A visual artificial intelligence algorithm arrangement method and device, equipment and medium

Abstract

The present application relates to the field of artificial intelligence, and particularly relates to a visual artificial intelligence algorithm arrangement method and device, equipment and medium. The method comprises the following steps: developing an operator to solidify various inputs, outputs, training, reasoning and evaluation in an artificial intelligence process into an operator component; loading the operator component to a visual arrangement page built based on a react framework, and receiving, by the visual arrangement page, combination and configuration of the operator component by a user to generate a target model; receiving, by the visual arrangement page, training parameter configuration of the target model by the user, and calling a container to train the target model based on the training parameter configuration; and publishing the trained target model into a model application to provide model service externally. The scheme of the present application enables the user to realize visual operation in the arrangement of training, evaluation, reasoning and the like, reduces understanding and application cost, and improves model production efficiency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and in particular to a method, apparatus, device, and medium for visualizing artificial intelligence algorithm programming. Background Technology

[0002] As artificial intelligence continues to develop, various computer vision algorithms have emerged, including those for imaging principles, boundaries and curves, image classification, image segmentation, object detection, shape analysis, texture analysis, image reconstruction, image generation, and face recognition. However, most of these algorithms run on Python interpreters, which presents a learning curve for those with a weak foundation in artificial intelligence. Visualization platforms such as AI Studio and algorithm-visualizer are also more difficult to use, making it hard for beginners to understand the complete process of model training, evaluation, and inference. Furthermore, they rely on the internet and cannot be used offline.

[0003] Currently, with the growing popularity of artificial intelligence and the deepening research on artificial intelligence algorithms, various business algorithms have emerged on the Internet, creating an urgent need for simple and efficient methods for producing artificial intelligence algorithm models. Summary of the Invention

[0004] In view of this, it is necessary to provide a method, apparatus, device and medium for visual artificial intelligence algorithm arrangement to address the above technical problems.

[0005] According to a first aspect of the present invention, a method for orchestrating visual artificial intelligence algorithms is provided, the method comprising:

[0006] Operators are developed to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence process into operator components;

[0007] The operator components are loaded into a visual orchestration page built on the React framework, and the visual orchestration page receives the user's combination and configuration of the operator components to generate the target model;

[0008] The visualization orchestration page receives the user's configuration of training parameters for the target model, and calls the container to train the target model based on the configuration of training parameters;

[0009] The trained target model is published as a model application to provide model services to external users.

[0010] In some embodiments, the method further includes;

[0011] During training, the backend interface service is called to store the model file, file address, model description, and model evaluation mAP into the model library.

[0012] In some embodiments, the method further includes:

[0013] Provides a process insight interface to show users the training data, evaluation data, inference process data, and result data generated during model training.

[0014] In some embodiments, operators are developed to solidify various inputs, outputs, training, inference, and evaluation processes in an artificial intelligence workflow into operator components, including:

[0015] The operator configuration is input into the operator development system and stored in the database;

[0016] The operator classification, anchor point type, algorithm callback function, algorithm parameter type, and algorithm script function are stored in the database as the main algorithm parameter information;

[0017] A list of operators is generated based on data from the database, which users can drag and drop to display.

[0018] In some embodiments, the operator component is loaded into a visual orchestration page built on the React framework, and the visual orchestration page receives user combinations and configurations of the operator component to generate a target model, including:

[0019] Use jsplumb, jquery, echarts, crypto-js, and axios to design page algorithm libraries and process modeling pages;

[0020] Load common operators and algorithms from the page algorithm library, and load the operator list from the database in the left-hand menu.

[0021] During the initialization of the process modeling page, nodes are cached as node information arranged on the page, and connects are cached as node connection information arranged on the page.

[0022] In the React framework's componentDidUpdate, execute the rendering of operator nodes in the page based on nodes, the rendering of operator connection information in the page based on connects, and configure the entire orchestration panel to be draggable.

[0023] It receives operator components that users drag from the left menu to the process modeling page, as well as directed connection segments that receive operator components, and connects to the fixed end node on the far right.

[0024] When an operator is dragged from the left menu to the panel, an object is constructed. The node's coordinates, index key, algorithm information, anchor type, and node name are added as an object to the page's nodes array, which is then used as a node array.

[0025] During node rendering, the rendering is performed according to the algorithm anchor point type on the node. The page connects two anchor points and adds the line segment information at the time of connection to the connects information on the page as a connection array. When the page saves, it saves nodes, connects, process ID, and update time to the database.

