters and their order of importance in the third parameter dataset;

[0152] The processor 300 is further configured to: perform backpropagation based on the loss function and adjust the parameters of the mask compression reconstruction network.

[0153] It should be noted that the video compression and reconstruction device provided in this application embodiment can implement all the method steps implemented in the above video compression and reconstruction method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0154] like Figure 4 As shown, this application embodiment also provides a video compression and reconstruction device, including a transceiver 410, a processor 400, a memory 420, and a program or instructions stored in the memory 420 and executable on the processor 400; when the processor 400 executes the program or instructions, it implements the above-described video compression and reconstruction method.

[0155] The transceiver 410 is used to receive and send data under the control of the processor 400.

[0156] Among them, Figure 4 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 400 and memory represented by memory 420. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. Transceiver 410 can be multiple elements, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium. For different video compression and reconstruction devices, the user interface 430 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0157] The processor 400 is responsible for managing the bus architecture and general processing, while the memory 420 can store the data used by the processor 400 when performing operations.

[0158] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the steps in the video compression and reconstruction method described above and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0159] The processor mentioned above is the processor in the video compression and reconstruction device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0160] It should be further noted that the electronic devices described in this specification include, but are not limited to, smartphones, tablets, etc., and many of the described functional components are referred to as modules in order to more specifically emphasize the independence of their implementation.

[0161] In this embodiment, the module can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.

[0162] In practice, an executable code module can be a single instruction or many instructions, and can even be distributed across multiple different code segments, different programs, and across multiple memory devices. Similarly, operational data can be identified within the module and can be implemented in any suitable form and organized within any suitable type of data structure. This operational data can be collected as a single dataset or distributed across different locations (including different storage devices), and can exist, at least in part, solely as electronic signals within the system or network.

[0163] When a module can be implemented using software, considering the current level of hardware technology, modules that can be implemented in software can be implemented using hardware circuits by those skilled in the art to achieve the corresponding functions, without considering cost. These hardware circuits include conventional very-large-scale integrated circuits (VLSI) or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules can also be implemented using programmable hardware devices, such as field-programmable gate arrays, programmable array logic, and programmable logic devices.

[0164] The exemplary embodiments described above are with reference to the accompanying drawings. Many different forms and embodiments are feasible without departing from the spirit and teachings of this application. Therefore, this application should not be construed as limiting the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make this application complete and convey the scope of this application to those skilled in the art. In these drawings, component dimensions and relative dimensions may be exaggerated for clarity. The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. As used herein, unless clearly indicated otherwise, the singular forms “a,” “an,” and “the” are intended to include all such forms. It will be further understood that the terms “comprising” and / or “including”, when used in this specification, indicate the presence of the stated features, integers, steps, operations, components, and / or elements, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, and / or groups thereof. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of the range and any subranges in between.

[0165] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method of video compression reconstruction, characterized in that, The method comprises: obtaining video data and a first parameter data set related to the video data, the first parameter data set comprising a key quality indicator (KQI) parameter data set and / or a key performance indicator (KPI) parameter data set; measuring the importance of each type of parameter in the first parameter data set; based on the measurement result, reconstructing the video data to obtain a reconstructed image, comprising: inputting the video data, the first parameter data set and the measurement result into a mask compression reconstruction network to obtain the reconstructed image; wherein the measurement result comprises an importance ranking of each type of parameter in the first parameter data set, and the importance of each type of parameter is positively correlated with the weight of the type of parameter in the image reconstruction process.

2. The method of claim 1, wherein, The importance of each type of parameter in the first parameter data set is measured, comprising: inputting the first parameter data set into a random forest classification module to obtain an importance order of each type of parameter in the first parameter data set.

3. The method of claim 2, wherein, The random forest classification module is used for: in each base learner of the random forest classification module, determining a target node set using a target class parameter, the target class parameter being any type of parameter in the first parameter data set; measuring the importance of the target class parameter to each node in the target node set; determining the importance of the target class parameter to the base learner in which the target node set is located according to the importance of the target class parameter to each node; determining the importance of the target class parameter to the random forest classification module according to the importance of the target class parameter to each base learner; ranking the importance of each type of parameter to the random forest classification module, and outputting the importance order of each type of parameter in the first parameter data set.

4. The method of claim 1, wherein, The mask compression reconstruction network comprises an encoder, a mask compression module and a decoder; wherein: the encoder is used for encoding the video data according to the first parameter data set and the measurement result, and outputting a latent representation of the video data; the mask compression module is used for converting and compressing the latent representation using a mask; wherein the mask is related to channel correlation; the decoder is used for image reconstruction based on the converted and compressed latent representation to obtain the reconstructed image.

5. The method of claim 4, wherein, The mask compression reconstruction network further comprises a classifier module, which is used for outputting a classification label.

6. The method of claim 1, wherein, The method further comprises: when training the mask compression reconstruction network, obtaining a second parameter data set related to video training data, an importance order of each type of parameter in the second parameter data set, a reconstructed training image obtained by reconstructing the video training data, and a third parameter data set related to the reconstructed training image; determining a loss function according to each type of parameter in the second parameter data set, each type of parameter in the third parameter data set, the importance order of each type of parameter in the second parameter data set, and each type of parameter and importance order in the third parameter data set; performing back propagation according to the loss function to adjust the parameters of the mask compression reconstruction network.

7. A video compression reconstruction apparatus, characterized by comprising: The method comprises: The first acquisition module is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset. The metric module is used to measure the importance of various parameters in the first parameter dataset; The reconstruction module is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image; The reconstruction module is used to: input the video data, the first parameter dataset, and the measurement results into a mask compression reconstruction network to obtain the reconstructed image; wherein the measurement results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

8. The apparatus of claim 7, wherein, The metric module is used to: input the first parameter dataset into the random forest classification module to obtain the importance order of various parameters in the first parameter dataset.

9. The apparatus of claim 7, wherein, The device further includes: The second acquisition module is used to acquire, during the training of the mask compression reconstruction network, a second parameter dataset related to the video training data, the importance order of various parameters in the second parameter dataset, the reconstructed training image obtained by reconstructing the video training data, and a third parameter dataset related to the reconstructed training image. The determination module is used to determine the loss function based on the various parameters in the second parameter dataset, the various parameters in the third parameter dataset, the importance order of the various parameters in the second parameter dataset, and the various parameters and importance order in the third parameter dataset. An adjustment module is used to perform backpropagation based on the loss function and adjust the parameters of the mask compression reconstruction network.

10. A video compression reconstruction device, characterized by, Includes transceivers and processors; The transceiver is used to acquire video data and a first parameter dataset of video services related to the video data. The first parameter dataset includes a Key Quality Indicator (KQI) parameter dataset and / or a Key Performance Indicator (KPI) parameter dataset. The processor is used to measure the importance of various parameters in the first parameter dataset; The processor is used to compress and reconstruct the video data based on the measurement results to obtain a reconstructed image; Specifically, when the processor is used to reconstruct the video data based on the measurement results to obtain a reconstructed image, it is used for: The video data, the first parameter dataset, and the metric results are input into a mask compression reconstruction network to obtain the reconstructed image; wherein, the metric results include the importance ranking of various parameters in the first parameter dataset, and the importance of each type of parameter is positively correlated with the weight of that type of parameter in the image reconstruction process.

11. A video compression reconstruction device, comprising: A transceiver, a processor, a memory, and a program or instructions stored in the memory and executable on the processor; characterized in that, when the processor executes the program or instructions, it implements the video compression and reconstruction method as described in any one of claims 1 to 6.

12. A computer readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the video compression and reconstruction method as described in any one of claims 1 to 6.

Images