Multi-code-rate scheduling method and multi-code-rate scheduling device
A scheduling method and scheduling device technology, applied in the field of training devices, can solve problems such as real-time bit rate fluctuations, decision errors, and inability to prefetch live stream information, and achieve the effect of improving robustness
Pending Publication Date: 2022-02-08
BEIJING DAJIA INTERNET INFORMATION TECH CO LTD +1
10 Cites 1 Cited by
AI-Extracted Technical Summary
Problems solved by technology
[0002] With the popularization of variable bit rate video encoding methods, when encoding video, due to different video contents, the real-time bit rate of the encoded video block will fluctuate up and down at the target bit rate
Based on existing data statistics, the average jitter range...
Method used
Adaptive code rate agent is by collecting state factors such as historical network state, historical code rate decision-making and historical prediction network error, utilizes reinforcement learning to determine the optimal code rate of next video block before current video block is transcoded gear, and notify the server of the decision. Once the video block is transcoded, it will be delivered immediately, so as to better me...
Abstract
The invention provides a multi-code-rate scheduling method and a multi-code-rate scheduling device. The multi-code-rate scheduling method comprises the following steps: acquiring coding information of a current video block from a current live broadcast stream; predicting an output code rate of each preset code rate gear for transcoding the current video block based on the coding information of the current video block; selecting a code rate gear from the preset code rate gears according to the predicted output code rate; and transmitting information about the selected bit rate gear. According to the invention, the real-time code rate of the video code stream is predicted based on the granularity of the block, and the coding information of the video is considered during prediction, so that the video block with the corresponding code rate is more accurately scheduled during the adaptive code rate scheduling decision.
Application Domain
Selective content distributionNeural learning methods
Technology Topic
Self adaptiveStream +7
Image
Examples
- Experimental program(1)
Example Embodiment
[0045] In order to better understand the technical solution of the present disclosure, the technical solutions in the present disclosure will be described in contemplation in connection with the accompanying drawings.
[0046] The following description will be provided with reference to the drawings to help understand the embodiments of the present disclosure of the claims and their equivalents. It includes various specific details to help understand, but these details are considered to be exemplary. Accordingly, there will be various changes and modifications to the embodiments described herein without departing from the scope and spirit of the present disclosure. In addition, descriptions of well-known functions and structures are omitted for clarity and conciseness.
[0047] The terms and words used in the following description and claims are not limited to written meaning, but only by the inventors used to achieve clarity and consistent understanding of the present disclosure. Thus, those skilled in the art will clear, the following description of the various embodiments of the present disclosure is only provided to illustrate the disclosure of the present disclosure of claims and their equivalents.
[0048] It should be noted that the specification and claims of the present disclosure and the terms "first", "second", "second", etc. of the drawings are used to distinguish similar objects without having to describe a particular order or ahead order. It is to be understood that the data such as use can be interchangeable in appropriate, so that the embodiments of the present disclosure described herein can be implemented in the order other than those illustrated or described herein. The embodiments described in the exemplary embodiments described below do not represent all embodiments consistent with the present disclosure. Instead, they are only examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0049] Currently, some people have noticed the importance of forecasting in the live adaptive code rate scheduling system. They observe that there is a nonlinear relationship between the video content and the video bit rate, that is, the lower code rate in the simple scene can still achieve a higher video quality, and the higher video quality in complex scenes is only based on the high yard ratio. In order to achieve better performance in the audience, first, they designed a content extracted neural network to predict the video quality of the video block corresponding to each bit rate in the future. After that, they designed a new bit rate control algorithm based on deep learning to obtain high video quality and low delays in the live broadcast system. They also designed new experience QoE functions for the live broadcast system. They achieve a live ABR scheduling framework based on video quality prediction.
[0050] However, the existing framework only takes into account the effects of video content on video quality, but does not take into account the impact of video content on video block size. Different video contents have different differences in video quality of the video blocks of the same yard gear position, which will also cause the real code rate of the video block in the coding process to jitter up and down in the target code rate, resulting in an adaptive code rate decision. The more serious decision-making mistakes. Second, the fluctuation of the yield can be reflected in a certain extent, and these facts should also be considered in the adaptive code rate decision of the live broadcast system. In addition, existing forecasting models are also large, training and operation require higher operational resources and longer time, and there is also a disadvantage of deep neural network, that is, the test data set and verification data set height Relevant, there is a weaker generalization.