[0026] When the page loads, it retrieves the node and connect information based on the process ID and loads it onto the page to enable secondary development after saving.

[0027] In some embodiments, the visual orchestration page receives the user's configuration of training parameters for the target model, and calls a container to train the target model based on the configuration of training parameters, including:

[0028] Use jsplumb, jquery, echarts, crypto-js, and axios to design process control panels and process progress panels;

[0029] The workflow control panel in the visual orchestration page provides four buttons: Save, Refresh, Start, and Stop. You can use the Start button to select the container to start the workflow.

[0030] When starting a process, the process ID from the front end is passed to the back end. The back end retrieves node information and connect information based on the process ID, and then retrieves the input and output endpoints of the connects.

[0031] Sorting is performed using a topology algorithm, and the corresponding sorted nodes are obtained.

[0032] Based on the node sorting and algorithm information, load the page parameter information of each node to form complete process information, construct a context object, and start a thread from the thread pool to call the algorithm execution container.

[0033] The page starts a scheduled task to check whether a thread is running in the thread pool and whether a script is being executed in the container.

[0034] If the process is running, the running log will be fed back to the process progress panel and a progress bar will be provided to display the progress.

[0035] In some embodiments, the target model belongs to at least one of imaging principles, boundaries and curves, image classification, image segmentation, target detection, shape analysis, texture analysis, image reconstruction, image generation, and face recognition.

[0036] According to a second aspect of the present invention, a visual artificial intelligence algorithm orchestration apparatus is provided, the apparatus comprising:

[0037] The operator development module is configured to develop operators to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence workflow into operator components.

[0038] The orchestration module is configured to load the operator components into a visual orchestration page built on the React framework, and the visual orchestration page receives the user's combination and configuration of the operator components to generate the target model.

[0039] The training module is configured to receive the user's training parameter configuration for the target model from the visualization orchestration page, and to call a container to train the target model based on the training parameter configuration;

[0040] The model publishing module is configured to publish the trained target model as a model application to provide model services to external users.

[0041] According to a third aspect of the present invention, a computer device is also provided, the computer device comprising:

[0042] At least one processor; and

[0043] The memory stores computer programs that can run on the processor, which executes the aforementioned visualization artificial intelligence algorithm orchestration method when executing the program.

[0044] According to a fourth aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, performs the aforementioned visualization artificial intelligence algorithm orchestration method.

[0045] The aforementioned visual artificial intelligence algorithm orchestration method enables the orchestration of artificial intelligence algorithm processes using a visual platform under offline conditions. It imports each algorithm into the software system and uses a visual orchestration page built based on the React framework to orchestrate the algorithm production model through a certain connection order. It also enables the model to run and be called in runtime, allowing users to perform visual operations during orchestration, training, evaluation, and inference processes. This reduces the cost of understanding and application and improves the efficiency of model production.

[0046] In addition, the present invention also provides a visual artificial intelligence algorithm orchestration device, a computer device, and a computer-readable storage medium, which can achieve the above-mentioned technical effects, and will not be described in detail here. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart illustrating a visual artificial intelligence algorithm orchestration method provided in one embodiment of the present invention;

[0049] Figure 2A Page intent provided by a page algorithm library in one embodiment of the present invention;

[0050] Figure 2B A schematic diagram of a process modeling page provided in one embodiment of the present invention;

[0051] Figure 2C This is a schematic diagram of an operator parameter configuration interface provided in one embodiment of the present invention;

[0052] Figure 2D A schematic diagram of a process control panel is provided for one embodiment of the present invention;

[0053] Figure 2E A schematic diagram of a process insight page provided for one embodiment of the present invention.

[0054] Figure 2F A schematic diagram of the reasoning process page provided in one embodiment of the present invention;

[0055] Figure 2G This is a schematic diagram illustrating the evaluation information provided in one embodiment of the present invention.

[0056] Figure 2H This is a schematic diagram of a comparison interface between the original image and the inference result image provided in an embodiment of the present invention.

