Data processing method and device for training artificial intelligence model and related equipment

Abstract

The invention provides a data processing method and device for training an artificial intelligence model and related equipment. Relates to the technical field of computers, in particular to the technical fields of intelligent interaction, databases and the like. According to the specific implementation scheme, a to-be-processed data set in a source database is processed based on a first target ETL pipeline supporting visual configuration, so that a to-be-labeled task suitable for storage of a target database is split from the source database; data is stored in the source database in a file form; the target database stores data in the form of a data table; labeling the to-be-labeled task based on the target database to obtain labeled data; and processing the labeled data based on the second target ETL pipeline, and storing the labeled data in the source database.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technologies, and in particular, to the fields of intelligent interaction, databases, and other technologies. Background Art

[0002] Currently, the training of artificial intelligence models requires a large amount of data, so distributed file systems are often considered for such data. In addition, the data required for the training of artificial intelligence models often needs to be manually annotated in large quantities. Summary of the Invention

[0003] The present disclosure provides a data processing method, apparatus, and related equipment for training an artificial intelligence model to solve or alleviate one or more technical problems in the prior art.

[0004] In a first aspect, the present disclosure provides a data processing method for training an artificial intelligence model, including: Processing a set of data to be processed in a source database based on a first target ETL pipeline that supports visual configuration, so as to split out a task to be annotated suitable for storage in a target database from the source database; the source database stores data in the form of files; the target database stores data in the form of data tables; Annotating the task to be annotated based on the target database to obtain annotated data; Processing the annotated data based on a second target ETL pipeline and storing the annotated data in the source database.

[0005] In a second aspect, the present disclosure provides a data processing apparatus for training an artificial intelligence model, including: A first processing module, configured to process a set of data to be processed in a source database based on a first target ETL pipeline that supports visual configuration, so as to split out a task to be annotated suitable for storage in a target database from the source database; the source database stores data in the form of files; the target database stores data in the form of data tables; An annotation module, configured to annotate the task to be annotated based on the target database to obtain annotated data; A second processing module, configured to process the annotated data based on a second target ETL pipeline and store the annotated data in the source database.

[0006] In a third aspect, an electronic device is provided, including: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any method in the embodiments of the present disclosure.

[0007] In a fourth aspect, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute any method in the embodiments of the present disclosure.

[0008] In a fifth aspect, a computer program product is provided, including a computer program which, when executed by a processor, implements any method in the embodiments of the present disclosure.

[0009] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. Description of the Drawings

[0010] In the drawings, unless otherwise specified, the same reference numerals throughout the several drawings denote the same or similar components or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments provided in accordance with the present disclosure and should not be regarded as limiting the scope of the present disclosure.

[0011] Figure 1 is a schematic diagram of a scenario of a data processing method for training an artificial intelligence model according to a first embodiment of the present disclosure; Figure 2 is a schematic flowchart of a data processing method for training an artificial intelligence model according to a second embodiment of the present disclosure; Figure 3 is a schematic flowchart of a data processing method for training an artificial intelligence model according to a third embodiment of the present disclosure; Figure 4 is a schematic diagram of a visualization interface of a data processing method for training an artificial intelligence model according to a fourth embodiment of the present disclosure; Figure 5 is a schematic diagram of a visualization interface of a data processing method for training an artificial intelligence model according to a fifth embodiment of the present disclosure; Figure 6 is a schematic diagram of a visualization interface of a data processing method for training an artificial intelligence model according to a sixth embodiment of the present disclosure; Figure 7 is a schematic diagram of a visualization interface of a data processing method for training an artificial intelligence model according to a seventh embodiment of the present disclosure; Figure 8Schematic diagram of the visualization interface for the data processing method for training an artificial intelligence model according to the eighth embodiment of the present disclosure; Figure 9 Schematic diagram of the visualization interface for the data processing method for training an artificial intelligence model according to the ninth embodiment of the present disclosure; Figure 10 Schematic flow diagram of the data processing method for training an artificial intelligence model according to the tenth embodiment of the present disclosure; Figure 11 Schematic diagram of the visualization interface for the data processing method for training an artificial intelligence model according to the eleventh embodiment of the present disclosure; Figure 12 Schematic diagram of the visualization interface for the data processing method for training an artificial intelligence model according to the twelfth embodiment of the present disclosure; Figure 13 Schematic diagram of the data processing method for training an artificial intelligence model according to the thirteenth embodiment of the present disclosure; Figure 14 Schematic diagram of the structure of the data processing device for training an artificial intelligence model according to the fourteenth embodiment of the present disclosure; Figure 15 Block diagram of the electronic device for implementing the data processing method / home model design method for training an artificial intelligence model according to the embodiments of the present disclosure. Detailed implementation manners

[0012] The present disclosure will be further described in detail below with reference to the accompanying drawings. The same reference numerals in the drawings denote elements having the same or similar functions. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.

[0013] In addition, for a better illustration of the present disclosure, numerous specific details are given in the following detailed implementation manners. Those skilled in the art should understand that the present disclosure can be implemented without some of these specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art are not described in detail so as to highlight the gist of the present disclosure.

[0014] Furthermore, the terms "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present disclosure, "a plurality" means two or more unless otherwise specifically defined.

[0015] It should be noted that, unless it is clearly stated that there is a sequential execution order between different operations in the flowcharts shown in the embodiments of the present disclosure, or there is a sequential execution order between different operations in terms of technical implementation, the execution order between multiple operations can be unordered, and multiple operations can also be executed simultaneously.

