Automatic parameter adjusting method and device
By obtaining the module status information of the current module of the image signal processing system, using the parameter adjustment agent to determine parameters based on the parameter generation strategy and module status information, and adjusting the module parameters, solving the problems of poor flexibility, low parameter adjustment effect and low efficiency of the existing automatic parameter adjustment method, and achieving a more efficient automatic parameter adjustment effect.
Patent Information
- Application Number
- CN202510146916.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-08
- Publication Date
- 2025-06-10
AI Technical Summary
The existing automatic parameter adjustment method has poor flexibility, low parameter adjustment effect and efficiency in image signal processing systems. The training complexity of the method based on the proxy network is high, while the image quality is poor after adjustment based on the reinforcement learning method.
By obtaining the module status information of the current module of the image signal processing system, using the parameter adjustment agent to determine parameters based on the parameter generation strategy and module status information, and adjust the module parameters to improve the flexibility, effect and efficiency of automatic parameter adjustment.
It improves the flexibility, parameter adjustment effect and efficiency of automatic parameter adjustment, making the parameter adjustment of the image signal processing system more accurate and efficient.
Smart Images
Figure CN120128783A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of automatic parameter tuning. More specifically, the present disclosure relates to an automatic parameter tuning method and apparatus. Background Art
[0002] With the development of smart phones and digital cameras, people have higher and higher requirements for imaging quality. Image Signal Processing (ISP) is a very important link in camera imaging and plays a decisive role in imaging quality. Image signal processing usually involves multiple modules and a large number of parameters, such as lens correction, dead pixel correction, automatic exposure, noise reduction, color enhancement, etc.
[0003] Currently, the industry research focuses on general automatic parameter tuning methods. The automatic parameter tuning method based on a surrogate network and the automatic parameter tuning method based on reinforcement learning are two common methods.
[0004] Among them, the automatic parameter tuning method based on a surrogate network is: for all non-differentiable modules in the image signal processing system, train a differentiable surrogate network, and use the surrogate network to tune the parameters in the image signal processing system; when updating the modules in the image signal processing system, the surrogate network needs to be retrained; when there are more parameters, the training complexity increases. This results in poor flexibility in parameter tuning; while the automatic parameter tuning method based on reinforcement learning is: combine different modules in the image signal processing system to adapt to different scenario requirements; after such adjustment, the number of modules in the image signal processing system is small, and the parameters of each module cannot be adjusted, and the image quality is poor. Summary of the Invention
[0005] An exemplary embodiment of the present disclosure is to provide an automatic parameter tuning method and apparatus to improve the flexibility, tuning effect and efficiency of automatic parameter tuning.
[0006] According to an exemplary embodiment of the present disclosure, an automatic parameter tuning method is provided, including: obtaining module state information of a current module to be parameter-tuned in an image signal processing system, where the module state information of the current module at least includes a module one-hot encoding parameter for the current module and an output image of the previous module of the current module, and the module one-hot encoding parameter is used to indicate the module currently being adjusted; determining, by a parameter tuning agent based on a parameter generation policy and the module state information, a parameter for adjusting the current module, where the parameter tuning agent is a network structure for adjusting parameters of modules in the image signal processing system; and adjusting the parameters of the current module based on the parameter, so as to improve the flexibility, tuning effect and efficiency of automatic parameter tuning.
[0007] Optionally, the determining, by the parameter tuning agent, the parameters for adjusting the current module based on the parameter generation policy and the module state information may include: extracting features from the module state information to obtain the proxy state vector of the current module; generating an action vector for the current module based on the proxy state vector and the parameter generation policy; and parsing the action vector to obtain the parameters for adjusting the current module, thereby improving the accuracy of the parameters.
[0008] Optionally, the extracting features from the module state information to obtain the proxy state vector of the current module may include: extracting features from the output image of the previous module to obtain the semantic feature vector of the output image of the previous module; and performing feature embedding on the semantic feature vector of the output image of the previous module and the module hot encoding parameters of the current module to obtain the proxy state vector of the current module, thereby improving the flexibility of the automatic parameter tuning method by designing the proxy state vector without restricting the differentiability of the module.
[0009] Optionally, the generating an action vector for the current module based on the proxy state vector and the parameter generation policy may include: inputting the proxy state vector into a flexible actuation / evaluation model including the parameter generation policy to generate an action vector for the current module, thereby improving the accuracy of the action vector by the flexible actuation / evaluation model including the parameter generation policy.