[0057] Figure 3A This is a schematic diagram of another visualization artificial intelligence algorithm orchestration method provided in one embodiment of the present invention;

[0058] Figure 3B This is a schematic diagram of an operator parameter modification interface provided in another embodiment of the present invention;

[0059] Figure 3C A schematic diagram of the new project workflow interface provided in another embodiment of the present invention;

[0060] Figure 3E A schematic diagram of the startup process settings interface provided in another embodiment of the present invention;

[0061] Figure 3F A schematic diagram of an interface for observing the running progress provided in another embodiment of the present invention;

[0062] Figure 4 A schematic diagram of a visual artificial intelligence algorithm orchestration device provided in another embodiment of the present invention;

[0063] Figure 5 This is an internal structural diagram of a computer device according to another embodiment of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0065] It should be noted that all uses of "first" and "second" in the embodiments of the present invention are for the purpose of distinguishing two entities or parameters with the same name but different names. It is clear that "first" and "second" are only for the convenience of expression and should not be construed as limiting the embodiments of the present invention. Subsequent embodiments will not explain this in detail.

[0066] In one embodiment, please refer to Figure 1 As shown, the present invention provides a visual artificial intelligence algorithm orchestration method 100, specifically, the method includes the following steps:

[0067] Step 101: Develop operators to solidify various inputs, outputs, training, inference, and evaluation in the artificial intelligence process into operator components;

[0068] Step 102: Load the operator component into a visual orchestration page built on the React framework, and the visual orchestration page receives the user's combination and configuration of the operator component to generate the target model;

[0069] Step 103: The visualization orchestration page receives the user's training parameter configuration for the target model, and calls the container to train the target model based on the training parameter configuration;

[0070] Step 104: Publish the trained target model as a model application to provide model services to external users.

[0071] The aforementioned visual artificial intelligence algorithm orchestration method enables the orchestration of artificial intelligence algorithm processes using a visual platform under offline conditions. It imports each algorithm into the software system and uses a visual orchestration page built based on the React framework to orchestrate the algorithm production model through a certain connection order. It also enables the model to run and be called in runtime, allowing users to perform visual operations during orchestration, training, evaluation, and inference processes. This reduces the cost of understanding and application and improves the efficiency of model production.

[0072] In some embodiments, the method further includes;

[0073] During training, the backend interface service is called to store the model file, file address, model description, and model evaluation mAP into the model library.

[0074] In some embodiments, the method further includes:

[0075] Provides a process insight interface to show users the training data, evaluation data, inference process data, and result data generated during model training.

[0076] In some embodiments, step 101 above, developing operators to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence workflow into operator components, includes:

[0077] The operator configuration is input into the operator development system and stored in the database;

[0078] The operator classification, anchor point type, algorithm callback function, algorithm parameter type, and algorithm script function are stored in the database as the main algorithm parameter information;

[0079] A list of operators is generated based on data from the database, which users can drag and drop to display.

[0080] In some embodiments, step 102, which involves loading the operator component onto a visual orchestration page built on the React framework, and having the visual orchestration page receive user combinations and configurations of the operator component to generate a target model, includes:

[0081] Use jsplumb, jquery, echarts, crypto-js, and axios to design page algorithm libraries and process modeling pages;

[0082] Load common operators and algorithms from the page algorithm library, and load the operator list from the database in the left-hand menu.

[0083] During the initialization of the process modeling page, nodes are cached as node information arranged on the page, and connects are cached as node connection information arranged on the page.

[0084] In the React framework's componentDidUpdate, execute the rendering of operator nodes in the page based on nodes, the rendering of operator connection information in the page based on connects, and configure the entire orchestration panel to be draggable.

[0085] It receives operator components that users drag from the left menu to the process modeling page, as well as directed connection segments that receive operator components, and connects to the fixed end node on the far right.

[0086] When an operator is dragged from the left menu to the panel, an object is constructed. The node's coordinates, index key, algorithm information, anchor type, and node name are added as an object to the page's nodes array, which is then used as a node array.

[0087] During node rendering, the rendering is performed according to the algorithm anchor point type on the node. The page connects two anchor points and adds the line segment information at the time of connection to the connects information on the page as a connection array. When the page saves, it saves nodes, connects, process ID, and update time to the database.

[0088] When the page loads, it retrieves the node and connect information based on the process ID and loads it onto the page to enable secondary development after saving.

[0089] In some embodiments, step 103, where the visual orchestration page receives the user's configuration of training parameters for the target model and, based on the configuration of training parameters, invokes a container to train the target model, includes:

[0090] Use jsplumb, jquery, echarts, crypto-js, and axios to design process control panels and process progress panels;

[0091] The workflow control panel in the visual orchestration page provides four buttons: Save, Refresh, Start, and Stop. You can use the Start button to select the container to start the workflow.

