Live video parameter adjustment and optimization method and device, equipment and medium

Abstract

The application discloses a live video parameter adjustment and optimization method and device, equipment and a medium. The method comprises the following steps: configuring a first value for a single image optimization parameter of a live program in a terminal device to generate a first video stream for an image source; configuring a second value for the image optimization parameter of the live program in the terminal device to generate a second video stream for the first video stream; constructing the first video stream and the second video stream into a comparison group and pushing the comparison group to a plurality of users to determine a voting rate of the first video stream and the second video stream; and determining an optimal value for the image optimization parameter in the live program according to the voting rate. The application generates video streams of a single variable image optimization parameter at different values as a comparison group based on the same software and hardware environment, and obtains a voting rate corresponding to different values based on the comparison group, so that the voting rate can effectively measure the pros and cons of different values and is not affected by other variables, thereby comprehensively improving the configuration effect of the image optimization parameter.

Description

Technical Field

[0001] This application relates to the field of live streaming technology, and in particular to a live streaming video optimization method, a live streaming video parameter tuning method, and corresponding devices, electronic devices, and computer-readable storage media. Background Technology

[0002] In online live streaming scenarios, streamers push video streams to the live streaming room to achieve application purposes such as talent display, information sharing, and knowledge education. This allows streamers to participate in social labor and earn income through these activities, thus promoting overall social benefits.

[0003] Live streaming programs typically include image optimization modules, such as video capture, image encoding, beautification, and image enhancement modules. By setting one or more image optimization parameters in these modules, users can achieve the effect of providing high-quality video images that adapt to the shooting environment.

[0004] When setting image optimization parameters in a live streaming program, users need to observe the effects before and after adjusting the image optimization parameters in order to obtain comparative results. Ideally, the effect of adjusting a single image optimization parameter on the video should be examined, that is, the impact of a single variable on the video image effect should be examined.

[0005] However, the characteristic of live streaming is the immediacy of its video stream. Constrained by this immediacy, a single frame captured by the camera undergoes a series of processing steps before being encoded, packaged, and pushed to a remote server. This represents a dynamic and irreversible input from the camera. Therefore, during comparisons, there are always interfering factors. For example, even with the same camera model, background, and time of shooting, the influence of equipment differences cannot be eliminated. Similarly, even with the same equipment and background, shooting at different times cannot eliminate the interference from spatial and temporal differences in shooting. These shortcomings make it difficult to examine a single variable, leading to challenges in image quality evaluation and affecting the effectiveness of image optimization parameters. Summary of the Invention

[0006] The primary objective of this application is to address at least one of the aforementioned problems by providing a live video parameter tuning and optimization method, a live video parameter tuning method, and corresponding apparatus, electronic equipment, and computer-readable storage medium.

[0007] To achieve the various objectives of this application, the following technical solution is adopted:

[0008] A method for adjusting parameters in live video streaming, proposed to meet one of the purposes of this application, includes the following steps:

[0009] Configure a first value for a single image optimization parameter of the live streaming program in the terminal device, and use the first value to drive the live streaming program to optimize the image source and generate a first video stream;

[0010] Configure a second value for the image optimization parameters of the live streaming program in the terminal device, and use the second value to drive the live streaming program to optimize the first video stream and generate a second video stream;

[0011] The first video stream and the second video stream are constructed into a comparison group and pushed to multiple users to obtain feedback data generated by the multiple users' votes on the first video stream and the second video stream. The feedback data includes the voting rate of the first video stream and the second video stream.

[0012] Based on the voting rates of the first and second video streams in the feedback data, the optimal value of the voting rate is determined, and the image optimization parameters in the live streaming program are configured with the optimal value.

[0013] In optional embodiments, the live video parameter tuning method of this application further includes:

[0014] Mathematical modeling is performed based on the vote rates in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote rates to obtain a vote rate prediction model that generates vote rates based on the values ​​of the image optimization parameters.

[0015] In an optional embodiment, mathematical modeling is performed based on the vote percentages in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including:

[0016] The vote rate of each video stream in the feedback data of multiple comparison groups corresponding to the same image optimization parameters is normalized to a specific numerical range, which includes multiple segmented intervals, and each segmented interval is set with its corresponding level label.

[0017] The specific values ​​of the image optimization parameters of each video stream in the feedback data are used as training samples, and the level labels corresponding to the normalized vote rate of the video stream are used as supervision labels of the training samples. The training samples and their supervision labels are mapped to construct a training dataset.

[0018] The training samples and their supervision labels in the training dataset are called to iteratively train the preset vote rate prediction model until it converges, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

[0019] In an optional embodiment, mathematical modeling is performed based on the vote percentages in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including:

[0020] The vote rate of each video stream in the feedback data of the multiple comparison groups corresponding to multiple image optimization parameters is normalized to a specific numerical range, which includes multiple segmented intervals, and each segmented interval is set with its corresponding level label.

[0021] Different image optimization parameters are grouped according to the vote rate belonging to the same segment interval. The specific values ​​of different image optimization parameters belonging to the same video stream and with the vote rate belonging to the same segment interval are constructed into a parameter vector.

[0022] The parameter vector is used as a training sample, and the level label corresponding to the parameter vector is used as the supervision label of the training sample. The training sample and its supervision label are mapped to construct a training dataset.

[0023] The training samples and their supervision labels in the training dataset are called to iteratively train the preset vote rate prediction model until it converges, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

[0024] In an optional embodiment, after obtaining the vote rate prediction model that generates the vote rate based on the values ​​of the image optimization parameters, the process includes:

[0025] The target value generated by the user adjusting the image optimization parameters in the live streaming program on their terminal device is obtained, and the target value is input into the vote rate prediction model to predict the inferred vote rate corresponding to the target value.

[0026] The level labels corresponding to the segment intervals to which the reasoning vote rate belongs are displayed on the graphical user interface of the user's terminal device.

[0027] In an optional embodiment, after obtaining the vote rate prediction model that generates the vote rate based on the values ​​of the image optimization parameters, the process includes:

[0028] All selectable values ​​within the range of the image optimization parameters in the live streaming program run by the user are input into the vote rate prediction model to obtain the inferred vote rate corresponding to each selectable value;

[0029] The system filters out several selectable values ​​corresponding to the highest vote rate of the reasoning and displays them to the user's graphical user interface;

[0030] Obtain the optional value selected by the user, and configure the user-selected optional value as the specific value of the image optimization parameter in the live streaming program.