[0010] Optionally, the parsing the action vector to obtain the parameters for adjusting the current module may include: converting the action vector to the parameter space of the image signal processing system to obtain the parameters for adjusting the current module, thereby enabling the parameters to be applied to the image signal processing system.
[0011] Optionally, the automatic parameter tuning method may further include: generating scores for each dimension in multiple dimensions for the image signal processing system after parameter adjustment by an image quality evaluation network; generating an image quality score based on the scores for each dimension in the multiple dimensions and the weights for each dimension in the multiple dimensions; and updating the parameter generation policy based on the image quality score, thereby improving the accuracy of the parameter generation policy.
[0012] According to an exemplary embodiment of the present disclosure, an automatic parameter adjustment device is provided, including: a status information acquisition unit configured to acquire module status information of a current module to be parameter-adjusted in an image signal processing system, where the module status information of the current module at least includes a module thermal encoding parameter for the current module and an output image of a previous module of the current module, and the module thermal encoding parameter is used to indicate the module currently being adjusted; a parameter determination unit configured to determine parameters for adjusting the current module based on a parameter generation strategy and the module status information through a parameter adjustment agent, where the parameter adjustment agent is a network structure for adjusting parameters of modules in an image signal processing system; and a parameter adjustment unit configured to adjust the parameters of the current module based on the parameters.
[0013] Optionally, the parameter determination unit may be configured to: obtain a proxy status vector of the current module by performing feature extraction on the module status information; generate an action vector for the current module based on the proxy status vector and the parameter generation strategy; and obtain parameters for adjusting the current module by parsing the action vector.
[0014] Optionally, the parameter determination unit may be configured to: perform feature extraction on the output image of the previous module to obtain a semantic feature vector of the output image of the previous module; and perform feature embedding on the semantic feature vector of the output image of the previous module and the module thermal encoding parameter of the current module to obtain a proxy status vector of the current module.
[0015] Optionally, the parameter determination unit may be configured to: generate an action vector for the current module by inputting the proxy status vector into a flexible actuation / evaluation model including the parameter generation strategy.
[0016] Optionally, the parameter determination unit may be configured to: convert the action vector into a parameter space of the image signal processing system to obtain parameters for adjusting the current module.
[0017] Optionally, the automatic parameter adjustment device may further include a strategy update unit configured to: generate scores for each dimension in multiple dimensions for the image signal processing system after parameter adjustment through an image quality evaluation network; generate an image quality score based on the scores for each dimension in the multiple dimensions and weights for each dimension in the multiple dimensions; and update the parameter generation strategy based on the image quality score.
[0018] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, an automatic parameter adjustment method according to an exemplary embodiment of the present disclosure is implemented.
[0019] According to an exemplary embodiment of the present disclosure, there is provided a computing device, including: at least one processor; and at least one memory storing a computer program, which when executed by the at least one processor, implements an automatic parameter adjustment method according to an exemplary embodiment of the present disclosure.
[0020] According to an exemplary embodiment of the present disclosure, there is provided a computer program product, and instructions in the computer program product can be executed by a processor of a computer device to complete an automatic parameter adjustment method according to an exemplary embodiment of the present disclosure.
[0021] For the automatic parameter adjustment method and device according to an exemplary embodiment of the present disclosure, by obtaining module state information of a current module to be parameter-adjusted in an image signal processing system, where the module state information of the current module at least includes a module hot encoding parameter for the current module and an output image of a previous module of the current module, and the module hot encoding parameter is used to indicate the currently adjusted module, a parameter adjustment agent determines parameters for adjusting the current module based on a parameter generation strategy and the module state information, where the parameter adjustment agent is a network structure for adjusting parameters of modules in the image signal processing system, and adjusts the parameters of the current module based on the parameters, thereby improving the flexibility, parameter adjustment effect, and efficiency of automatic parameter adjustment.