[0092] When starting a process, the process ID from the front end is passed to the back end. The back end retrieves node information and connect information based on the process ID, and then retrieves the input and output endpoints of the connects.

[0093] Sorting is performed using a topology algorithm, and the corresponding sorted nodes are obtained.

[0094] Based on the node sorting and algorithm information, load the page parameter information of each node to form complete process information, construct a context object, and start a thread from the thread pool to call the algorithm execution container.

[0095] The page starts a scheduled task to check whether a thread is running in the thread pool and whether a script is being executed in the container.

[0096] If the process is running, the running log will be fed back to the process progress panel and a progress bar will be provided to display the progress.

[0097] In some embodiments, the target model belongs to at least one of imaging principles, boundaries and curves, image classification, image segmentation, target detection, shape analysis, texture analysis, image reconstruction, image generation, and face recognition.

[0098] In another embodiment, to facilitate understanding of the present invention, the implementation process of a visual artificial intelligence algorithm orchestration method is described in detail below using a scenario for object detection as an example. The principle is to divide the main contents of the artificial intelligence process into six parts: operator development, visual orchestration, task execution, model output, model deployment, and process insight. The following will describe each of these important parts in detail:

[0099] Part 1: Operator Development

[0100] Please refer to Figure 2A As shown, the main orchestration steps of the AI ​​algorithm visualization orchestration process involve inputting operator configurations into the operator development system, storing them in the operator repository, and classifying operators (e.g., platform classification, application scenario classification), anchor point type (specifying the number of input and output parameters of the operator), algorithm callback function (determining the type of backend service called by the algorithm), algorithm parameter type (determining the type of matching parameters of the algorithm), and algorithm script function (determining the address of the backend script file called by the algorithm) as the main algorithm parameter information, which are stored in the database. In the menu of the algorithm visualization orchestration process, the list of operators stored in the database can be retrieved for users to drag and drop for display. The visualization orchestration page can modify the parameters of each operator according to the above parameter types (e.g., YOLOv3, YOLOv5, mobilenet_v2, etc.).

[0101] The operator management section here provides common operators: "Select Dataset," "Model Evaluation," "Model Input," "Model Output," and "Model Inference." These are compatible operator components usable across multiple AI frameworks. Combining these common components with user-defined operator components forms the entire operator suite for this AI platform. Users can arrange these components sequentially on the visual orchestration page below, creating a runnable workflow to execute and generate tasks.

[0102] Part Two: Visual Layout

[0103] The visual orchestration page is based on the React framework and uses jsplumb, jQuery, echarts, crypto-js, and axios as its main design components. The main design pages include an algorithm library, a process modeling page, a process control panel, and a process progress panel. The page's algorithm library loads the common operators and algorithms configured in the operator development process. Common operators and algorithm modules configured in the operator development menu are loaded from the database and added to the left-hand menu. These modules are then dragged and dropped into the process design panel, connected as directed line segments, and finally connected to a fixed end node on the far right.

[0104] During initialization, the workflow orchestration page caches nodes as the node information for the orchestration on the page and connects as the connection information between the nodes. In the componentDidUpdate method in React, the page renders the operator nodes based on nodes and the operator connection information based on connects, and sets the panel to allow the entire orchestration panel to be dragged and dropped.

[0105] Please refer to Figure 2B As shown, during the arrangement process, an operator menu exists on the left. When the page is initialized, the operator menu loads the common components and algorithm components configured during operator development. Users select common components and algorithm components and drag them to the right, connecting them with line segments.

[0106] In the implementation process, from a code logic perspective: when an operator is dragged from the left menu to the panel, an object needs to be constructed. The node's coordinates, index key, algorithm information, anchor point type, and node name are added as an object to the page's `nodes` array, forming a node array. During node rendering, rendering is performed based on the algorithm anchor point type on the node. The page can connect two anchor points, adding the line segment information to the page's `connects` information, forming a connection array. When the page saves, it saves `nodes`, `connects`, process ID, and update time to the database. When the page loads, it retrieves the associated `nodes` and `connects` information based on the process ID and loads it onto the page, enabling secondary development after saving.

[0107] Please refer to Figure 2C As shown, each node is bound to an ondbclick event. Double-clicking an operator on the page will pop up a parameter configuration page of the specified type based on the algorithm parameter type.