[0031] A live video optimization method provided for one of the purposes of this application includes:

[0032] The live streaming program drives the camera unit to capture the live video stream;

[0033] The image optimization parameters of the live streaming program are configured using the live video parameter tuning method described in this application. The image optimization parameters are used to perform beautification processing on the face images in the live video stream.

[0034] The beautified live video stream is pushed to the online live streaming room for display.

[0035] A live video parameter tuning device provided for one of the purposes of this application includes:

[0036] The first optimization module is configured to configure a first value for a single image optimization parameter of the live streaming program in the terminal device, and use the first value to drive the live streaming program to optimize the image source and generate a first video stream.

[0037] The second optimization module is configured to configure a second value for the image optimization parameters of the live streaming program in the terminal device, and use the second value to drive the live streaming program to optimize the first video stream and generate a second video stream.

[0038] The comparison and evaluation module is configured to construct a comparison group of the first video stream and the second video stream and push it to multiple users to obtain feedback data generated by the multiple users' votes on the first video stream and the second video stream. The feedback data includes the voting rate of the first video stream and the second video stream.

[0039] The parameter application module is configured to determine the optimal value of the voting rate based on the voting rates of the first video stream and the second video stream in the feedback data, and to configure the image optimization parameters in the live streaming program with the optimal value.

[0040] A live video optimization device provided for one of the purposes of this application includes:

[0041] The live streaming startup module is configured to have the camera unit capture the live video stream, driven by the live streaming program.

[0042] An image optimization module is configured to use the live video parameter tuning device described in this application to configure the image optimization parameters of the live program, wherein the image optimization parameters are used to perform beautification processing on the face images in the live video stream.

[0043] The video push module is set to push the beautified live video stream to the online live streaming room for display.

[0044] An electronic device provided for one of the purposes of this application includes a central processing unit and a memory, the central processing unit being configured to invoke and run a computer program stored in the memory to perform the steps of the live video parameter tuning method described in this application.

[0045] A computer-readable storage medium is provided for another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the live video parameter tuning method, which, when invoked by a computer, performs the steps included in the method.

[0046] A computer program product provided for another purpose of this application includes a computer program / instructions that, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0047] Compared to existing technologies, this application utilizes the same terminal device and the same live streaming program to generate a first video stream based on the same image source and applying a first value of image optimization parameters. Then, a second video stream is generated based on the first video stream and applying a second value of image optimization parameters. This ensures that both video streams are generated using the same hardware and software environment, undergoing image optimization processing through the same live streaming program's acquisition, encoding, beautification, and enhancement algorithms. The two video streams are then constructed into a comparison group to obtain the voting rate from user voting feedback data. Based on this voting rate, the optimal image optimization parameters are selected and set as the default image optimization parameters for the live streaming program. This completes the parameter configuration related to image optimization for the live streaming program. Since the two video streams in the comparison group are generated based on the same hardware and software environment and the same algorithm, they are highly comparable, and the obtained feedback data better reflects the quality of the image optimization parameters, thus determining the optimal image optimization parameters. Configuring the live streaming program in this way ensures that it can output high-quality live video streams during online live streaming. Attached Figure Description

[0048] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0049] Figure 1 This is an exemplary network architecture used for the live streaming service in the live streaming scenario of this application;

[0050] Figure 2 This is a flowchart illustrating one embodiment of the live video parameter tuning method of this application;

[0051] Figure 3 A flowchart illustrating a live video production line for an exemplary live streaming program of this application;

[0052] Figure 4 This is a schematic diagram illustrating the process of training a vote prediction model suitable for serving the optimization parameters of a single image in an embodiment of this application.

[0053] Figure 5 This is a schematic diagram of the process for training a vote prediction model suitable for serving multiple image optimization parameters in an embodiment of this application;

[0054] Figure 6 This is a schematic diagram illustrating the process of using a vote-rate prediction model to provide users with level labels for the set image optimization parameters in an embodiment of this application;

[0055] Figure 7 This is a schematic diagram illustrating the process of recommending optional values ​​of image optimization parameters to users using a vote-rate prediction model in an embodiment of this application.

[0056] Figure 8 This is a flowchart illustrating one embodiment of the live video optimization method of this application;

[0057] Figure 9 This is a schematic block diagram of the live video parameter tuning device of this application;

[0058] Figure 10 This is a schematic block diagram of the live video optimization device of this application;

[0059] Figure 11 This is a schematic diagram of the structure of an electronic device used in this application. Detailed Implementation

[0060] Please see Figure 1 This application discloses an exemplary application scenario using a network architecture including a terminal device 80, a media server 81, and an application server 82. The terminal device 80 can run a live streaming program, allowing broadcasters or viewers to use the live streaming function. For example, a broadcaster can upload a live video stream to the media server 81 via their terminal device 80, or the media server 81 can push a target broadcaster's live video stream to a viewer's terminal device 80 for playback. The media server 81 is primarily responsible for pushing the live video streams uploaded by each broadcaster to their respective live streaming rooms. The application server 82 can be used to deploy a network live streaming service to maintain interaction between broadcasters and viewers based on the live streaming room.

[0061] The computer program product implemented according to the live video parameter tuning method of this application can run on the media server 81, the application server 82, the terminal device 80, or any other device. By running the computer program product, the various steps of the method are executed to realize the relevant technical solutions. Thus, the image optimization parameters required by the live program are determined according to the given parameters, so that the video images in the live video stream are optimized during the process of generating the live video stream when the live program is broadcasting on the network.

[0062] The computer program product implemented according to the live video optimization method of this application can run in the terminal device, for example, embedded in the live streaming program. Through the operation of the computer program product, the various steps of the method are executed to realize the relevant technical solutions. Thus, the live streaming program is configured according to the image optimization parameters to ensure that the live streaming program can optimize the video images in the live video stream during the process of generating the live video stream when it is broadcasting live online.