[0051] Based on this, the present disclosure proposes an adaptive code rate scheduling framework for the live stream of the unknown model based on the live stream of MAML, which can better predict the code rate jitter of the video block, thereby completing the adaptive code rate scheduling under live stream. At the same time, ensure the real-timeness of the code rate scheduling, avoiding the problem of constantly querying the server on the server.
[0052] Hereinafter, according to various embodiments of the present disclosure, the methods and apparatus of the present disclosure will be described in detail with reference to the accompanying drawings.
[0053] figure 1 It is a diagram of an application environment for multi-rate scheduling according to an embodiment of the present disclosure.
[0054] Refer figure 1 The application environment 100 includes terminal 110, terminal 120, and server 130.
[0055] Terminal 110 and terminal 120 can be the terminal where different users are located. For example, an anchor uses terminal 110 to perform live recording and upload live video, the viewer uses terminal 130 to view the live video of the anchor. Terminal 110 and terminal 120 can be at least one of a smartphone, a tablet, a laptop, a desktop computer, and the like. Terminal 110 and terminal 120 can be installed with a target application, such as application software for live, video conferencing, and the like.
[0056] Although this embodiment will only show two terminals, those skilled in the art can be known that the number of terminals may be two or more. The embodiment of the present disclosure does not limit the number and device type of the terminal.
[0057] The terminal 110 and terminal 120 can be connected to the server 130 through a wireless network such that the terminal 110 and the terminal 120 can perform real-time communication between the terminal 130. For example, the network can include Bluetooth, local area network (LAN), WAN (WAN), wireless link, intranet, internet, or combination thereof. Here, the server 130 can include a streaming server for transcoding and a content distribution network for distributing video.
[0058] Live with the anchor using terminal 110 and the viewer uses terminal 120 to view live as an example. The terminal 110 transmits the encoded information of the live video to the code rate prediction network while transmitting live video to the server 130. Here, the code rate prediction network can be disposed in the server 130 or in the terminal 120. Before the server completes the transcoding of the received video block, the code rate prediction network predicts the output code rate of each predetermined rate gear of the transcodation, and then transmits the predicted output rate to adaptive. Code Rate Scheduling Agent (also known as an adaptive code rate model). Here, the adaptive code rate scheduling agent can be disposed in the terminal 120 or in the server 130. The adaptive code rate scheduling agent can use the enhanced learning to be transcoded by the video block by means of a hunting network error and historical network status, historical code rate decisions, and historical prediction network errors. The code rate gear position and the determination results will be notified of the server 130. Thus, the server 130 can be issued to the terminal 120 after the optimal rate gear position is transferred to the terminal 120.
[0059] figure 2 It is a flow chart of a multi-rate scheduling method according to an embodiment of the present disclosure. figure 2 The method shown is based on the granularity of the block to predict the real-time code rate of the video code stream, and one video block is described as an example to describe how many magnification scheduling of the present disclosure.
[0060] Refer figure 2 In step S201, encoding information of the current video block is acquired from the current live stream. For example, when an anodol is broadcast using an electronic device, encoding information encoding each video block in the live stream can be acquired from the electronic device in units of video blocks.
[0061] A video block can include a preset number of video frames, for example, each video block can be set in advance including 48 video frames. The encoding parameters may include the residual frame information and quantization parameters of each video frame during encoding. However, the above example is merely exemplary, and the present disclosure is not limited thereto.
[0062] In step S202, the output rate of each predetermined rate gear position for transcoding the current video block is predicted based on the encoded information of the current video block. Each preset rate gear position corresponds to a predicted output rate.
[0063] In order to improve the real time of the rate scheduling, the prediction can be performed for a specific number of video frames in a video block. At least one video frame can be selected from the current video block, and then the output code rate of each predetermined code rate gear position is predicted based on the encoding information of the selected at least one video frame. For example, six video frames can be uniformly selected from the current video block, and then the encoding parameters of the six video frames can be used to predict the respective output rate of each predetermined code rate gear position of the current video block.
[0064] According to an embodiment of the present disclosure, the parameters of the neural network for predicting the code rate can be obtained using a meta-learning method of unknown models. Hereinafter Figure 4 and Figure 5 To detail how to train the neural network of this disclosure.
[0065] In step S203, the code rate gear position is selected from the respective predetermined code rate gear based on the predicted output code rate.