[0022] Additional aspects and / or advantages of the general concept of the present disclosure will be partially set forth in the following description, and some will be apparent from the description, or can be learned through the implementation of the general concept of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Through the following description with reference to the drawings of exemplary embodiments shown, the above and other objects and features of the exemplary embodiments of the present disclosure will become clearer, where:
[0024] Figure 1 A flowchart showing an automatic parameter adjustment method according to an exemplary embodiment of the present disclosure;
[0025] Figure 2 A schematic diagram showing a module hot encoding parameter according to an exemplary embodiment of the present disclosure;
[0026] Figure 3 A structure diagram showing a parameter adjustment agent according to an exemplary embodiment of the present disclosure;
[0027] Figure 4 A schematic diagram showing the generation of an agent state vector according to an exemplary embodiment of the present disclosure;
[0028] Figure 5Schematic diagram showing a flexible actuation / evaluation framework according to an exemplary embodiment of the present disclosure;
[0029] Figure 6 Schematic diagram showing auto - parameter tuning of an image signal processing system according to an exemplary embodiment of the present disclosure;
[0030] Figure 7 Block diagram showing an auto - parameter tuning device according to an exemplary embodiment of the present disclosure; and
[0031] Figure 8 Schematic diagram showing a computing device according to an exemplary embodiment of the present disclosure. Detailed Description of the Invention
[0032] Reference will now be made in detail to the exemplary embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings, wherein the same reference numerals always refer to the same components. The embodiments will be described below with reference to the accompanying drawings to explain the present disclosure.
[0033] To solve the difficult problems in parameter tuning of existing image signal processing systems, the present disclosure proposes an auto - parameter tuning method based on reinforcement learning. For an image signal processing system that needs parameter tuning, first, the state information to be tracked in the image signal processing system is designed. Parameter information is generated based on the state information, and a parameter parser generates parameters that can ultimately be applied to the image signal processing system according to the parameter information. The parameter tuning system also scores the effects of the generated parameters, thereby updating the strategy for generating parameters to make the quality of the generated parameters higher. The state information is only related to the input image and output image of the module, and has nothing to do with the differentiability of the module. Secondly, each module in the image signal processing system is adjusted step by step in sequence, so there is no limit on the number of parameters to be adjusted in each module. In summary, the present disclosure provides an auto - parameter tuning method with high flexibility and high parameter tuning efficiency.
[0034] Figure 1 Flowchart showing an auto - parameter tuning method according to an exemplary embodiment of the present disclosure. Figure 2 Schematic diagram showing module one - hot encoding parameters according to an exemplary embodiment of the present disclosure. Figure 3 Structure showing a parameter tuning agent according to an exemplary embodiment of the present disclosure. Figure 4 Schematic diagram showing generating an agent state vector according to an exemplary embodiment of the present disclosure. Figure 5 Schematic diagram showing a flexible actuation / evaluation framework according to an exemplary embodiment of the present disclosure.
[0035] Refer to Figure 1, in step S101, obtain the module status information of the current module in the image signal processing system that needs to have its parameters adjusted. Here, the module status information of the current module at least includes the module hot encoding parameter for the current module and the output image of the previous module of the current module. The module hot encoding parameter is used to indicate the module that is currently being adjusted. The module status information is only related to the module hot encoding parameter of the current module and the output image of the previous module, and has nothing to do with the differentiability of the module. Here, when the current module is the first module, the output image of the previous module can be, for example, but not limited to, the input original image.
[0036] The module hot encoding parameter is used to indicate the module that is currently being adjusted. The module hot encoding parameter of the currently adjusted module is set to 1, and the remaining module parameters are set to 0. Only 1 module out of N modules is currently being adjusted. As Figure 2 shown, when Module 1 is the module that is currently being adjusted, the module hot encoding parameter is: hot_encoding(Module 1) = [1, 1,..., 1, 0, 0,..., 0,..., 0, 0,..., 0]; when Module 2 is the module that is currently being adjusted, the module hot encoding parameter is: hot_encoding(Module 2) = [0, 0,..., 0, 1, 1,..., 1,..., 0, 0,..., 0]; when Module N is the module that is currently being adjusted, the module hot encoding parameter is: hot_encoding(Modeule N) = [0, 0,..., 0, 0, 0,..., 0,..., 1, 1,..., 1]. The module status information of the current module can be expressed as, for example, but not limited to, s module = {I prev_output , hot_encoding(Modeule), I pipeline_output}. Here, s module represents the module status information of the current module, I prev_output represents the output image of the previous module of the current module (here, when the current module is the first module, I prev_output represents the input original image), hot_encoding(Module) represents the module hot encoding parameter of the current module, and I pipeline_output represents the output parameter of the image signal processing system.
[0037] In step S102, determine the parameters for adjusting the current module through the parameter tuning agent based on the parameter generation strategy and the module status information. Here, the parameter tuning agent is a network structure used to adjust the parameters of each module in the image signal processing system. Here, the parameter generation strategy can be, for example, but not limited to, a decision network that generates the action vector of the current module based on the module status vector.
