Labeling and training method based on integrated platform
By synchronously generating training task frameworks and automatically converting data formats on an integrated platform, the inefficiency of traditional separate processes is solved, achieving efficient integration of annotation and model training, and improving the development efficiency and performance of artificial intelligence models.
Patent Information
- Application Number
- CN202511062257.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-31
- Publication Date
- 2025-11-14
AI Technical Summary
The traditional separation of data annotation and model training processes leads to low efficiency, high error rates, and difficulty in meeting real-time requirements. The lack of effective collaboration mechanisms also affects the development efficiency and performance of artificial intelligence models.
This paper presents an annotation and training method based on an integrated platform. By creating annotation tasks in the annotation task management module, an associated training task framework is generated synchronously, data formats are automatically converted and training parameters are configured, and the training process is monitored until the target model is obtained.
It achieves seamless integration of annotation and model training, improving efficiency and data quality, and enhancing the development efficiency and performance of artificial intelligence models.
Smart Images
Figure CN120952202A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of machine learning technology, and in particular to a labeling and training method based on an integrated platform. Background Technology
[0002] In today's digital age, the rapid development of artificial intelligence technology has placed higher demands on data labeling and model training.
[0003] Traditional data annotation and model training processes are usually separate. Annotated data needs to undergo tedious processing and transformation before it can be used for model training. This process is not only inefficient but also prone to errors. At the same time, the quality of annotated data is difficult to guarantee, and the lack of effective communication and collaboration mechanisms between the annotation team and the model training team leads to information asymmetry, making it difficult to meet the real-time requirements of model training.
[0004] These problems severely impact the development efficiency and performance of artificial intelligence models, urgently requiring a solution that integrates data annotation and model training to improve the efficiency and data quality of the entire process and better adapt to the rapid development of artificial intelligence technology. Summary of the Invention
[0005] In view of this, the purpose of this invention is to provide a labeling and training method based on an integrated platform.
[0006] In a first aspect, embodiments of the present invention provide a labeling and training method based on an integrated platform, the method comprising:
[0007] In the annotation task management module, an annotation task is created, and an associated training task framework is generated synchronously. The training task framework is pre-bound to the annotation type and algorithm model type, and inherits the material classification system of the annotation task.
[0008] In response to the annotation completion command, the annotation data and the original image are automatically converted into the target algorithm format corresponding to the algorithm model type and injected into the data repository of the algorithm model type;
[0009] Respond to algorithm management commands and configure training parameters; training parameters should include at least the number of cycle counts, the maximum number of unoptimized cycles, and the number of training threads.
[0010] In response to the training command, the server management interface is invoked according to the algorithm model type to start the training task;
[0011] Monitor the training process until the target model is obtained after training.
[0012] In conjunction with the first aspect, the steps for creating a labeling task in the labeling task management module and simultaneously generating the associated training task framework include:
[0013] In response to image management commands, add or delete images to be labeled according to preset material categories;
[0014] In response to the annotation command, switch to the material annotation interface, which includes a material list area, a canvas area, and an annotation preview area arranged in sequence;
[0015] In response to a selection operation on a target material in the material list area, display the target material in the canvas area;
[0016] In response to editing operations on a canvas area, add annotations to the target material;
[0017] In response to the save operation, the labeled target material is saved and synchronized to the database of the labeling task management module.
[0018] In conjunction with the first aspect, before the step of adding or deleting images to be labeled in a preset material category in response to image management instructions, the following steps are also included:
[0019] In response to the instruction to claim a labeling task, update the claiming status of the labeling task to "claimed".
[0020] In conjunction with the first aspect, following the step of adding annotations to the target material in response to editing operations on the canvas area, it also includes:
[0021] In response to the adjustment operation, adjust the image attributes of the target material that has been annotated.
[0022] In conjunction with the first aspect, after the steps of creating a labeling task and synchronously generating the associated training task framework in the labeling task management module, it also includes:
[0023] In response to adjustments made to the algorithm management module, update the algorithm model type with the adjusted algorithm model type of the target algorithm.
[0024] In conjunction with the first aspect, the steps for monitoring the training process include:
[0025] When a training anomaly caused by a labeling error is detected, the problematic data is automatically sent back to the original labeling task queue.