[0063] Based on the above exemplary scenarios and related principle descriptions, please refer to Figure 2 In one embodiment of the live video parameter tuning method of this application, the following steps are included:

[0064] Step S1100: Configure a first value for a single image optimization parameter of the live streaming program in the terminal device, and use the first value to drive the live streaming program to optimize the image source and generate a first video stream;

[0065] A live streaming program is installed in the terminal device. When the live streaming program is invoked and run by the central processing unit of the terminal device, it constructs multiple image optimization function modules, including but not limited to, such as... Figure 3 The live video stream production line shown includes a video acquisition module, a beautification module, an image enhancement module, and an image encoding module. Of course, depending on specific functional requirements, the beautification module and image enhancement module can be selectively used or omitted. The video acquisition module is mainly used to acquire video images from image sources to generate the live video stream. The beautification module is mainly used to optimize the face images detected by the face detection module from the video images. The image enhancement module is mainly used to perform image enhancement processing on the acquired video images. The encoding module is mainly used to encode the video images to adapt to network conditions and generate a live video stream, which is then pushed to a media server. The media server then pushes the stream to the live stream room of the broadcaster user logged into the terminal device, allowing viewers in the live stream room to receive and play the live video stream.

[0066] Each of the above image optimization modules can adjust the effect of its output video image through one or more image optimization parameters. Therefore, the effect of the final generated live video stream can be controlled by controlling any image optimization parameter in any image optimization module. The image source acquired by the video acquisition module can be either an image file or a video file, and the video file can be a video file acquired by the camera unit of the terminal device. In one embodiment, the video image generated by the camera can be uniformly in a specific video image format, such as YUV format. Considering that image files and video files are usually in RGBA four-channel format, and RGBA format to YUV format is usually relatively efficient and easy, converting image files and video files from RGBA to YUV can achieve a more efficient live video stream generation.

[0067] To examine the impact of a single variable on the generation of a live video stream, an image source is first determined. The live streaming program then initiates its live video stream production line, where the video acquisition module captures the image source to generate a corresponding video image output. This video image is then processed by the beautification module and the image enhancement module, and finally encoded in the video encoding module to generate a live video stream, defined here as the first video stream. During the generation of the first video stream, the live streaming program can operate according to the pre-configured default values ​​of the image optimization parameters of its various image optimization function modules. For the single variable to be examined, i.e., a specific image optimization parameter, its default value is used as its first value. Therefore, it can be determined that the first video stream is obtained by setting the image optimization parameter to the first value. Of course, the first value can also be a different, non-default value.

[0068] It is not difficult to understand that by setting the image optimization parameter of a single variable to a first value, the live streaming program can be driven to generate a first video stream based on the image source.

[0069] Step S1200: Configure a second value for the image optimization parameters of the live streaming program in the terminal device, and use the second value to drive the live streaming program to optimize the first video stream and generate a second video stream;

[0070] Furthermore, the values ​​of the image optimization parameters under investigation can be adjusted in the live streaming program, for example, by setting them to a second value. Using the first video stream generated in the previous step as the image source, the live streaming program's live video stream production line performs image optimization processing on the first video stream. After image acquisition, image beautification, image enhancement, and image encoding, a second video stream is generated. It can be seen that the second video stream is the live video stream generated after setting the second value for the image optimization parameters under investigation, using the first video stream as the image source. However, both the first and second video streams contain the same image content, and the live video stream production lines processed by both are identical in all environmental factors except for the specific values ​​of the image optimization parameters under investigation. This ensures that the difference in effect between the first and second video streams depends entirely on the changes in the image optimization parameters under investigation, making the first and second video streams highly comparable.

[0071] The process of a live streaming program generating a live video stream is essentially the program's broadcast startup process. From this perspective, the generation of the first and second video streams not only masks the differences in the acquisition devices, such as the camera units in the terminal devices, but also unifies the broadcast startup process, preventing differences in the startup process from introducing variables that could affect comparability. To obtain the first and second video streams during the process of the live streaming program generating the live video stream according to its broadcast startup process, a video recording module can be implemented in the live streaming program. This module records the entire live video stream production line, i.e., the first and second video streams generated during the broadcast startup process.

[0072] Based on the above process, it can be understood that the first and second video streams can actually be obtained through two live broadcasts in the aforementioned live broadcast program. The first broadcast primarily generates the first video stream as a sample object. After encoding, a copy of the stream is made. Taking the H.264 encoding protocol as an example, the stream needs to start from frames containing SPS / PPS and IDR. The video recording module encapsulates the bitstream to generate the video file required by the image acquisition module. During the second broadcast, the video file recorded and synthesized from the first broadcast is used. After decoding, it replaces the camera's captured frames and is sent to the subsequent broadcast process. At this time, the specific values ​​of the image optimization parameters being examined can be adjusted to generate the second video stream.

[0073] Step S1300: Construct the first video stream and the second video stream into a comparison group and push it to multiple users to obtain feedback data generated by the multiple users' votes on the first video stream and the second video stream. The feedback data includes the voting rate of the first video stream and the second video stream.

[0074] The first and second video streams are generated using the same single variable of image optimization parameters, thus forming a comparison group. In practice, multiple such comparison groups can be generated following the above process. Furthermore, different comparison groups can be generated based on different image sources or different image optimization parameters. In short, the image source and image optimization parameters can be changed as needed to generate a massive number of comparison groups. Of course, the two video streams in the same comparison group are generated by processing the same image source twice based on different values ​​of the same image optimization parameters.

[0075] For any comparison group, subjective evaluation information from users can be obtained through user feedback. Therefore, these comparison groups can be pushed to multiple users for selection, and the voting rate of different video streams in the same comparison group can be determined based on the user selection results.

[0076] Specifically, taking a single comparison group as an example, the first and second video streams are arranged on the same display page, and the display page is pushed to the multiple users. Each user will select one of them, thereby obtaining the vote count for the first and second video streams. Based on the ratio of the vote count to the number of users who participated in the vote, the voting rate of the first and second video streams can be obtained. These two voting rates are encapsulated into feedback data corresponding to the comparison group, and the user's perception of each video stream can be identified through the voting rate of each video stream in the feedback data.

[0077] Step S1400: Determine the optimal value of the voting rate based on the voting rates of the first video stream and the second video stream in the feedback data, and configure the image optimization parameters in the live streaming program with the optimal value.

[0078] For a single comparison group, since the feedback data already provides the voting rates for the first and second video streams, and the first and second video streams correspond to two different values ​​of the single image optimization parameter being examined (i.e., the first value and the second value), by comparing the voting rates, the video stream with the highest voting rate is the one with the best user rating. The specific value of the image optimization parameter corresponding to this video stream is the optimal value that can be used to configure the live streaming program to obtain a high-quality live video stream. After configuring the live streaming program with the optimal value, when the live streaming program runs and starts broadcasting, it can provide broadcasters with a relatively high-quality live video stream. Depending on the image optimization parameter being examined, the effect provided by the live video stream will also be different. For example, when the image optimization parameter being examined is a parameter of the beautification module, what is optimized is the face image in the live video stream, mainly manifested as a good beautification effect on the face image. As another example, when the image optimization parameter being examined is a parameter of the video encoding module, such as any parameter such as resolution, bitrate, or frame rate, it can manifest as a change in the image quality of the live video stream. Such measures ensure that one or more image optimization parameters of the live streaming program are configured well.