[0066] As an example, the code rate gear can be selected from the respective predetermined code rate gear positions from each predetermined code rate gear based on the predicted output code rate and each state factor. Here, each state factor can include at least one of history network status information, historical code rate selection information, and historical prediction error, and historical network status information includes a variety of network states, and historical code rate selection information includes a variety of code rate decision information. Historical predictive error information includes a plurality of prediction errors for a variety of predictive networks. For example, a historical network state can be obtained from a user terminal (such as the terminal to watch live broadcast) for predicting the current state of the network. History code rate decision information and code rate prediction error information can be obtained from the results of the previously performing code rate schedule to determine the optimal code rate gear position as reference information.
[0067] Enhanced learning can be made through constant attempts, in different states, to obtain instantaneous rewards of different operations in the same state, and state rewards of the next state jump. After a constant attempt. Agent Agent has gradually convergeful for future expectation incentives for different actions in each state, and ultimately selects the optimal decision path based on future reward expectations of different states and future rewards of different motions.
[0068] In the embodiment of the present disclosure, a proxy agent can be first trained, and the agent constantly collects environmental information to decision the code rate gear selection of the next video block, depending on the code rate gear selection, will get the next video block. A instantaneous reward evaluation of the code rate decision and the system state of the next decision-making time, different code rate decisions brings different instantaneous rewards and different future, and strengthen learning Agent through the continuous interaction and exploration of the environment, ultimately get Each state's expectation reward and each state can obtain the highest expected award corresponding action action.
[0069] For example, various state factors (such as historical network status information, historical code rate selection information, and various states included in history predictive error) are used as a "state" that enhances the learning. "Action", Agent, can determine the predicted output rate rate in each state factor in each preset code rate gear position, and then determine the preset code rate gear position corresponding to the maximum incentive value as The optimum code rate gear.
[0070]For example, the proximal strategy can be utilized to enhance the algorithm PPO2 to select the optimum rate gear position suitable for the current environment based on the predicted output code rate and each state factor. However, the above example is merely exemplary, and the present disclosure is not limited thereto.
[0071] By strengthening the learning model, only status factors such as the predicted value of the network state and video block code rate are required, the code rate gear decision of the video block can be automatically generated.
[0072] In step S204, information about the selected code rate gear is transmitted. For example, the determined optimal code rate gear position can be notified to the server where the transferred video block will be issued. Once the current video block completes the transcod, immediately after the optimum rate gear is transferred. The video block is sent to the corresponding user terminal.
[0073] The present disclosure can predict the optimal rate gear of the current viewing environment for different user terminals in different network environments / network conditions, so that users can better view live video.
[0074] image 3 It is a flow diagram of a multi-rate scheduling method according to an embodiment of the present disclosure. image 3 The method shown in the introduction of H.264 encoded information guidance H.265 encoded video blocks of video blocks, if subsequent H.266 encoding is developed, can also use H.265 encoded information guidance H.266 video The adaptive code rate decision of the block. In fact, the higher the advanced coding mode, the higher the requirements of the device, so most of the code stream uploaded in the average, most of the code streams encoded in accordance with the latest encoding. image 3 Taking H.264 video transcoding as a H.265 video as an example.
[0075] Refer image 3 It is assumed that the anchor uses the device 1 to broadcast, and the device 1 can upload the video encoded with the H.264 to 1.2Mbps to the streaming server 2 arranged at the server, and upload the encoded information of the video to the code rate prediction network. . Here, the code rate prediction network can be arranged in the server, such as flow media servers 2, or in the client, such as image 3 Various terminals shown in.
[0076] The present disclosure is predicted based on the output code rate of the video stream based on the granularity of the block, and below is described as an example of a video block. The code rate prediction network can predict the output code rate for each preset encoded gear of the current video block to the H.265 format, and the prediction of the output code rate will complete the current video blocks in the streaming media server. The coding operation of the coding rate is completed before completion, and the code rate prediction network can give the prediction result of the H.265 video stream to the adaptive code rate scheduling agent. Although image 3 The adaptive code rate scheduling agent is shown in the client, but the present disclosure is not limited thereto, and the adaptive code rate scheduling agent can also be arranged in the client.
[0077] Adaptive Code Rate Agents By collecting historical network status, historical code rate decisions, and historical prediction network errors, using strengthening learning to determine the optimal rate gear of the next video block before the current video block is transferred. And will notify the server. Once the video block completes the transfer code, it is immediately issued to better meet the real-time requirements under the live framework.