[0026] In conjunction with the first aspect, it also includes a server management module; the method further includes:
[0027] In response to server management operations, establish the association between the server and the target training task, and / or correct the server configuration information.
[0028] Secondly, this application provides an annotation and training device based on an integrated platform, the device comprising:
[0029] The creation module is used to create annotation tasks in the annotation task management module and synchronously generate associated training task frameworks. The training task framework is pre-bound to annotation type and algorithm model type and inherits the material classification system of the annotation task.
[0030] The first response module is used to respond to the annotation completion command, automatically convert the annotation data and the original image into the target algorithm format corresponding to the algorithm model type, and inject it into the data repository of the algorithm model type.
[0031] The second response module is used to respond to algorithm management instructions and configure training parameters; the training parameters include at least the number of cycles, the maximum number of unoptimized cycles, and the number of training threads.
[0032] The third response module is used to respond to training instructions, call the server management interface according to the algorithm model type, and start the training task;
[0033] The monitoring module is used to monitor the training process until the target model is obtained after training is complete.
[0034] Thirdly, this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to cause the electronic device to perform the above-described method.
[0035] Fourthly, this application provides a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the above-described method.
[0036] The embodiments of this invention bring the following beneficial effects: This application provides an annotation and training method based on an integrated platform. The method includes: creating an annotation task in the annotation task management module and synchronously generating an associated training task framework; the training task framework pre-binding the annotation type and algorithm model type, and inheriting the material classification system of the annotation task; responding to the annotation completion instruction, automatically converting the annotation data and the original image into the target algorithm format corresponding to the algorithm model type, and injecting it into the data repository of the algorithm model type; responding to the algorithm management instruction, configuring training parameters; the training parameters include at least the configuration cycle count, the maximum unoptimized count, and the number of training threads; responding to the training instruction, calling the server management interface according to the algorithm model type to start the training task; and monitoring the training process until the target model is obtained after training.
[0037] The annotation and training method based on an integrated platform provided in this application realizes material annotation and model training on the same platform without the need for cumbersome processing and conversion, which helps to improve the efficiency of annotation and model training, thereby improving the development efficiency and performance of artificial intelligence models.
[0038] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.
[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0041] Figure 1 A schematic diagram of the annotation and training method based on an integrated platform provided in an embodiment of the present invention;
[0042] Figure 2 Example diagram of the material management interface of the integrated platform provided in the embodiments of the present invention;
[0043] Figure 3 Example diagram of the annotation task interface of the integrated platform provided in the embodiments of the present invention;
[0044] Figure 4 Example diagram of the material annotation interface of the integrated platform provided in the embodiments of the present invention;
[0045] Figure 5 Example diagram of the training task interface of the integrated platform provided in the embodiments of the present invention;
[0046] Figure 6 Example diagram of the algorithm model interface of the integrated platform provided in the embodiments of the present invention;
[0047] Figure 7 Example diagram of the algorithm version interface of the integrated platform provided in the embodiments of the present invention;
[0048] Figure 8 Example diagram of the server management interface of the integrated platform provided in the embodiments of the present invention;
[0049] Figure 9 A structural diagram of an annotation and training device based on an integrated platform provided in an embodiment of the present invention;
[0050] Figure 10 This is a schematic diagram of the electronic device structure provided in an embodiment of the present invention.
[0051] Figure label:
[0052] 10 - Creation Module, 20 - First Response Module, 30 - Second Response Module, 40 - Third Response Module, 50 - Monitoring Module;
[0053] 130 - Processor, 131 - Memory, 132 - Bus, 133 - Communication interface. Detailed Implementation
[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0055] To facilitate understanding of this embodiment, the application scenarios and design concepts of this application embodiment will be briefly introduced below.
[0056] Traditionally, data annotation and model training processes are usually separate. Annotated data needs to undergo tedious processing and transformation before it can be used for model training, which leads to low training efficiency. First, the lack of collaboration in the model training process makes it difficult to obtain a good model.
[0057] Based on this, this application provides an annotation and training method based on an integrated platform.