[0079] When selecting the best value from feedback data obtained by applying different values ​​to multiple comparison groups for the same image source and the same image optimization parameters, the same principle can be applied to a single comparison, where the optimal value of the image optimization parameter is determined by comparing the vote rates of each video stream.

[0080] When it is necessary to adjust parameters for multiple image optimizations, the above process can be followed to select the best option based on feedback data from multiple comparison groups, which will not be elaborated further.

[0081] As can be seen from the above embodiments, this application utilizes the same terminal device and the same live streaming program to generate a first video stream based on the same image source and applying a first value of image optimization parameters. Then, a second video stream is generated based on the first video stream and applying a second value of image optimization parameters. This ensures that both video streams are generated based on the same hardware and software environment, undergoing image optimization processing through the same live streaming program's acquisition algorithm, encoding algorithm, beautification algorithm, and enhancement algorithm. The two video streams are then constructed into a comparison group to obtain the voting rate from the user voting feedback data. Based on the voting rate, the optimal image optimization parameters are selected and set as the default image optimization parameters for the live streaming program. This completes the parameter configuration related to image optimization for the live streaming program. Since the two video streams in the comparison group are generated based on the same hardware and software environment and the same algorithm, they are highly comparable, and the obtained feedback data better reflects the quality of the image optimization parameters, thus enabling the determination of the optimal image optimization parameters. Configuring the live streaming program in this way ensures that it can output high-quality live video streams during network live streaming.

[0082] Based on any embodiment of this application, the live video parameter tuning method of this application further includes:

[0083] Step S1500: Based on the vote rate in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote rate, perform mathematical modeling to obtain a vote rate prediction model that generates the vote rate based on the values ​​of the image optimization parameters.

[0084] This application addresses multiple comparison groups obtained by changing image sources with different image optimization parameters. Feedback data is obtained after user voting, thus revealing a mapping relationship between the image optimization parameters used in each comparison group, the specific values ​​of those parameters, and the vote share of the video stream based on those specific values. This mapping relationship primarily reflects the vote share obtained under different specific values ​​of the image optimization parameters—that is, the mapping relationship between specific values ​​and vote shares. This mapping relationship data can be used to implement mathematical modeling, obtaining one or more vote share prediction models. These models can then predict the corresponding vote share based on the values ​​of the image optimization parameters.

[0085] In one embodiment, based on the massive mapping relationship data obtained from the comparison groups, a single-parameter vote rate prediction model can be mathematically modeled for specific image optimization parameters. This allows the obtained vote rate prediction model to predict the vote rate for the specific values ​​of the image optimization parameters. As needed, a corresponding vote rate prediction model can be generated for each image optimization parameter.

[0086] In another embodiment, based on the massive amount of mapping relationship data obtained from the comparison group, the specific values ​​of multiple target image optimization parameters belonging to the same segment interval can be found by aggregating the vote rates belonging to the same segment interval. These specific values ​​are used as inputs, and the corresponding vote rate segment intervals are used as outputs. Mathematical modeling is performed using this input-output relationship to obtain a vote rate prediction model, which can predict the corresponding vote rate based on a set of target image optimization parameters.

[0087] The vote-getting prediction model can be either a machine learning model or a deep learning model.

[0088] As can be seen from the above embodiments, since the specific values ​​of each image optimization parameter and the mapping relationship data between their vote rates are all generated based on the same live video stream production line, that is, the same broadcast process, the massive mapping relationship data centrally reflects the changes in subjective evaluation information of video images caused by a single variable under the same video stream hardware and software production environment. Therefore, when using these mapping relationship data to build a mathematical model, the vote rate prediction model can accurately learn the mapping relationship between the changes in the values ​​of image optimization parameters and the high or low vote rates, thereby obtaining more accurate prediction capabilities. This can serve the configuration of image optimization parameters in live broadcast programs, providing recommendation and evaluation information for the configuration process.

[0089] Based on any embodiment of this application, please refer to Figure 4 Based on the vote percentages in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages, mathematical modeling is performed to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including:

[0090] Step S1511: Normalize the vote rate of each video stream in the feedback data of multiple comparison groups corresponding to the same image optimization parameter to a specific numerical range. The specific numerical range includes multiple segmented intervals, and each segmented interval is set with its corresponding level label.

[0091] Based on the same image optimization parameter, the mapping relationship data between the vote rate in the feedback data of the comparison group corresponding to different values ​​and the specific numerical values ​​corresponding to those values ​​can be used to construct a vote rate prediction model serving the image optimization parameter. For this purpose, mapping relationship data belonging to the target image optimization parameter can be first filtered from the mapping relationship data corresponding to the comparison group.

[0092] Since the vote percentage is the ratio of the number of votes for each video stream in the same comparison group to the total number of users who participated in the vote, it is actually a normalized value, for example, belonging to a specific numerical range of [0,1]. To facilitate rating, this specific numerical range can be divided into multiple segments, for example, dividing it into 5 segments with each segment representing a 20% rating. Then, each segment can be assigned a corresponding rating label to indicate the rating corresponding to different vote percentages.

[0093] Step S1512: Take the specific values ​​of the image optimization parameters of each video stream in the feedback data as training samples, take the level label corresponding to the normalized vote rate of the video stream as the supervision label of the training samples, and map the training samples and their supervision labels to construct a training dataset.

[0094] In order to construct the training dataset required for mathematical modeling, it is necessary to first construct training sample and supervision label pairs. Therefore, each specific value in the mapping relationship data selected by the same image optimization parameters can be used as a training sample, and the level label corresponding to the specific value can be used as a supervision label. These can be stored in the training dataset for training the vote rate prediction model during the mathematical modeling process.