[0038] In an exemplary embodiment of the present disclosure, determining parameters for adjusting the current module through a tuning proxy based on a parameter generation policy (e.g., a predetermined decision network) and the module state information may include: obtaining a proxy state vector of the current module by performing feature extraction on the module state information; generating an action vector for the current module based on the proxy state vector and the parameter generation policy; and obtaining parameters for adjusting the current module by parsing the action vector, thereby improving the accuracy of the parameters.
[0039] In an exemplary embodiment of the present disclosure, obtaining a proxy state vector of the current module by performing feature extraction on the module state information may include: performing feature extraction on the output image of the previous module to obtain a semantic feature vector of the output image of the previous module; and performing feature embedding on the semantic feature vector of the output image of the previous module and the module one-hot encoding parameters of the current module to obtain a proxy state vector of the current module, thereby using the obtained proxy state vector to adjust the parameters of the module and improving the flexibility of the automatic tuning method.
[0040] In an exemplary embodiment of the present disclosure, generating an action vector for the current module based on the proxy state vector and the parameter generation policy may include: generating an action vector for the current module by inputting the proxy state vector into a flexible actuation / evaluation model including the parameter generation policy, thereby improving the accuracy of the action vector by the flexible actuation / evaluation model including the parameter generation policy.
[0041] In an exemplary embodiment of the present disclosure, obtaining parameters for adjusting the current module by parsing the action vector may include: converting the action vector into the parameter space of an image signal processing system to obtain parameters for adjusting the current module, thereby enabling the parameters to be applied to the image signal processing system. For example, the parameters of the current module may be obtained through the formula a t = Norm(tanh(a′ t ))). Here, a t represents the parameters of the current module, a′ t represents the action vector, Norm represents normalization, and tanh represents the hyperbolic tangent function.
[0042] Such as Figure 3As shown, the structure of the parameter tuning agent can include three parts: an agent state vector construction part, a Soft Actor-Critic (SAC) framework, and a parameter parsing part. The Soft Actor-Critic framework is a stochastic policy algorithm trained using the off-policy method, which is based on the maximum entropy framework, that is, the goal of policy learning is to maximize the entropy of the policy at each moment in addition to maximizing the reward. In this application, to solve the parameter tuning task of the image signal processing system, an agent state vector is designed in the structure of the parameter tuning agent.
[0043] As Figure 4 shown, the agent state vector construction part first resizes the output image of the previous module, extracts semantic feature vectors from the resized image, and then performs feature embedding on the semantic feature vectors and the module hot encoding parameters of the current module to obtain the agent state vector of the current module. The agent state vector can be expressed as, for example, but not limited to, s agent = concat(f agent , hot_encoding(Module)). Here, s agent represents the agent state vector, concat represents feature embedding, f agent represents the semantic feature vector, and hot_encoding(Module) represents the module hot encoding parameters of the current module. Figure 4 TinyViT in
[0044] As Figure 5 shown, the Soft Actor-Critic (SAC) framework is mainly used to generate action vectors in the action space, that is, the parameters in the pipeline of the image signal processing system. The Soft Actor-Critic (SAC) framework mainly includes a pair of Q networks (i.e., two Q networks) and a decision network. The decision network is responsible for generating action vectors in the action space, the Q network is responsible for evaluating the actions and used to update the decision network, and the update formula of the decision network is: Here, represents the agent state vector at time t, represents the action vector generated by the decision network according to the agent state vector, Denote the evaluation value of the Q network for the agent state vector at time t, k denote the entropy ratio factor, D denote the experience replay pool, and E denote the expected return. The update formula of the Q network is as follows:
[0045] Here, denote the agent state vector at time t, denote the action vector generated by the decision-making network according to the agent state vector at time t, denote the evaluation value of the Q network for the agent state vector at time t, k denote the entropy ratio factor, D denote the experience replay pool, E denote the expected return, r denote the return value obtained by executing the action under the current agent state vector, γ denote the discount factor, denote the agent state vector at time t + 1, denote the evaluation value of the Q network for the agent state vector at time t + 1, denote the action vector generated by the decision-making network according to the agent state vector at time t + 1.
[0046] In step S103, adjust the parameters of the current module based on the parameters.