[0078] In order to better enhance the real-time, the present disclosure uses a lightweight code rate prediction network. In order to better deal with the complexity of live content, when new video occurs, such as when a live broadcast begins, the MAML method can be used to quickly train the code rate prediction network. For example, 5 new video samples and 5 round training can be quickly trained to quickly train new code rate prediction networks. In addition, the new code rate prediction network can also be further fine-tuning according to any additional data of the new video.
[0079] Figure 4 It is a flow chart of a training method of a code rate prediction network according to an embodiment of the present disclosure.
[0080] When the initial training rate prediction network, small sample live data in the historical live broadcast can quickly train the code rate prediction network. For example, a plurality of video blocks and video blocks of each live stream can be obtained from the previous multi-field live broadcast, and an initial code rate prediction is trained for each live stream of video data for each live stream. The internet. For example, the top five video block data of multiple video streams can be imported into the initial code rate prediction network, so that the parameters of the initial code rate prediction model are fine-tuned after fine-tuning at different video streams. MES. Minimum.
[0081] Thereafter, when a new live broadcast occurs, the meta-learning method can be used in real time to further update the parameters of the code rate predictive network in real time to apply to the code rate prediction under the current live scene.
[0082] When using the new video data update code rate prediction network, the training can be used in steps S401 to S404 below.
[0083] Refer Figure 4 In step S401, encoding information of the current video block in the live stream is obtained. The video block can include a preset number of video frames, and the encoding parameters can include residual frame information and quantization parameters of each video frame.
[0084] In step S402, the output code rate of each predetermined code rate gear of the current video block is predicted by the code rate prediction network based on the encoded information of the current video block.
[0085] As an example, at least one video frame can be selected from the current video block, the feature parameters of the at least one video frame are obtained by the code rate prediction network based on the residual frame information of the selected at least one video frame, according to the feature parameters of the at least one video frame. And the corresponding quantization parameters utilize the code rate prediction network to predict the output rate of each predetermined rate gear position of the current video block.
[0086] In step S403, the loss function is constructed based on the predicted output code rate and the actual output rate of the current video block. For example, the mean square error MES loss function can be constructed based on the predicted code rate with the actual code rate.
[0087] In step S404, the parameters of the code predict the network are trained according to the loss calculated by the loss function. For example, the parameters of the code rate prediction network are updated by minimizing the mean square error between the predicted code rate and the actual code rate.
[0088] When the code rate prediction network is well trained, when a new video appears, the video block of the new video in step S401 can be used as a first video block in the video, and then an unknown model can be used. Metaching methods use the predetermined number of video blocks from the current video block to update the parameters of the code rate prediction network for new video from the current video block.
[0089] According to an embodiment of the present disclosure, the training of the code rate prediction network can be mainly divided into two stages. The first phase uses a multi-live stream of history (e.g., the top five video blocks) to quickly train an initial code rate prediction network (ie the initial model of the code rate prediction network), for example, using a number of videos using history. The data set of the stream is minimized after the code rate prediction network is fine-tuning, thereby obtaining the total MES minimum, resulting in the initial model of the code rate prediction network; the second phase uses the data set of the new live stream to update the parameters of this initial model. For example, the parameters of this initial model implement MES minimal after fine-tuning under the data set of new video streams.
[0090] Figure 5 It is a schematic diagram of the structure of the code rate prediction network according to the embodiment of the present disclosure.
[0091] Refer Figure 5 The residual frame of the video block and the corresponding quantization parameter are input to the code rate prediction network to obtain the predicted code rate value of each preset rate gear position of the video block.
[0092] As an example, a residual frame of a video frame in the video block, the individual coding blocks in the residual frame are characterized, and the extracted features are fused to obtain the fusion characteristics of the video frame. For example, Figure 5 Box 4 × 4Transform Block and 8 × 8transform Block represent the process of performing feature extraction for different sizes of encoding blocks, respectively. Figure 5 Box MacRoblock and box frame can represent a feature fusion process of a video frame.
[0093] exist Figure 5 Among the six video frames in a video block can be used to perform a prediction of the code rate. In block chun, the predicted value of each predetermined code rate of the video block is obtained from the respective quantization parameter (QP) of the fusion feature of the six video frames.