[0095] Step S1513: Call the training samples and their supervision labels in the training dataset to iteratively train the preset vote rate prediction model until convergence, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

[0096] When training the vote-getting prediction model, each time a training sample is used as the input of the vote-getting prediction model, the inference component infers and obtains feature information. Then, the multi-classifier at the end maps the feature information to a preset classification space, obtains the classification probability corresponding to each level label in the classification space, and then calculates the classification loss value of the obtained level label using the supervision label corresponding to the training sample. The vote-getting prediction model is then updated with gradients based on the classification loss value. This process is repeated iteratively until the vote-getting prediction model is trained to a convergent state.

[0097] Once the vote-getting prediction model has been trained to convergence, it is used in the inference phase. When a specific value of the image optimization parameter is input into it, it can output the classification probability of each level label. The level label with the highest classification probability is the level label corresponding to the specific value, and the classification probability corresponding to the level label can represent the inferred vote-getting rate corresponding to the specific value.

[0098] As can be easily understood from the above embodiments, when mathematical modeling is performed based on the same image optimization parameters to obtain a vote rate prediction model, the selection of the vote rate prediction model is simpler. It can be implemented using traditional machine learning models, which is efficient, accurate, and easy to implement.

[0099] Based on any embodiment of this application, please refer to Figure 5 Based on the vote percentages in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages, mathematical modeling is performed to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including:

[0100] Step S1521: Normalize the vote rate of each video stream in the feedback data of the multiple comparison groups corresponding to the multiple image optimization parameters to a specific numerical range. The specific numerical range includes multiple segmented intervals, and each segmented interval is set with its corresponding level label.

[0101] Mathematical modeling of vote rate prediction models can also be achieved using multiple image optimization parameters. Similar to the embodiment of mathematical modeling based on a single image optimization parameter, each image optimization parameter, using the mapping relationship data between the vote rate in the feedback data of the comparison group corresponding to its different values ​​and the specific numerical values ​​corresponding to the values, can be used to construct a vote rate prediction model serving the image optimization parameter.

[0102] Similarly, since the vote percentage is the ratio of the number of votes for each video stream in the same comparison group to the total number of users who participated in the vote, it is actually a normalized value, for example, belonging to a specific value range of [0,1]. To facilitate rating, this specific value range can be divided into multiple segments, for example, dividing it into 5 segments with each segment representing a 20% rating. Then, each segment can be assigned a corresponding rating label to indicate the rating corresponding to different vote percentages.

[0103] Step S1522: According to the vote rate belonging to the same segment interval, different image optimization parameters are grouped together. The specific values ​​of different image optimization parameters belonging to the same video stream and whose vote rate belongs to the same segment interval are constructed into a parameter vector.

[0104] Unlike the implementation of single-image optimization parameter modeling, the vote rates of different image optimization parameters are not inherently comparable. However, by setting segmented intervals for the vote rates, even though the mapping relationship data are obtained from different image optimization parameters, as long as their vote rates belong to the same segmented interval, the segmented interval can provide flexible tolerance for different image optimization parameters. In other words, even if the specific values ​​of two different image optimization parameters are different, as long as their vote rates in the mapping relationship data belong to the same segmented interval, they can be considered to have the same level of adjustment effect. Based on this principle, these different image optimization parameters can be aggregated, and mapping relationship data that belong to the same segmented interval based on their vote rates can be merged into a subset of mapping relationship data. Then, for each subset of mapping relationship data, a parameter vector mapping to the level label of its respective segmented interval can be constructed.

[0105] The parameter vector can be constructed by randomly selecting a specific value for each image optimization parameter from the same subset of mapping data, thereby obtaining a specific value for each of the image optimization parameters. These specific values ​​can then be sorted in a certain order to construct the parameter vector.

[0106] Step S1523: Use the parameter vector as a training sample, use the level label corresponding to the parameter vector as the supervision label of the training sample, and map the training sample and its supervision label to construct a training dataset.

[0107] Similarly, in order to construct the training dataset required for mathematical modeling, it is necessary to first construct training sample and supervision label pairs. Therefore, each parameter vector can be used as a training sample, and the level label corresponding to the segment interval to which the parameter vector belongs can be used as a supervision label and stored in the training dataset for training the vote rate prediction model during the mathematical modeling process.

[0108] Step S1524: Call the training samples and their supervision labels in the training dataset to iteratively train the preset vote rate prediction model until convergence, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

[0109] Similarly, when training the vote-getting prediction model, each time a training sample is used as the input to the model, the inference component obtains feature information through inference, and the multi-classifier at its end maps the feature information to a preset classification space. In the classification space, the classification probability corresponding to each level label is obtained. Then, the classification loss value of the obtained level label is calculated using the supervision label corresponding to the training sample. The vote-getting prediction model is then updated with gradients based on the classification loss value. This process is repeated iteratively until the vote-getting prediction model is trained to a convergent state.

[0110] Once the vote-getting prediction model has been trained to convergence, it is put into the inference stage. When a set of specific values ​​corresponding to all the image optimization parameters are input into it, it can output the classification probability of each level label. The level label with the highest classification probability is the level label corresponding to the set of specific values, and the classification probability corresponding to the level label can represent the inferred vote-getting rate corresponding to the set of specific values.

[0111] Based on the above embodiments, it is easy to understand that when mathematical modeling is performed based on multiple image optimization parameters to obtain a vote rate prediction model, the vote rate prediction model can be implemented using a deep learning model based on neural networks. By utilizing the reasoning ability of the deep learning model, the inferred vote rate corresponding to multiple sets of image optimization parameter values ​​can be obtained, thereby obtaining the evaluation results of multiple image optimization parameter values.

[0112] Based on any embodiment of this application, please refer to Figure 6 After obtaining the vote rate prediction model based on the values ​​of image optimization parameters, the process includes:

[0113] Step S2100: Obtain the target value generated by the user adjusting the image optimization parameters in the live streaming program on their terminal device, input the target value into the vote rate prediction model, and predict the inferred vote rate corresponding to the target value.

[0114] After obtaining the vote rate prediction model through modeling, whether it is a vote rate prediction model serving a single image optimization parameter or a vote rate prediction model serving multiple image optimization parameters, it can be used to serve the process of anchor users adjusting the corresponding image optimization parameters in the live broadcast program.

[0115] Specifically, when a user inputs one or more image optimization parameters of an image optimization function model into the live streaming program on their terminal device to determine the corresponding target value, the target value of the one or more image optimization parameters can be input into the corresponding vote rate prediction model. With the help of the inference ability obtained during the training of the vote rate prediction model, the inference vote rate corresponding to the target value can be output. This inference vote rate serves to measure the quality of the live video stream corresponding to the target value set by the user.