[0047] In an exemplary embodiment of the present disclosure, in steps S101, S102, and S103, for each parameter-adjustable module among multiple parameter-adjustable modules in the image signal processing system, a same parameter adjustment agent sequentially adjusts the parameters of each parameter-adjustable module based on a parameter generation strategy and the module state information of each parameter-adjustable module, so that the number of parameters to be adjusted in each module can be unrestricted by gradually adjusting each module in sequence.
[0048] In an exemplary embodiment of the present disclosure, the automatic parameter adjustment method may further include: generating scores for each dimension among multiple dimensions for the image signal processing system after parameter adjustment through an image quality evaluation network; generating an image quality score based on the scores for each dimension among the multiple dimensions and the weights for each dimension among the multiple dimensions; updating the parameter generation strategy based on the image quality score, so as to improve the accuracy of the parameter generation strategy. Here, the multiple dimensions may include, for example, but are not limited to, details, color, contrast, brightness, etc. The human eye is not equally sensitive to information in different dimensions of an image. Compared with the details of the image edge, the human eye is more sensitive to the overall brightness of the image. The evaluation function in the present disclosure fully considers this point. As an example, first generate scores for each dimension through the image quality evaluation network, and the user determines the weights for each dimension through a configuration file to generate the final image quality score.
[0049] The automatic parameter tuning method according to an exemplary embodiment of the present disclosure can enable the tuned image to achieve higher scores or effects in dimensions such as details, color, brightness, and contrast, while improving the efficiency of automatic parameter tuning.
[0050] Figure 6 A schematic diagram showing automatic parameter tuning of an image signal processing system according to an exemplary embodiment of the present disclosure. In Figure 6 it, the image signal processing system includes an image signal processing pipeline, a parameter tuning agent, and an image quality evaluation network. The same parameter tuning agent is used to Figure 6 perform automatic parameter tuning on all modules in the image signal processing pipeline in
[0051] As Figure 6 shown, in step S601, the original image is input into module 1 of the image signal processing pipeline. In step S602, based on the output of module 1 and the module hot-encoding parameters for module 2, new module state information of module 2 is determined, and the new module state information of module 2 is input into the parameter tuning agent. In step S603, the parameter tuning agent determines the parameters of module 2, and updates the parameters of module 2 according to the determined parameters. In step S604, based on the output of module 3 and the module hot-encoding parameters for module 4, new module state information of module 4 is determined, and the new module state information of module 4 is input into the parameter tuning agent. In step S605, the parameter tuning agent determines the parameters of module 4, and updates the parameters of module 4 according to the determined parameters. In step S606, based on the output of module N - 1 and the module hot-encoding parameters for module N, new module state information of module N is determined, and the new module state information of module N is input into the parameter tuning agent. In step S607, the parameter tuning agent determines the parameters of module N, and updates the parameters of module N according to the determined parameters. In step S609, the output image corresponding to the original image (e.g., an RGB image) is obtained. In step S609, the output image is evaluated by the image quality evaluation network to obtain an image quality score, and the parameter generation strategy in the parameter tuning agent is updated according to the image quality score.
[0052] The above has described the automatic parameter tuning method according to an exemplary embodiment of the present disclosure in conjunction with Figures 1 to 6 In the following, the automatic parameter tuning device and its units according to an exemplary embodiment of the present disclosure will be described with reference to Figure 7 A block diagram showing an automatic parameter tuning device according to an exemplary embodiment of the present disclosure.
[0053] Figure 7 Refer to
[0054] Refer to Figure 7, the automatic parameter adjustment device includes a status information acquisition unit 71, a parameter determination unit 72, and a parameter adjustment unit 73.
[0055] The status information acquisition unit 71 is configured to acquire the module status information of the current module that needs to be parameter-adjusted in the image signal processing system. Here, the module status information of the current module at least includes the module thermal encoding parameter for the current module and the output image of the previous module of the current module, and the module thermal encoding parameter is used to indicate the module that is currently being adjusted.
[0056] The parameter determination unit 72 is configured to determine, through a parameter adjustment proxy, the parameters for adjusting the current module based on a parameter generation strategy and the module status information. Here, the parameter adjustment proxy is a network structure for adjusting the parameters of the modules in the image signal processing system.
[0057] In an exemplary embodiment of the present disclosure, the parameter determination unit 72 may be configured to: obtain the proxy state vector of the current module by performing feature extraction on the module status information; generate an action vector for the current module based on the proxy state vector and the parameter generation strategy; and obtain the parameters for adjusting the current module by parsing the action vector.