[0058] Example 1
[0059] This application provides a labeling and training method based on an integrated platform, combining... Figure 1 As shown, the method includes:
[0060] S110, create a labeling task in the labeling task management module and generate an associated training task framework simultaneously; the training task framework is pre-bound to the labeling type and algorithm model type, and inherits the material classification system of the labeling task.
[0061] S120, in response to the annotation completion command, automatically converts the annotation data and the original image into the target algorithm format corresponding to the algorithm model type, and injects it into the data repository of the algorithm model type.
[0062] S130 responds to algorithm management commands and configures training parameters; training parameters include at least the number of cycles, the maximum number of unoptimized cycles, and the number of training threads.
[0063] S140, in response to the training command, calls the server management interface according to the algorithm model type to start the training task.
[0064] S150 monitors the training process until the target model is obtained after training.
[0065] The method provided in this application first creates a labeling task and generates an associated training task framework. Then, it automatically transforms the data and injects it into a repository. Based on the configured training parameters, it starts the training task after receiving the training instruction and monitors the training process. In this way, labeling and model training are performed on the same platform, reducing the sense of separation in model training operations. Reducing manual intervention helps to improve efficiency, lower the error rate, and improve the accuracy of the model.
[0066] In an integrated platform, multiple annotation tools are typically integrated to meet different task requirements. Before annotation, a dataset must be created and the original data (images to be annotated in this embodiment) uploaded. Step S110 constructs the annotation task and associates it with the training task framework to link annotation with model training. This allows for model training using the annotated material and pre-bound algorithm model type after annotation. Common annotation types include image annotation, speech annotation, text annotation, and video annotation. In this embodiment, the material is the image to be annotated, and the corresponding annotation type is image annotation. In practical applications, the annotation type is determined based on the actual material. An algorithm model refers to a mathematical or computational framework used to solve a specific problem. Common algorithm model types are mainly divided according to task requirements, typically including regression models, classification models, clustering models, dimensionality reduction models, forced learning models, etc., which can be determined based on the model type corresponding to the actual pre-bound algorithm model. The material classification system is usually a hierarchical structure designed to organize and manage data, facilitating efficient retrieval and use. Its typical features include knowledge graph-based classification, classification by data type (different annotation methods and application scenarios correspond to different material types), and classification based on capability systems. It is evident that annotation types, algorithm model types, and material classification systems respectively cover the core aspects of data processing, model training, and resource management, jointly supporting the development, training, and application of models.
[0067] In conjunction with the first aspect, step S110 includes:
[0068] S111, responding to image management commands, adds or deletes images to be labeled in preset material categories.
[0069] S112, in response to the annotation command, switches to the material annotation interface, which includes a material list area, a canvas area, and an annotation preview area arranged in sequence.
[0070] S113, in response to a selection operation on a target material in the material list area, displays the target material in the canvas area.
[0071] S114, responds to editing operations on a canvas area by adding annotations to the target material.
[0072] S115, in response to the save operation, saves the labeled target material and synchronizes it to the database of the labeling task management module.
[0073] The integrated platform provided in this application includes a material management module. This module serves as the parent node and also provides corresponding child nodes for material management, material classification, and task labeling. Clicking on a material management child node generates and displays a material management interface. Figure 2 As shown, this interface displays materials categorized differently from those currently stored in the material library. Users can quickly search for materials by name, status, or category using the search bar at the top. The interface also allows for adding, deleting, batch importing, and batch exporting of materials. The material status indicates whether it has been labeled; if labeled, the status is "Success."
[0074] When a sub-node of the annotation task is clicked, the annotation task interface generated and displayed is as follows: Figure 3 As shown, the interface provides an image management module (combined with "Manage Image Batch" in the figure). Users can enter the image management process by clicking on the image management module, and then upload, delete, and modify the material categories to be labeled through the system interface or API. The system supports dynamically adjusting the material classification system, which is convenient for subsequent inheritance to the training task.
[0075] Subsequently, when the user initiates a labeling task (e.g., clicking...), Figure 3 After clicking the "Enter Annotation Interface" function key, you will be automatically redirected to the material annotation interface. Figure 4 As shown, the interface has the following functional areas: a material list area, a canvas area, and an annotation preview area. The material list area displays thumbnails or identifiers of all materials to be annotated in the current task; the canvas area displays the currently selected material for annotation; and the annotation preview area displays the added annotations and their attribute information in real time.