[0116] Step S2200: Display the level label corresponding to the segment interval to which the reasoning vote rate belongs to the graphical user interface of the user's terminal device.

[0117] As mentioned earlier, there is a correspondence between the vote rate and the segment intervals, and each segment interval also has a corresponding level label. Therefore, the level label corresponding to the segment interval to which the inferred vote rate belongs can be obtained as a prompt for the user to set the corresponding target value. The level label is displayed in the graphical user interface of the user's terminal device. Based on the level label in these prompts, the user can make a choice about the target value they set and finally decide whether to use the corresponding target value.

[0118] When the vote prediction model serves a single specific image optimization parameter, the user can quickly obtain the corresponding grade label by selecting a single value for that specific image optimization parameter. When the vote prediction model serves multiple specific image optimization parameters, the live streaming program can construct a parameter vector in sequence based on the target values ​​corresponding to all specific image optimization parameters after the user sets one of the image optimization parameters, and submit it to obtain the corresponding grade label.

[0119] It can be seen that by using the vote rate prediction model of this application to serve the user parameter tuning during the live broadcast program, when the user determines the corresponding target value for one or more image optimization parameters, the system can instantly generate the level label corresponding to one or more target values ​​as prompt information, which plays a timely information interaction role, can change the user experience, and ensure that the user maintains a good quality of live video stream during the broadcast.

[0120] Based on any embodiment of this application, please refer to Figure 7 After obtaining the vote rate prediction model based on the values ​​of image optimization parameters, the process includes:

[0121] Step S3100: Input all selectable values ​​within the range of the image optimization parameters in the live streaming program run by the user into the vote rate prediction model to obtain the inferred vote rate corresponding to each selectable value;

[0122] When a live streamer runs the live streaming program on their terminal device, enters the online live streaming room, and starts the live streaming process, the voting rate prediction model can be used to provide specific values ​​for one or more image optimization parameters for the live streaming process, in order to assist in the automatic configuration of the live streaming program.

[0123] Specifically, each image optimization module in the live streaming program may contain one or more configurable image optimization parameters. Each image optimization parameter has a value range, and each value range typically contains multiple natural values. For example, for the image optimization parameter corresponding to the whitening effect, its value range can be any natural number between [0, 100]. To evaluate the display efficiency of the live video stream corresponding to these optional values ​​of the image optimization parameters, the live streaming program process can input each optional value of the image optimization parameter into the vote rate prediction model in the background to predict its corresponding inferred vote rate.

[0124] In one embodiment, if the vote rate prediction model is suitable for serving a single image optimization parameter setting, the specific values ​​of each image optimization parameter that needs to be configured can be input into the vote rate prediction model one by one in the manner described above, and the corresponding inferred vote rate can be predicted for each specific value.

[0125] In another embodiment, for cases where the vote rate prediction model is suitable for serving multiple image optimization parameter settings, each of the multiple image optimization parameters can be randomly selected to form the same set of parameter vectors, and the multiple parameter vectors can be input into the vote rate prediction model to obtain the corresponding multiple inferred vote rates.

[0126] Step S3200: Select and display the several selectable values ​​corresponding to the highest reasoning vote rate to the user's graphical user interface;

[0127] The predicted vote rate model, which predicts the inferred vote rate for one or a set of specific values, can be used to measure the quality of the live video stream generated after configuring corresponding image optimization parameters. However, since the inferred vote rate corresponds to the level labels of the segment intervals, in one embodiment, typically only one or a set of specific values ​​whose inferred vote rate belongs to the segment interval representing the level label corresponding to the highest level can be selected as the optional values ​​for the corresponding image optimization parameters. There may be multiple inferred vote rates falling into the same segment interval. Therefore, a parameter list can be constructed from one or a set of specific values ​​corresponding to multiple inferred vote rates and displayed on the user's graphical user interface for the user to select.

[0128] In another embodiment, the role of the rating labels can be disregarded. Instead, the multiple inference vote rates obtained from the vote prediction model can be sorted in reverse order, and the top few inference vote rates can be selected. The specific values ​​corresponding to these selected inference vote rates, one or a group of specific values, can be displayed as optional values ​​on the user's graphical user interface for selection.

[0129] Step S3300: Obtain the optional value selected by the user, and configure the user-selected optional value as the specific value of the image optimization parameter in the live streaming program.

[0130] After the user selects the corresponding optional values ​​for the image optimization parameters they need to set from the graphical user interface, they can configure one or a group of specific values ​​for the corresponding image optimization parameters. The image optimization function module that applies the specific values ​​will perform image optimization processing on the live video stream during the broadcast process according to the specific values, thereby improving the live broadcast quality of the live video stream.

[0131] As can be seen from the above embodiments, the vote rate prediction model can be used to recommend and configure the image optimization parameters of the live broadcast program, which greatly improves the efficiency of users adjusting the image optimization parameters during the broadcast and significantly enhances the user experience.

[0132] Please see Figure 8 A live video optimization method provided for one of the purposes of this application includes:

[0133] Step S4100: The live streaming program drives the camera unit to acquire the live video stream;

[0134] When a user launches a live streaming application on their device, enters their live stream room, and begins broadcasting, the application's process drives the device's camera unit to capture video images, thus acquiring a live video stream corresponding to the local environment. This camera unit is typically a camera connected to the device; for example, when the device is a smartphone, it could be the smartphone's built-in camera.

[0135] Step S4200: Configure the image optimization parameters of the live streaming program using the live video parameter tuning method described in this application. The image optimization parameters are used to perform beautification processing on the face images in the live video stream.

[0136] To facilitate users in improving the efficiency of beautification processing of their facial images, the functional module implemented by the live video parameter tuning method described in this application can be pre-installed in the live streaming program, and thus start with the live streaming program. In this way, the functional module can configure the specific values ​​of one or more image optimization parameters in the beautification processing module of the live streaming program for the broadcaster user according to the process described above, so that the broadcaster user can avoid the trouble of setting the relevant image optimization parameters by himself.

[0137] Step S4300: Push the beautified live video stream to the online live streaming room for display.

[0138] Once the image optimization parameters are configured according to their specific values, the beautification module can perform beautification on the facial images in the video images captured during the broadcast process, thereby enhancing the aesthetic effect of the facial images in the video images. Finally, the corresponding live video stream is obtained during encoding and pushed to the broadcaster's online live broadcast room for display.