[0094] exist Figure 5 In the middle, FC represents the full connectivity layer in the neural network, and AvgPool represents the average cellular layer in the neural network.
[0095] exist Figure 5 In the middle, the prediction model is a lightweight model to better meet the requirements of the project real-time.
[0096] Image 6 It is a schematic structural diagram of a multi-rate scheduling device of the hardware operating environment of the present disclosure.
[0097] like Image 6 As shown, the multi-rate scheduling device 600 can include a processing component 601, a communication bus 602, a network interface 603, an input / output interface 604, a memory 605, and a power supply component 606. The communication bus 602 is used to implement connection communication between these components. The input / output interface 604 can include a video display (such as a liquid crystal display), a microphone, and a speaker, and a user interaction interface (such as a keyboard, mouse, touch input device, etc.), optionally, input / output interface 604 can also include a standard wired interface. Wireless interface. The network interface 603 is optional, including a standard wired interface, a wireless interface, such as a wireless fidel interface. Memory 605 can be a high-speed random access memory, or a stable non-volatile memory. Memory 605 can also be a storage device independently of the foregoing processing assembly 601.
[0098] Those skilled in the art will appreciate that Image 6 The structure shown is not constituied to the multi-rate scheduling device 600, which may include more or fewer components illustrated, or in combination of certain components, or different components.
[0099] like Image 6 As shown, memory 605 as a storage medium may include an operating system (such as a MAC operating system), a data storage module, a network communication module, a user interface module, a real-time communication program, and a database.
[0100] exist Image 6 In the multi-yard scheduling device 600, the network interface 603 is mainly used to communicate with the external electronic device / terminal; the input / output interface 604 is mainly used to interact with the user; the processing component in the multi-rate scheduling device 600 601. Memory 605 can be disposed in the multi-yard scheduling device 600, the multi-rate scheduling device 600 is modified by the processing component 601 calling the multi-yard scheduling method program stored in the memory 605 and various APIs provided by the operating system. Multi-code rate scheduling method or training method provided by the present embodiment.
[0101] Processing component 601 can include at least one processor, stored in memory 605 to perform instruction sets, and perform a multi-rate scheduling method or training method according to the present disclosure, when the computer can perform an instruction set . Further, the processing assembly 601 can perform encoding operations and decoding operations, and the like. However, the above example is merely exemplary, and the present disclosure is not limited thereto.
[0102]As an example, the processing assembly 601 can acquire the encoding information of the current video block from the current live stream, predict the output rate of each predetermined code rate gear position for transcoding the current video block based on the encoded information of the current video block. The optimum code rate gear position is selected from the respective predetermined code rate gear positions from each predetermined code rate gear position according to the predicted output code rate and each state factor, and information about the optimal code rate gear is transmitted.
[0103] As another example, the processing assembly 601 acquires the encoding information of the current video block in the live stream, using the code rate prediction network based on the encoded information of the current video block, to predict the respective predetermined code rate gear of the current video block. The output code rate is constructed based on the predicted output code rate and the actual output code rate of the current video block, and the parameters of the code rate prediction network are trained according to the loss of the loss function. Further, when a new video occurs, the processing component 601 can update the parameters of the training-welld code rate prediction network using the MAML method.
[0104] The multi-rate scheduling device 600 can perform a corresponding multi-rate scheduling method or training method as an electronic device (such as a first electronic device) or server. The multi-rate scheduling device 600 can receive data from the electronic device by inputting output interface 604 and can send / forward data.
[0105] As an example, the multi-rate scheduling device 600 can be a PC computer, a flat panel device, a personal digital assistant, a smartphone, or a device that can perform the above instruction set. Here, the multi-rate scheduling device 600 does not have to be a single electronic device, and may be any assembly of a device or circuit capable of performing the above instructions (or instruction sets) alone or in combination. The multi-rate scheduling device 600 can also be part of an integrated control system or system manager, or can be configured to interface with local or remote (e.g., via wireless transmission) to interface interconnection portable electronic devices.
[0106] In the multi-rate scheduling device 600, the processing component 601 can include a central processor (CPU), a graphics processor (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example, the processing assembly 601 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, and the like.
[0107] Processing component 601 can run instructions or code stored in memory, where memory 605 can also store data. The instructions and data can also be transmitted and received via the network interface 603, wherein the network interface 603 can employ any known transport protocol.