[0058] In an exemplary embodiment of the present disclosure, the parameter determination unit 72 may be configured to: perform feature extraction on the output image of the previous module to obtain the semantic feature vector of the output image of the previous module; and perform feature embedding on the semantic feature vector of the output image of the previous module and the module thermal encoding parameter of the current module to obtain the proxy state vector of the current module.
[0059] In an exemplary embodiment of the present disclosure, the parameter determination unit 72 may be configured to: generate an action vector for the current module by inputting the proxy state vector into a flexible actuation / evaluation model including the parameter generation strategy.
[0060] In an exemplary embodiment of the present disclosure, the parameter determination unit 72 may be configured to: convert the action vector into the parameter space of the image signal processing system to obtain the parameters for adjusting the current module.
[0061] The parameter adjustment unit 73 is configured to adjust the parameters of the current module based on the parameters.
[0062] In an exemplary embodiment of the present disclosure, for each parameter - adjustable module in a plurality of parameter - adjustable modules in an image signal processing system, by using a status information acquisition unit 71, a parameter determination unit 72, and a parameter adjustment unit 73, parameter adjustment is sequentially performed on each parameter - adjustable module based on a parameter generation strategy and the module status information of each parameter - adjustable module.
[0063] In an exemplary embodiment of the present disclosure, the automatic parameter - tuning device may further include a strategy update unit (not shown), configured to: generate scores for each dimension in a plurality of dimensions for the image signal processing system after parameter adjustment through an image quality evaluation network; generate an image quality score based on the scores for each dimension in the plurality of dimensions and the weights for each dimension in the plurality of dimensions; and update the parameter generation strategy based on the image quality score.
[0064] In addition, according to an exemplary embodiment of the present disclosure, there is also provided a computer - readable storage medium, on which a computer program is stored. When the computer program is executed, an automatic parameter - tuning method according to an exemplary embodiment of the present disclosure is implemented.
[0065] In an exemplary embodiment of the present disclosure, the computer - readable storage medium may carry one or more programs. When the computer program is executed, the following steps may be implemented: acquire the module status information of the current module to be parameter - tuned in the image signal processing system, where the module status information of the current module at least includes the module hot - encoding parameter for the current module and the output image of the previous module of the current module, and the module hot - encoding parameter is used to indicate the module currently being adjusted; determine, through a parameter - tuning agent, the parameters for adjusting the current module based on a parameter generation strategy and the module status information, where the parameter - tuning agent is a network structure for adjusting the parameters of modules in the image signal processing system; and adjust the parameters of the current module based on the parameters, thereby improving the flexibility, parameter - tuning effect, and efficiency of automatic parameter - tuning.
[0066] A computer-readable storage medium may, for example, but is not limited to, be a system, apparatus, or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In an embodiment of the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a computer program, and the computer program may be used by or in combination with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable storage medium may be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above. The computer-readable storage medium may be contained in any device; it may also exist alone without being assembled into the device.
[0067] In addition, according to an exemplary embodiment of the present disclosure, there is also provided a computer program product, and instructions in the computer program product can be executed by a processor of a computer device to complete the method of automatic parameter adjustment according to the exemplary embodiment of the present disclosure.
[0068] The above has been combined with Figure 7 to describe the automatic parameter adjustment device according to the exemplary embodiment of the present disclosure. Next, in combination with Figure 8 to describe the computing device according to the exemplary embodiment of the present disclosure.
[0069] Figure 8 A schematic diagram showing a computing device according to an exemplary embodiment of the present disclosure.
[0070] Referring to Figure 8 , a computing device 8 according to an exemplary embodiment of the present disclosure includes a memory 81 and a processor 82. A computer program is stored on the memory 81, and when the computer program is executed by the processor 82, the automatic parameter adjustment method according to the exemplary embodiment of the present disclosure is implemented.
[0071] In an exemplary embodiment of the present disclosure, when the computer program is executed by the processor 82, the following steps may be implemented: obtaining module status information of a current module to be parameter-tuned in an image signal processing system, where the module status information of the current module at least includes a module hot encoding parameter for the current module and an output image of the previous module of the current module, and the module hot encoding parameter is used to indicate the module currently being adjusted; determining, by a parameter tuning agent, a parameter for adjusting the current module based on a parameter generation strategy and the module status information, where the parameter tuning agent is a network structure for adjusting parameters of modules in the image signal processing system; and adjusting the parameters of the current module based on the parameter, so as to improve the flexibility, tuning effect, and efficiency of automatic parameter tuning.