[0139] As can be seen from the above embodiments, the live video optimization method of this application can configure image optimization parameters according to the live video parameter tuning method, which can significantly improve video quality, broadcasting efficiency and user experience.

[0140] Please see Figure 9 A live video parameter tuning device provided to meet one of the purposes of this application includes a first optimization module 1100, a second optimization module 1200, a comparison and evaluation module 1300, and a parameter application module 1400. The first optimization module 1100 is configured to configure a first value for a single image optimization parameter of a live streaming program in a terminal device, using the first value to drive the live streaming program to optimize the image source and generate a first video stream. The second optimization module 1200 is configured to configure a second value for the image optimization parameter of the live streaming program in the terminal device, using the second value to drive the live streaming program to optimize the first video stream and generate a second video stream. The comparison and evaluation module 1300 is configured to construct a comparison group of the first and second video streams and push it to multiple users, obtaining feedback data generated by the multiple users' votes on the first and second video streams, the feedback data including the voting rate of the first and second video streams. The parameter application module 1400 is configured to determine the optimal value of the voting rate based on the voting rate of the first and second video streams in the feedback data, and configure the image optimization parameter in the live streaming program with the optimal value.

[0141] Based on any embodiment of this application, the live video parameter tuning device of this application further includes: a modeling implementation module, configured to perform mathematical modeling based on the vote rate in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote rate, to obtain a vote rate prediction model that generates the vote rate based on the values ​​of the image optimization parameters.

[0142] Based on any embodiment of this application, the modeling implementation module includes: a label sorting unit, configured to normalize the vote rate of each video stream in the feedback data of multiple comparison groups corresponding to the same image optimization parameter to a specific numerical range, wherein the specific numerical range includes multiple segmented intervals, and each segmented interval is provided with its corresponding level label; a sample construction unit, configured to use the specific value of the image optimization parameter of each video stream in the feedback data as a training sample, and use the level label corresponding to the normalized vote rate of the video stream as the supervision label of the training sample, and map the training sample and its supervision label to construct a training dataset; and a training implementation unit, configured to call the training sample and its supervision label in the training dataset to iteratively train a preset vote rate prediction model to a convergent state, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given value of the image optimization parameter.

[0143] Based on any embodiment of this application, mathematical modeling is performed according to the vote rate in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote rate to obtain a vote rate prediction model that generates the vote rate based on the value of the image optimization parameters. This model includes: a label processing unit, configured to normalize the vote rate of each video stream in the feedback data of multiple comparison groups corresponding to multiple image optimization parameters to a specific numerical range, wherein the specific numerical range includes multiple segmented intervals, and each segmented interval is assigned its corresponding level label; and a data collection unit, configured to collect different images according to their vote rates belonging to the same segmented interval. For example, the optimization parameters are segmented into intervals, where the specific values ​​of different image optimization parameters belonging to the same video stream and with the same vote rate segment are constructed into parameter vectors. The sample construction unit is configured to use the parameter vectors as training samples and the corresponding level labels of the parameter vectors as supervision labels for the training samples, and to map the training samples and their supervision labels to construct a training dataset. The training implementation unit is configured to call the training samples and their supervision labels in the training dataset to iteratively train the preset vote rate prediction model until convergence, so that the vote rate prediction model is suitable for predicting the corresponding inferred vote rate based on the given values ​​of the image optimization parameters.

[0144] Based on any embodiment of this application, the live video parameter tuning device of this application further includes: a model inference module, configured to obtain the target value generated by the user adjusting the image optimization parameters in the live program on the user's terminal device, input the target value into the vote rate prediction model, and predict the inferred vote rate corresponding to the target value; and a result display module, configured to display the level label corresponding to the segment interval to which the inferred vote rate belongs to the graphical user interface of the user's terminal device.

[0145] Based on any embodiment of this application, the live video parameter tuning device of this application further includes: an automatic prediction module, configured to input all selectable values ​​within the range of the image optimization parameters in the live program run by the user into the vote rate prediction model to obtain the inferred vote rate corresponding to each selectable value; a recommendation display module, configured to filter out several selectable values ​​corresponding to the highest inferred vote rate and display them to the user's graphical user interface; and an automatic configuration module, configured to obtain the selectable value selected by the user and configure the user-selected selectable value as the specific value of the image optimization parameters in the live program.

[0146] Please see Figure 10 A live video optimization device provided to meet one of the purposes of this application includes a live broadcast startup module 4100, an image optimization module 4200, and a video push module 4300. The live broadcast startup module 4100 is configured to drive a camera unit to acquire a live video stream via a live broadcast program. The image optimization module 4200 is configured to configure image optimization parameters of the live broadcast program using the live video parameter adjustment device described in this application. These image optimization parameters are used to perform beautification processing on facial images in the live video stream. The video push module 4300 is configured to push the beautified live video stream to a network live broadcast room for display.

[0147] To address the aforementioned technical problems, embodiments of this application also provide an electronic device. For example... Figure 11The diagram shows the internal structure of an electronic device. This electronic device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When the computer-readable instructions are executed by the processor, they can cause the processor to implement a live video parameter tuning method or a live video optimization method. The processor of this electronic device provides computing and control capabilities to support the operation of the entire electronic device. The memory of this electronic device may store computer-readable instructions. When these computer-readable instructions are executed by the processor, they can cause the processor to execute the live video parameter tuning method of this application. The network interface of this electronic device is used for communication with a terminal. Those skilled in the art will understand that… Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0148] In this embodiment, the processor is used to execute... Figure 9 and / or Figure 10 The system contains the specific functions of each module and unit, and the memory stores the program code and various data required to execute the aforementioned modules or units. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / units in the live video parameter tuning device and / or live video optimization device of this application. The server can call the server's program code and data to execute the functions of all units.

[0149] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the live video parameter tuning method and / or live video optimization method of any embodiment of this application.

[0150] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.

[0151] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0152] In summary, this application generates video streams of image optimization parameters with different values ​​under the same hardware and software environment as a comparison group. Based on the comparison group, the voting rate corresponding to different values ​​is obtained, so that the voting rate can effectively measure the merits of different values ​​without being affected by other variables, thereby comprehensively improving the configuration effectiveness of image optimization parameters.