[0108] Memory 605 can integrate with the processing component 601, for example, arranged RAM or flash memory within an integrated circuit microprocessor or the like. Additionally, memory 605 can include separate devices, such as other storage devices that can be used by an external disk drive, a storage array, or any database system. Memory and processing component 601 can be coupled, or may communicate with each other, for example, by an I / O port, a network connection, such that processing assembly 601 can read data stored in memory 605.
[0109] Figure 7 A block diagram of a multi-rate scheduling device according to an embodiment of the present disclosure. Figure 7 The desired multi-rate scheduling device can be used as part of the streaming server or a separate device. The multi-rate scheduling device can be connected to the streaming server for real-time communication.
[0110] Refer Figure 7 The multi-rate scheduling device 700 can include a acquisition module 701, a prediction module 702, a determination module 703, and a transmission module 704. Each module in the multi-rate scheduling device 700 can be implemented by one or more modules, and the name of the corresponding module can vary depending on the type of the module. In various embodiments, some of the modules in the multi-rate scheduling apparatus 700 may be omitted, or additional modules can also be included. Further, the module / element according to various embodiments of the present disclosure can be combined to form a single entity, and thus the functionality of the corresponding module / element can be performed equivalently before combining.
[0111] The acquisition module 701 can acquire encoded information of the current video block from the current live stream. The encoding information may include the residual frame information of the current video block and the quantization parameters.
[0112] The prediction module 702 can predict the output code rate for each predetermined code rate gear of the current video block based on the encoded information of the current video block. The prediction module 702 can select the output code rate of the respective predetermined code rate gear positions using the neural network based on the encoded information of the selected at least one video frame. The parameters of the neural network used to predict the rate of code can be trained using the model-independent element learning method. The specific model training process can be referred to Figure 4 description of.
[0113] The determination module 703 can select the code rate gear position from each predetermined code rate gear according to the predicted output code rate. The determination module 703 can select the optimum code rate gear position from each predetermined code rate gear based on the predicted output code rate and each state factor.
[0114] Status factors may include historical network status information, at least one of the historical code rate selection information, and historical prediction errors. For example, determining module 703 can determine that the predicted output code rate is obtained under various state factors in various state factors, and will determine the preset code rate gear position corresponding to the maximum reward value as the most The code rate gear.
[0115] The transmission module 704 can transmit information about the selected code rate gear. For example, the transmission module 704 can notify the determined optimum code rate to give the streaming media server, so after the current video block is completed, it can directly remove the optimal code rate gear position to the corresponding user terminal. Video block.
[0116] Figure 8 A block diagram of a training apparatus of a code rate prediction network according to an embodiment of the present disclosure. Refer Figure 8 The training device 800 can include acquiring module 801 and training module 802. Each module in the training device 800 can be implemented by one or more modules, and the name of the corresponding module can vary depending on the type of the module. In various embodiments, some modules in the training device 800 may be omitted, or additional modules may also be included. Further, the module / element according to various embodiments of the present disclosure can be combined to form a single entity, and thus the functionality of the corresponding module / element can be performed equivalently before combining.
[0117] The acquisition module 801 acquires the encoding information of the current video block in the live stream. The current video block can include a preset number of video frames, and the encoding parameters can include residual frame information and quantization parameters of each video frame.
[0118] The training module 802 can utilize the output code rate of each predetermined code rate gear of the current video block based on the code rate prediction network based on the encoded information of the current video block, based on the predicted output code rate and the actual video block. The output code rate is used to construct the loss function, and the code rate prediction network parameters are trained according to the loss calculated by the loss function. According to an embodiment of the present disclosure, the current video block may be a first video block in a live stream, and a metamorphosis of a predetermined number of video block codes starting from the current video block can be used using a meta-learning method based on an unknown model. The parameters of the network are trained.
[0119] As an example, the training module 802 can select the characteristic parameters of the at least one video frame based on the residual frame information of the selected at least one video frame based on the selected at least one video frame information, according to the at least one The feature parameters of the video frame and the corresponding quantization parameters utilize the code rate prediction network to predict the output rate of each predetermined rate gear position of the current video block. The code rate prediction network of new video can be obtained based on five samples and 5 rounds of the MAML method and 5 rounds of rapid training.
[0120] According to an embodiment of the present disclosure, an electronic device can be provided. Figure 9 It is a block diagram of an electronic device according to an embodiment of the present disclosure, which may include at least one memory 902 and at least one processor 901, which stores a computer executable instruction set, when the computer executable instruction set is When at least one processor 901 is executed, a multi-rate scheduling method or a training method according to an embodiment of the present disclosure is performed. For example, electronic device 900 can act as an electronic device or as a server.