[0108] The "Select Dataset" node provides a list of datasets to choose from, along with the file storage location for the algorithm, the dataset configuration file generation parameters, and the selection of the dataset that meets the data requirements for model training.

[0109] The "Model Input" node provides a list to select from the existing pre-trained model library in the system as the pre-trained model for model training.

[0110] The "Algorithm" type node can provide a multi-tab configuration file text editing page, which allows users to freely edit various configuration attributes of the algorithm. The image below shows the model training configuration file for the Paddle framework, where the training model is a YOLOv3 network structure.

[0111] The "Model Output" node provides a form for users to fill in the model category, name, and description when outputting the model. After successful model training, the model can be stored in the local model repository based on the entered information, for use in model inference and secondary training. Parameter configuration information is saved in the background and loaded during process execution.

[0112] Part Three: Task Execution

[0113] The workflow control panel in the visual orchestration page provides four buttons: Save, Refresh, Start, and Stop. You can use the Start button to select to start the container and start the workflow.

[0114] Please refer to Figure 2D As shown, when starting the process, the process ID from the front end is passed to the back end. The back end retrieves the node and connection information based on the process ID, sorts the connections according to their input and output endpoints using a topology algorithm, obtains the corresponding node sorting, and loads the page parameter information for each node based on the node sorting and algorithm information to form complete process information. A context object is then constructed, and a thread is started from the thread pool to execute the algorithm container. A scheduled task is started on the page, checking whether the thread in the thread pool is running and whether the script in the container is executing. If the process is running, the runtime logs are fed back to the process progress panel, and a progress bar is provided for display.

[0115] The model training, evaluation, and inference processes are all carried out in a single directory generated in the environment. The training, evaluation, and inference scripts will set the files to be generated in the current directory, which ensures that the contents of the generated files are all in the specified directory.

[0116] When the model output module is executed, it calls the backend interface service to store the trained model file, file address, model description, and model evaluation mAP (mean accuracy) into the model library.

[0117] Part Four: Process Insights

[0118] Please refer to Figure 2E As shown, after the process is completed, you can switch to the Insights page via the Insights tab at the top of the Orchestration page.

[0119] Please refer to Figure 2F As shown, the Insights main page displays data and results from the training, evaluation, and inference processes during the workflow. During training, users can view the training dataset, logs, and the generated model. During evaluation, users can view evaluation logs and mAP (mean accuracy). During inference, users can view the original inference image and the resulting inference image.

[0120] The process is as follows: clicking on a node on the insight page retrieves historical training information from the backend service based on the node's ID information, loads the corresponding directory file content from the file server, and returns it to the page for display. Depending on the content clicked on the page, model evaluation will load and display evaluation information; please refer to [link / reference]. Figure 2G As shown. The model inference will display a comparison chart of the model inference results; please refer to it. Figure 2H As shown.

[0121] Please combine Figure 3A As shown below, the process of generating a model using another visualization artificial intelligence algorithm orchestration method 200 will be explained in detail:

[0122] Step 201, maintain the algorithm, load the algorithm information template into the system, please refer to... Figure 3B As shown;

[0123] Step 202: Create a new project flow. Drag operators from the flow menu panel into the right-hand flow panel to form flow nodes. Please refer to [link / reference]. Figure 3C As shown;

[0124] Step 203: Save the settings. Double-click a node in the graph to configure the parameters for each node. These parameters are from the configuration files of the respective frameworks (Paddle, TensorFlow, and PyTorch). For detailed configuration instructions, please refer to the official websites of each framework. Figure 2C As shown; it should be noted that this is the YOLOv3 configuration file of the PyTorch framework. The properties can be found and learned from the official website, and will not be elaborated here.

[0125] Step 204, Startup Process: In the startup panel, select the runtime environment (such as a Docker container) and the runtime. Please refer to [link / reference needed]. Figure 3E As shown;

[0126] Step 205: Observe the progress. After the process is complete, click "Insights" to observe the model training and inference status. Please refer to... Figure 3F As shown;

[0127] Step 206: Output the model to the model repository through the model output operator, and finally publish the model as a model application to provide model services to the outside world.