[0072] The computing device in the embodiments of the present disclosure may include, but is not limited to, devices such as mobile phones, laptop computers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), desktop computers, etc. Figure 8 The illustrated computing device is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.
[0073] As described above with reference to Figures 1 to 8 the automatic parameter tuning method and device according to the exemplary embodiments of the present disclosure. However, it should be understood that: Figure 7 the automatic parameter tuning device and its units shown in Figure 8 may be respectively configured as software, hardware, firmware, or any combination of the above for performing specific functions,
[0074] The computing device shown in
[0075] is not limited to including the components shown above, but may add or delete some components as needed, and the above components may also be combined. Although the present disclosure has been specifically shown and described with reference to its exemplary embodiments, those skilled in the art should understand that various changes in form and details may be made thereto without departing from the spirit and scope of the present disclosure defined by the claims.
Claims
1. An automatic parameter adjustment method, comprising: Acquire module state information of a current module of an image signal processing system that needs to be adjusted, wherein the module state information of the current module at least includes module hot encoding parameters for the current module and an output image of a previous module of the current module, and the module hot encoding parameters are used to indicate the module currently being adjusted; Determining, by a parameter adjustment agent, parameters for adjusting the current module based on a parameter generation strategy and the module state information, wherein the parameter adjustment agent is a network structure for adjusting parameters of a module in an image signal processing system; The parameters of the current module are adjusted based on the parameters.
2. The automatic parameter adjustment method according to claim 1, wherein: The determining, by the parameter adjustment agent based on the parameter generation strategy and the module state information, the parameters for adjusting the current module includes: Obtaining the agent state vector of the current module by performing feature extraction on the module state information; generating an action vector for the current module based on the agent state vector and the parameter generation strategy; The parameters for adjusting the current module are obtained by parsing the action vector.
3. The automatic parameter adjustment method according to claim 2, wherein: The step of obtaining the agent state vector of the current module by extracting features from the module state information includes: Performing feature extraction on the output image of the previous module to obtain a semantic feature vector of the output image of the previous module; Feature embedding is performed on the semantic feature vector of the output image of the previous module and the module hot encoding parameters of the current module to obtain the agent state vector of the current module.
4. The automatic parameter adjustment method according to claim 2, wherein: The generating an action vector for the current module based on the agent state vector and a parameter generation strategy comprises: An action vector for the current module is generated by inputting the agent state vector into a flexible actuation / evaluation model including the parameter generation strategy.
5. The automatic parameter adjustment method according to claim 2, wherein: The step of obtaining a parameter for adjusting the current module by parsing the action vector includes: The motion vector is converted into a parameter space of an image signal processing system to obtain parameters for adjusting the current module.
6. The automatic parameter adjustment method according to claim 1, further comprising: Generate a score for each of the multiple dimensions for the parameter-adjusted image signal processing system using an image quality assessment network; generating an image quality score based on the score for each of the plurality of dimensions and the weight for each of the plurality of dimensions; The parameter generation strategy is updated based on the image quality score.
7. An automatic parameter adjustment device, comprising: a state information acquisition unit, configured to acquire module state information of a current module of the image signal processing system that needs to be adjusted, wherein the module state information of the current module at least includes a module hot encoding parameter for the current module and an output image of a previous module of the current module, and the module hot encoding parameter is used to indicate the module currently being adjusted; a parameter determination unit configured to determine, through a parameter adjustment agent, parameters for adjusting the current module based on a parameter generation strategy and the module state information, wherein the parameter adjustment agent is a network structure for adjusting parameters of modules in an image signal processing system; and A parameter adjustment unit is configured to adjust the parameters of the current module based on the parameters.
8. The automatic parameter adjustment device according to claim 7, wherein: The parameter determination unit is configured to: Obtaining the agent state vector of the current module by performing feature extraction on the module state information; generating an action vector for the current module based on the agent state vector and the parameter generation strategy; The parameters for adjusting the current module are obtained by parsing the action vector.
9. A computer-readable storage medium storing a computer program, wherein: When the computer program is executed by a processor, the automatic parameter adjustment method according to any one of claims 1 to 6 is implemented.
10. A computing device comprising: at least one processor; At least one memory stores a computer program, and when the computer program is executed by the at least one processor, the automatic parameter adjustment method according to any one of claims 1 to 6 is implemented.