[0076] The left side of the material annotation interface is the material list area, such as... Figure 4 The interface displays "Current annotation task: test2," along with a list of materials below, with their annotation status shown on the right. The canvas area is located in the center of the material annotation interface. This canvas area typically occupies a larger area than or equal to the sum of the other two areas to enlarge the materials for easier annotation. Figure 3 The borders on the material are the annotations added by the user; the right side of the material annotation interface is the annotation preview area, used to display and preview the annotation content and attribute information added by the user in the canvas area in real time, such as... Figure 4 The attribute information corresponding to the rectangular annotation frame on this material is 1.1, 2.2, 3.3, 4.4, and 5.5. It can be understood that the numbers or letters in the above attribute information can be information with preset meanings. For example, the number "1" represents the presence of a weld, and the letter "D" represents the presence of a dent, etc. The meanings can be preset according to the actual annotation requirements. This is just an example and is not limited.
[0077] Then, based on the user's selection of the target material, the target material is displayed on the canvas area. Click the "Draw Rectangle" function button below the canvas area to add annotations, click the "Edit Mode" function button to add annotation information, click the "Clear All" function button to delete all annotations in the canvas area, and click the "Save" function button to save, thus completing the annotation operation of the target material.
[0078] In conjunction with the first aspect, prior to step S111, the following also includes:
[0079] S101, in response to the instruction to claim the labeling task, update the claiming status of the labeling task to the claimed status.
[0080] Combination Figure 3 The annotation task interface shown also provides a task claiming module (see “Claim Task” in the figure), which allows users to claim annotation image content for that time period or a specified time period according to their work schedule.
[0081] Combination Figure 3 The integrated platform also provides a download module (see "Download" in the image), allowing users to download labeled materials (preferred to be in JSON or TXT format) and the original image content. It also provides a tag management module (see "Manage Tags" in the image), allowing users to manage added tags by modifying their format and specifying appropriate labeling content, such as allowing only numeric tags and prohibiting letter tags. Furthermore, the platform offers a recycling module (see "Recycle" in the image) for recycling materials claimed by users but without added labels; a synchronization module (see "Synchronize" in the image) for synchronizing and reloading materials based on the current material category; and a deletion module (see "Delete" in the image) for deleting labeling tasks.
[0082] In conjunction with the first aspect, after step S114, the following also includes:
[0083] S116, in response to the adjustment operation, adjusts the image attributes of the target material that has been annotated.
[0084] After the user marks the target material, they can click on it as shown in the image. Figure 4 Use the "Image Adjustment" function key below the canvas area shown to adjust image properties, such as image contrast, brightness, and pixel values.
[0085] In conjunction with the first aspect, after step S110, the following also includes:
[0086] In response to adjustments made to the algorithm management module, update the algorithm model type with the adjusted algorithm model type.
[0087] The integrated annotation platform also provides an algorithm management module, which acts as a parent node. The child nodes corresponding to this parent node include algorithm model modules and training task modules. These modules are generated and displayed by clicking on the algorithm model child node. Figure 6 The algorithm model interface shown displays the names of multiple algorithm models, their corresponding ports (port mapping is a crucial step in deploying large machine learning models; by configuring environment variables and starting the HTTP service, the model can be bound to a specified port to provide services), IP addresses, paths, status (which can refer to configuration status, with "success" indicating successful configuration), and remarks. By selecting a particular algorithm model, the user determines the target algorithm model based on their needs, thus defining the algorithm model type. Subsequent labeling and training operations are then performed using this target algorithm model. The algorithm model type is synchronized to the integrated platform during selection, and the selected target algorithm model is initialized.
[0088] It's worth noting that before adding or using a new algorithm model, it needs to be registered first. This involves building and storing the association between the new algorithm model and its type. For details, please click [link / link]. Figure 6 The "Add" function key allows you to add algorithms, which will not be elaborated upon here. Similarly, you can quickly search by algorithm model name, status, or other information directly in the search bar above, or the author can modify the notes by clicking the "Edit" function key, or delete a saved algorithm model by clicking the "Delete" function key, which will not be elaborated upon here.