[0128] This embodiment of the visualized artificial intelligence algorithm orchestration method has at least the following beneficial technical effects: It enables the import of various algorithms into the software system using a visualization platform under offline conditions, and the execution of each algorithm through a certain connection sequence, realizing the operation of the algorithms and their runtime invocation. This allows users to perform visualized operations during orchestration, training, evaluation, and inference processes, reducing the cost of understanding and application, improving model production efficiency, and increasing the efficiency of building the model production line. Simultaneously, the constructed model production line can quickly train new models, improving model production capacity. Furthermore, it allows users to more easily observe the parameters in each operator during the process and view the runtime data of each operator on the page. It provides low coupling for each step in the production line and, according to certain rules, can be adapted to other platforms.

[0129] In some embodiments, please refer to Figure 4 As shown, the present invention also provides a visual artificial intelligence algorithm orchestration device 300, the device comprising:

[0130] Operator development module 301 is configured to develop operators to solidify various inputs, outputs, training, inference, and evaluation in the artificial intelligence process into operator components;

[0131] The orchestration module 302 is configured to load the operator components onto a visual orchestration page built on the React framework, and the visual orchestration page receives user combinations and configurations of the operator components to generate a target model.

[0132] The training module 303 is configured to receive the user's training parameter configuration for the target model from the visualization orchestration page, and to call a container to train the target model based on the training parameter configuration;

[0133] Model publishing module 304 is configured to publish the trained target model as a model application to provide model services to external users.

[0134] The aforementioned visual artificial intelligence algorithm orchestration device enables the orchestration of artificial intelligence algorithm processes using a visual platform under offline conditions. It imports various algorithms into the software system and uses a visual orchestration page built based on the React framework to orchestrate algorithm production models through a certain connection order. It also enables the running of models and runtime invocation, allowing users to perform visual operations during orchestration, training, evaluation, and inference processes, reducing the cost of understanding and application, and improving model production efficiency.

[0135] It should be noted that the specific limitations regarding the visual AI algorithm orchestration device can be found in the limitations of the visual AI algorithm orchestration method described above, and will not be repeated here. Each module in the aforementioned visual AI algorithm orchestration device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0136] According to another aspect of the present invention, a computer device is provided, which may be a server, and its internal structure diagram is shown below. Figure 5 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the aforementioned visual artificial intelligence algorithm orchestration method. Specifically, the method includes the following steps:

[0137] Operators are developed to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence process into operator components;

[0138] The operator components are loaded into a visual orchestration page built on the React framework, and the visual orchestration page receives the user's combination and configuration of the operator components to generate the target model;

[0139] The visualization orchestration page receives the user's configuration of training parameters for the target model, and calls the container to train the target model based on the configuration of training parameters;

[0140] The trained target model is published as a model application to provide model services to external users.

[0141] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, it implements the above-described visualization artificial intelligence algorithm orchestration method, specifically including the following steps:

[0142] Operators are developed to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence process into operator components;

[0143] The operator components are loaded into a visual orchestration page built on the React framework, and the visual orchestration page receives the user's combination and configuration of the operator components to generate the target model;

[0144] The visualization orchestration page receives the user's configuration of training parameters for the target model, and calls the container to train the target model based on the configuration of training parameters;

[0145] The trained target model is published as a model application to provide model services to external users.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0147] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for orchestrating visual artificial intelligence algorithms, characterized in that, The method includes: Operators are developed to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence process into operator components. This includes: inputting operator configurations into the operator development system and storing them in a database; storing operator classifications, anchor point types, algorithm callback functions, algorithm parameter types, and algorithm script functions as the main algorithm parameter information in the database; and generating an operator list based on the data in the database for users to drag and drop for display. The operator components are loaded into a visual orchestration page built on the React framework. The visual orchestration page receives user input and configuration of the operator components to generate the target model. This includes: using jsplumb, jQuery, echarts, crypto-js, and axios to design a page algorithm library and a process modeling page; loading common operators and algorithms into the page algorithm library and loading the operator list from the database (left-hand menu); during initialization of the process modeling page, caching nodes as the orchestrated node information and connections as the node connection information; and executing `componentDidUpdate` in the React framework to render operator nodes based on nodes and operator connection information based on connections, and configuring the entire orchestration panel as available. The page supports drag-and-drop functionality. It receives operator components dragged from the left menu to the process modeling page, along with directed connection segments and connections to the fixed end node on the right. When an operator is dragged from the left menu to the panel, an object is constructed, adding the node's coordinates, index key, algorithm information, anchor point type, and node name as an object to the page's `nodes` array. During node rendering, the page renders according to the algorithm anchor point type, connects two anchor points, and adds the connection segment information to the page's `connects` array. When saving, the page saves `nodes`, `connects`, process ID, and update time to the database. Upon loading, the page retrieves the `nodes` and `connects` information based on the process ID and loads it to enable post-save secondary development. The visualization orchestration page receives the user's configuration of training parameters for the target model, and calls the container to train the target model based on the configuration of training parameters; The trained target model is published as a model application to provide model services to external users.