Claims

1. A method for adjusting parameters in live video streaming, characterized in that, Includes the following steps: Configure a first value for a single image optimization parameter in the live streaming program in the terminal device, and use the first value to drive the image optimization function module in the live streaming program to optimize the image source and generate a first video stream. The image optimization parameter includes parameters for beautifying the face image in the live video stream. Configure a second value for the image optimization parameters of the live streaming program in the terminal device, and use the second value to drive the live streaming program to optimize the first video stream and generate a second video stream; The first video stream and the second video stream are constructed into a comparison group and pushed to multiple audience users. Feedback data generated by the multiple audience users' votes on the first video stream and the second video stream is obtained. The feedback data includes the vote rate of the first video stream and the second video stream. Based on the voting rates of the first video stream and the second video stream in the feedback data, the optimal value of the voting rate is determined, and the image optimization parameters in the live streaming program are configured with the optimal value. Based on the voting rates in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the voting rates, mathematical modeling is performed to obtain a voting rate prediction model that generates the voting rate based on the values ​​of the image optimization parameters. When the anchor user adjusts the image optimization parameters in the image optimization function module of the live broadcast program on the terminal device to determine the corresponding target values, the voting rate prediction model is used to determine the inferred voting rate for the image optimization parameters. The inferred voting rate is displayed through a graphical user interface for the anchor user to set the image optimization parameters.

2. The live video parameter tuning method according to claim 1, characterized in that, Based on the vote percentages in the feedback data from multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages, mathematical modeling is performed to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including: The vote rate of each video stream in the feedback data of multiple comparison groups corresponding to the same image optimization parameters is normalized to a specific numerical range, which includes multiple segmented intervals, and each segmented interval is set with a corresponding level label. The specific values ​​of the image optimization parameters of each video stream in the feedback data are used as training samples, and the level labels corresponding to the normalized vote rate of the video stream are used as supervision labels of the training samples. The training samples and supervision labels are mapped to construct a training dataset. The training samples and the supervised labels in the training dataset are used to iteratively train the preset vote rate prediction model until it converges, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

3. The live video parameter tuning method according to claim 1, characterized in that, Based on the vote percentages in the feedback data from multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the vote percentages, mathematical modeling is performed to obtain a vote percentage prediction model that generates the vote percentage based on the values ​​of the image optimization parameters, including: The vote rate of each video stream in the feedback data of the multiple comparison groups corresponding to multiple image optimization parameters is normalized to a specific numerical range, which includes multiple segmented intervals, and each segmented interval is set with a corresponding level label. Different image optimization parameters are grouped according to the vote rate belonging to the same segment interval. The specific values ​​of different image optimization parameters belonging to the same video stream and with the vote rate belonging to the same segment interval are constructed into a parameter vector. The parameter vector is used as a training sample, and the level label corresponding to the parameter vector is used as the supervision label of the training sample. The training samples and supervision labels are mapped to construct a training dataset. The training samples and the supervised labels in the training dataset are used to iteratively train the preset vote rate prediction model until it converges, so that the vote rate prediction model is suitable for predicting the corresponding inference vote rate based on the given values ​​of image optimization parameters.

4. The live video parameter tuning method according to any one of claims 1 to 3, characterized in that, After obtaining the vote rate prediction model that generates the vote rate based on the values ​​of image optimization parameters, the following steps are included: The target value generated by the anchor user adjusting the image optimization parameters in the live streaming program on the terminal device is obtained, and the target value is input into the vote rate prediction model to predict the inferred vote rate corresponding to the target value. The rating labels corresponding to the segment intervals to which the reasoning vote rate belongs are displayed on the graphical user interface of the anchor user's terminal device.

5. The live video parameter tuning method according to any one of claims 1 to 3, characterized in that, After obtaining the vote rate prediction model that generates the vote rate based on the values ​​of image optimization parameters, the following steps are included: All selectable values ​​within the range of the image optimization parameters in the live streaming program run by the anchor user are input into the vote rate prediction model to obtain the inferred vote rate corresponding to each selectable value; The system filters out several selectable values ​​corresponding to the highest vote rate of the reasoning and displays them to the broadcaster's graphical user interface. Obtain the optional value selected by the broadcaster user, and configure the optional value determined by the broadcaster user as the specific value of the image optimization parameter in the live streaming program.

6. A method for optimizing live video, characterized in that, include: The live streaming program drives the camera unit to capture the live video stream; The image optimization parameters of the live streaming program are configured using the live video parameter tuning method as described in any one of claims 1 to 5, and the image optimization parameters are used to perform beautification processing on the face images in the live video stream. The beautified live video stream is pushed to the online live streaming room for display.

7. A live video parameter adjustment device, characterized in that, include: The first optimization module is configured to configure a first value for a single image optimization parameter of the live streaming program in the terminal device, and use the first value to drive the image optimization function module in the live streaming program to optimize the image source and generate a first video stream. The image optimization parameter is used to perform beautification processing on the face image in the live video stream. The second optimization module is configured to configure a second value for the image optimization parameters of the live streaming program in the terminal device, and use the second value to drive the live streaming program to optimize the first video stream and generate a second video stream. The comparison and evaluation module is configured to construct a comparison group of the first video stream and the second video stream and push it to multiple audience users to obtain feedback data generated by the multiple audience users' votes on the first video stream and the second video stream. The feedback data includes the voting rate of the first video stream and the second video stream. The parameter application module is configured to determine the optimal value of the voting rate based on the voting rates of the first video stream and the second video stream in the feedback data, and to configure the image optimization parameters in the live streaming program with the optimal value. The modeling implementation module is configured to perform mathematical modeling based on the voting rates in the feedback data of multiple comparison groups and the specific values ​​of the image optimization parameters corresponding to the voting rates, to obtain a voting rate prediction model that generates the voting rate based on the values ​​of the image optimization parameters. When the anchor user adjusts the image optimization parameters in the image optimization function module to determine the corresponding target values ​​for the live video stream in the live streaming program on the terminal device, the voting rate prediction model is used to determine the corresponding inferred voting rate for the image optimization parameters, and the inferred voting rate is displayed through a graphical user interface for the anchor user to set the image optimization parameters.

8. A live video optimization device, characterized in that, include: The live streaming startup module is configured to have the camera unit capture the live video stream, driven by the live streaming program. An image optimization module is configured to apply the live video parameter tuning device as described in claim 7 to configure the image optimization parameters of the live program, wherein the image optimization parameters are used to perform beautification processing on the face images in the live video stream. The video push module is set to push the beautified live video stream to the online live streaming room for display.

9. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 6, which, when invoked by a computer, executes the steps included in the corresponding method.

Images