[0121] Processor 901 can include a central processor (CPU), a graphics processor (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example, the processor 901 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, and the like.
[0122] Memory 902 as a storage medium can include an operating system (eg, a MAC operating system), a data storage module, a network communication module, a user interface module, a multi-rate scheduling method, a training method, and a database.
[0123] Memory 902 can be integrated with processor 901, for example, RAM or flash memory can be arranged within an integrated circuit microprocessor or the like. Additionally, memory 902 can include separate devices, such as other storage devices that can be used by an external disk drive, a storage array, or any database system. Memory 902 and processor 901 can be coupled, or may communicate with each other, for example, by an I / O port, a network connection, etc., such that processor 901 can read files stored in memory 902.
[0124] Further, the electronic device 900 can also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, mouse, touch input device, etc.). All components of the electronic device 900 can be connected to each other via a bus and / or network.
[0125] Those skilled in the art will appreciate that Figure 9 The structure shown is not limited by the pair, which may include more or less components illustrated, or in combination of certain components, or different components.
[0126] According to an embodiment of the present disclosure, a computer readable storage medium of a storage instruction can also be provided, wherein at least one processor is executed or training in accordance with the multi-code rate scheduling method according to the present disclosure when the instruction is operated by at least one processor. method. Examples of computer readable storage media here include: read-only memory (ROM), random access programmable read only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM) , Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Flash, Nonvolatile Memory, CD-ROM, CD-R, CD + R, CD-RW, CD + RW, DVD-ROM , DVD-R, DVD + R, DVD-RAM, DVD + RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or CD Memory, Hard Drive (HDD), Solid State Hard disk (SSD), card memory (such as multimedia card, secure digital (SD) card or speed number (XD) card), tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, Other devices configured to store computer programs in a non-temporary manner and any associated data, data files, and data structures and provide the computer program and any associated data, data files, and data structures. Give the processor or computer such that the processor or computer can perform the computer program. Computer programs in the above-described computer readable storage medium can be run in an environment deployed in computer devices such as clients, hosts, proxy, servers, in addition, computer programs and any associated data, data files And data structures are distributed on the networked computer system such that computer programs and any associated data, data files, and data structures are stored, accessed, and executed by one or more processors or computers.
[0127]According to an embodiment of the present disclosure, a computer program product can also be provided, and instructions in the computer program product can be performed by the processor of the computer device to complete the above multi-code rate scheduling method or training method.
[0128] Other embodiments of the present disclosure will be readily apparent to those skilled in the art.The present application is intended to cover any variations, uses, or adaptive changes in the present disclosure, these variations, uses, or adaptive changes follow the general principles of the present disclosure and include known common sense or customary techniques in the art from the present disclosure..The specification and examples are only considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.
[0129] It should be understood that the present disclosure is not limited to the exact structure described above and shown in the drawings, and various modifications and changes may be made without departing from their extent.The scope of the present disclosure is limited only by the appended claims.
PUM


Description & Claims & Application Information
We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
Similar technology patents
Audience rating survey system and audience rating information searching method based on audio frequency characteristics
Owner:SHANDONG UNIV
Bird nest type constellation and design method thereof
Owner:SHANGHAI AEROSPACE SYST ENG INST
Variational mode decomposition-based method for locating leakage point of pipeline by acoustic emission
Owner:NANJING COLLEGE OF INFORMATION TECH
Revolution body object pose vision measurement method
Owner:CHINA ACAD OF AEROSPACE AERODYNAMICS
Driver fatigue detection method based on blink rule recognition
Owner:NO 15 INST OF CHINA ELECTRONICS TECH GRP
Classification and recommendation of technical efficacy words
- Improve robustness
System and method for implementing a dynamic cache for a data storage system
Owner:NETWORK APPLIANCE INC
Automatic parking method based on fusion of vision and ultrasonic perception
Owner:ZHEJIANG LEAPMOTOR TECH CO LTD
Finite time convergence time-varying sliding mode attitude control method
Owner:BEIJING INSTITUTE OF TECHNOLOGYGY
Scene-based hybrid invasion detection method and system
Owner:AUTOMATION RES & DESIGN INST OF METALLURGICAL IND