[0089] Combination Figure 5 As shown, when the user clicks on the training task module (i.e. Figure 5 When selecting a "training task" (as shown in the image), the training task interface is displayed, which provides an attribute editing module, specifically as follows: Figure 5 The "Modify" and "Delete" functions shown above allow you to modify or delete task names or notes using the aforementioned modules.
[0090] Combination Figure 5As shown, the training task interface also provides a module for managing algorithm versions (combined with...). Figure 7 In the "Manage Algorithm Versions" section, click on this module to switch to the algorithm version interface. Figure 7 The interface shown is for configuring the algorithm model. Specifically, this interface provides a module for configuring training parameters (combined with...). Figure 7 In the "Configure Training Parameters" section, clicking this module allows you to configure necessary training information such as the number of iterations, maximum unoptimized iterations, and number of training threads within a pop-up window. The configured information is then sent to the algorithm for data processing and model training during the training process. Figure 7 As shown, this provides a logging module (in conjunction with...) Figure 7 The "View Training Log" option allows you to retrieve information returned by the algorithm, such as error messages and accuracy rates.
[0091] Combination Figure 7 As shown, the training task interface also displays the selected test version, remarks, the selected target algorithm model, and the training progress displayed in a specified graphical format. It also provides a "Delete" function key, which can be clicked to delete the training task; a "Manage Materials" function key, which can be clicked to switch to the material management interface to import, delete, and modify labeled materials; and a "Start Training" function key, which can be clicked to start the model training process.
[0092] Understandably, step S110 is used to create the annotation task and associate the training task framework. Then, after the annotation is completed in step S120, the model is transformed to adapt to the algorithm model type. Then, based on the training parameters configured in step S130, step S140 is used to train the model. And step S150 is used to monitor the training process until the target model is obtained.
[0093] In conjunction with the first aspect, step S150, monitoring the training process, specifically includes:
[0094] S151: When a training anomaly caused by a labeling error is detected, the problematic data is automatically sent back to the original labeling task queue.
[0095] Understandably, when anomalies occur during training, the cause of the anomaly should be analyzed based on the log information fed back by the algorithm platform. If the training anomaly is caused by labeling errors, the problematic data should be sent back to the original labeling task queue so that the user can correct the labeling before model training can proceed.
[0096] In conjunction with the first aspect, it also includes a server management module; the method further includes:
[0097] In response to server management operations, establish the association between the server and the target training task, and / or correct the server configuration information.
[0098] Understandably, different algorithm models may be deployed on different servers. Therefore, by managing server addresses, different algorithm models can be selected. "Server Management" serves as the parent node, with child nodes including "Server" and "Server Configuration." Clicking the "Server" node generates... Figure 8 The server management interface shown allows users to modify the associations of Prometheus tasks associated with the server by clicking the "Modify" button, and to remove these associations by clicking "Delete". Clicking the "Server Configuration" node generates a configuration information display interface (not shown), which displays the corresponding algorithm's operation type, unit, number of decimal places, etc. The display status can be modified using the "Modify" button, and the configuration information can be deleted by clicking "Delete". This is a standard technique and will not be elaborated upon here.
[0099] Secondly, this application provides a labeling and training device based on an integrated platform, combining... Figure 9 As shown, the device includes: a creation module 10, a first response module 20, a second response module 30, a third response module 40, and a monitoring module 50.
[0100] The creation module 10 is used to create annotation tasks in the annotation task management module and synchronously generate associated training task frameworks. The training task framework is pre-bound to annotation type and algorithm model type and inherits the material classification system of the annotation task.
[0101] The first response module 20 is used to respond to the annotation completion command, automatically convert the annotation data and the original image into the target algorithm format corresponding to the algorithm model type, and inject it into the data repository of the algorithm model type.
[0102] The second response module 30 is used to respond to algorithm management instructions and configure training parameters; the training parameters include at least the number of configuration cycles, the maximum number of unoptimized cycles, and the number of training threads.
[0103] The third response module 40 is used to respond to training instructions, call the server management interface according to the algorithm model type, and start the training task.
[0104] The monitoring module 50 is used to monitor the training process until the target model that has been trained is obtained.
[0105] Thirdly, embodiments of this application provide an electronic device, combined with Figure 10 As shown, the electronic device includes a memory 131 and a processor 130. The memory 131 stores a computer program, and the processor 130 runs the computer program to make the electronic device perform the above-described method.