2. The visualization artificial intelligence algorithm orchestration method according to claim 1, characterized in that, The method further includes; During training, the backend interface service is called to store the model file, file address, model description, and model evaluation mAP into the model library.

3. The visualization artificial intelligence algorithm orchestration method according to claim 1, characterized in that, The method further includes: Provides a process insight interface to show users the training data, evaluation data, inference process data, and result data generated during model training.

4. The visualization artificial intelligence algorithm orchestration method according to claim 1, characterized in that, The visual orchestration page receives the user's configuration of training parameters for the target model, and calls a container to train the target model based on the configuration of training parameters, including: Use jsplumb, jquery, echarts, crypto-js, and axios to design process control panels and process progress panels; The workflow control panel in the visual orchestration page provides four buttons: Save, Refresh, Start, and Stop. You can use the Start button to select the container to start the workflow. When starting a process, the process ID from the front end is passed to the back end. The back end retrieves node information and connect information based on the process ID, and then retrieves the input and output endpoints of the connects. Sorting is performed using a topology algorithm, and the corresponding sorted nodes are obtained. Based on the node sorting and algorithm information, load the page parameter information of each node to form complete process information, construct a context object, and start a thread from the thread pool to call the algorithm execution container. The page starts a scheduled task to check whether a thread is running in the thread pool and whether a script is being executed in the container. If the process is running, the running log will be fed back to the process progress panel and a progress bar will be provided to display the progress.

5. The visualization artificial intelligence algorithm arrangement method according to claim 1, characterized in that, The target model belongs to at least one of the following: imaging principle, boundary and curve, image classification, image segmentation, target detection, shape analysis, texture analysis, image reconstruction, image generation, and face recognition.

6. A visual artificial intelligence algorithm orchestration device, characterized in that, The device includes: The operator development module is configured to develop operators to solidify various inputs, outputs, training, inference, and evaluation processes in the artificial intelligence workflow into operator components. The orchestration module is configured to load the operator components into a visual orchestration page built on the React framework, and the visual orchestration page receives user combinations and configurations of the operator components to generate the target model. The training module is configured to receive the user's training parameter configuration for the target model from the visualization orchestration page, and to call a container to train the target model based on the training parameter configuration; The model publishing module is configured to publish the trained target model as a model application to provide model services to external users. The operator development module is also used to input the operator configuration into the operator development system and store it in the database; store the operator classification, anchor point type, algorithm callback function, algorithm parameter type, and algorithm script function as the main algorithm parameter information in the database; and generate an operator list based on the data in the database for users to drag and drop to display. The orchestration module is also used to design page algorithm libraries and process modeling pages using jsplumb, jquery, echarts, crypto-js, and axios; load public operators and algorithms in the page algorithm library, and load the operator list from the database to the left-hand menu; during the initialization of the process modeling page, nodes are cached as the node information for the orchestration on the page, and connects are cached as the node connection information for the orchestration on the page; in the componentDidUpdate method of the React framework, the operator nodes on the page are rendered according to nodes, and the operator connection information on the page is rendered according to connects, and the entire orchestration panel is configured to be draggable; it receives operator components dragged by the user from the left-hand menu to the process modeling page. The page includes directed connection segments that receive operator components and connect to the fixed end node on the far right. When an operator is dragged from the left menu to the panel, an object is constructed, and the node's coordinates, index key, algorithm information, anchor point type, and node name are added as an object to the page's nodes array, which serves as the node array. During node rendering, the page renders the nodes according to the algorithm anchor point type. The page connects two anchor points and adds the connection segment information to the page's connects information, which serves as the connects array. When the page saves, it saves nodes, connects, process ID, and update time to the database. When the page loads, it retrieves the associated nodes and connects information based on the process ID and loads it onto the page to enable secondary development after saving.

7. A computer device, characterized in that, include: At least one processor; as well as A memory storing a computer program executable in the processor, wherein the processor executes the program to perform the method according to any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it performs the method described in any one of claims 1-5.

Images