[0106] Furthermore, combined Figure 10 The electronic device shown also includes a bus 132 and a communication interface 133, with the processor 130, the communication interface 133 and the memory 131 connected via the bus 132.
[0107] The memory 131 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 133 (which can be wired or wireless), such as the Internet, wide area network, local area network, or metropolitan area network. The bus 132 may be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 10 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.
[0108] Processor 130 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 130 or by instructions in software form. Processor 130 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 131, and processor 130 reads the information in memory 131 and, in conjunction with its hardware, completes the steps of the method described in the foregoing embodiments.
[0109] Fourthly, embodiments of this application provide a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the above-described method.
[0110] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0111] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.
[0112] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0113] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.
[0114] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A labeling and training method based on an integrated platform, characterized in that, The method includes: In the annotation task management module, an annotation task is created, and an associated training task framework is generated synchronously. The training task framework is pre-bound to the annotation type and the algorithm model type, and inherits the material classification system of the annotation task. In response to the annotation completion command, the annotation data and the original image are automatically converted into the target algorithm format corresponding to the algorithm model type and injected into the data repository of the algorithm model type; In response to algorithm management instructions, configure training parameters; the training parameters include at least the number of cycle counts, the maximum number of unoptimized cycles, and the number of training threads. In response to the training instruction, the server management interface is invoked according to the algorithm model type to start the training task; Monitor the training process until the target model is obtained after training.
2. The method according to claim 1, characterized in that, The steps for creating a labeling task and synchronously generating the associated training task framework in the labeling task management module include: In response to image management commands, add or delete images to be labeled according to preset material categories; In response to the annotation command, switch to the material annotation interface, which includes a material list area, a canvas area, and an annotation preview area arranged in sequence; In response to a selection operation on a target material in the material list area, the target material is displayed in the canvas area; In response to an editing operation on the canvas area, add annotations to the target material; In response to the save operation, the labeled target material is saved and synchronized to the database of the labeling task management module.
3. The method according to claim 2, characterized in that, Before the steps of adding or deleting images to be labeled in the preset material categories in response to image management commands, the following are also included: In response to the instruction to claim a labeling task, the claiming status of the labeling task is updated to "claimed".
4. The method according to claim 2, characterized in that, Following the step of adding annotations to the target material in response to an editing operation on the canvas area, the method further includes: In response to the adjustment operation, adjust the image attributes of the target material that has been annotated.
5. The method according to claim 1, characterized in that, The integrated platform also includes an algorithm management module; After the steps of creating a labeling task and synchronously generating the associated training task framework in the labeling task management module, the following are also included: In response to the adjustment operation of the algorithm management module, the algorithm model type is updated with the algorithm model type of the adjusted target algorithm.
6. The method according to claim 1, characterized in that, The steps for monitoring the training process include: When a training anomaly caused by a labeling error is detected, the problematic data is automatically sent back to the original labeling task queue.
7. The method according to claim 1, characterized in that, It also includes a server management module; the method further includes: In response to server management operations, establish the association between the server and the target training task, and / or correct the configuration information of the server.
8. A labeling and training device based on an integrated platform, characterized in that, The device includes: A creation module is used to create annotation tasks in the annotation task management module and synchronously generate associated training task frameworks; the training task framework is pre-bound to annotation type and algorithm model type, and inherits the material classification system of the annotation task; The first response module is used to respond to the annotation completion command, automatically convert the annotation data and the original image into the target algorithm format corresponding to the algorithm model type, and inject it into the data repository of the algorithm model type. The second response module is used to respond to algorithm management instructions and configure training parameters; the training parameters include at least the number of configuration cycles, the maximum number of unoptimized cycles, and the number of training threads. The third response module is used to respond to training instructions by calling the server management interface according to the algorithm model type to start the training task; The monitoring module is used to monitor the training process until the target model is obtained after training is complete.
9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program and the processor running the computer program to cause the electronic device to perform the method of any one of claims 1 to 7.
10. A storage medium, characterized in that, The storage medium stores computer program instructions, which, when read and executed by a processor, perform the method described in any one of claims 1 to 7.