[0016] Since a distributed file system can store a large amount of data, the unlabeled data used for training an artificial intelligence model is generally stored using a distributed file system.

[0017] In order to quickly and conveniently perform annotation processing on the unlabeled data in the distributed file system, data transmission between two databases is involved in the embodiments of the present disclosure, including a source database and a target database. The source database stores data in the form of files. Exemplarily, it can be HDFS (Hadoop Distributed File System), which is a distributed file system with high fault tolerance and suitable for storing large-scale data. As the core component of the Hadoop ecosystem, HDFS adopts a master-slave structure, divides data into multiple blocks, and stores them distributedly on multiple nodes, supporting data redundancy and fault tolerance across servers. HDFS is suitable for storing large-scale batch processing data, and there is a problem that when a part of the data is updated, the entire data needs to be updated.

[0018] Among them, the target database stores data in the form of data tables. Exemplarily, the target database can be a MySQL database. MySQL is an open-source relational database management system that supports the storage and management of structured data. It uses SQL (Structured Query Language) to perform operations such as adding, deleting, modifying, and querying data, and is widely used in data persistence and transaction processing scenarios in fields such as the Internet, finance, and e-commerce. Compared with a database stored in the form of files, the MySQL database is easier to update data, but its storage capacity may be smaller.

[0019] The embodiments of the present disclosure propose a data processing method for training an artificial intelligence model, as Figure 1 shown in the schematic diagram of the application scenario of this method. Figure 1 It includes a server 11 and a terminal device 12.

[0020] Among them, the terminal device 12 and the server 11 are connected through a wireless or wired network. The terminal device 12 includes, but is not limited to, electronic devices such as desktop computers, mobile phones, mobile computers, tablet computers, media players, smart wearable devices, and smart TVs. The server 11 can be a single server, a server cluster composed of several servers, or a cloud computing center. The server 11 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0021] In the embodiments of the present disclosure, the terminal device 12 is supported to configure an ETL (Extract, Transform, Load) pipeline, and provide it to the source database and the target database in the server 11 for data transmission. The terminal device 12 is also used to provide a visual interface to perform annotation processing on the data transmitted to the target database.

[0022] Such as Figure 2 shown, it is a schematic flowchart of a data processing method for training an artificial intelligence model provided by the embodiments of the present disclosure, including the following content: S201, process the set of data to be processed in the source database based on the first target ETL pipeline that supports visual configuration, so as to split out the annotation tasks to be labeled that are suitable for storage in the target database; the source database stores data in the form of files; the target database stores data in the form of data tables.

[0023] During implementation, it is necessary to transfer the set of data to be processed originally stored in the source data to the target database. Therefore, an ETL pipeline is established between the source database and the target database for data transmission between the two databases.

[0024] The ETL pipeline configured based on the visual interface enables non-technical personnel to directly configure parameters on the visual interface to complete the construction of the data processing flow, reducing the dependence on professional developers and lowering the usage threshold. In addition, visual configuration enables users to quickly complete the configuration and adjustment of the ETL pipeline. Modifying and adjusting in the visual interface makes the entire configuration process more intuitive and efficient.

[0025] S202, perform annotation on the annotation tasks to be labeled based on the target database to obtain the annotated data.

[0026] Among them, annotating the task to be annotated refers to the process of manually or automatically adding auxiliary information for training a machine learning model to the original data (such as images, texts, voices, videos, etc.). The annotation information includes but is not limited to class labels, bounding boxes, key points, semantic segmentation labels, sentiment tendencies, named entities, etc. Data annotation is the premise for constructing a supervised learning model, and the annotation quality has a crucial impact on the model performance, and is widely used in fields such as computer vision, natural language processing, and speech recognition.

[0027] In the home furnishing field, the types of tasks to be annotated include but are not limited to annotating the position of the home furnishing model in the picture (for example, using a bounding box to annotate the position of the sofa in the image), annotating the category of the home furnishing product picture (such as living room furniture, bedroom furniture, kitchen utensils, etc.), and the attributes of the home furnishing model (such as annotating the color, material, style, etc. of the sofa).

[0028] S203, process the annotated data based on the second target ETL pipeline and store the annotated data in the source database.

[0029] Among them, the second target ETL pipeline is the reverse operation of the first target ETL pipeline. For example, the first target ETL pipeline is responsible for transferring the data in the source database to the target database, while the second target ETL pipeline is responsible for transferring the data in the target database to the source database and storing it according to the requirements of the corresponding fields in the source database, and the requirements include but are not limited to naming and storage methods.

[0030] In the embodiments of the present disclosure, since the source database is stored in the form of files, any update to a field in this type of database requires the update of the entire file. Therefore, this type of database cannot be compatible with and support field-level annotation tasks. Alternatively, when this type of database supports field-level annotation tasks, updating this type of database will waste a lot of computing resources. Moreover, the source database usually stores a large amount of raw data. Directly performing frequent data update operations on the source database may have a greater impact on its performance, resulting in slower query and write operations. By transferring data to the target database through the ETL pipeline for annotation and update, the direct write operations to the source database can be reduced. In addition, by splitting the data in the source database through the ETL pipeline and storing it in the target database for annotation, the target database is stored in the form of data tables. However, the storage capacity of this type of database is small and it supports field-level updates. Therefore, only the tasks to be annotated can be processed and updated, rather than performing large-scale update operations on the entire database. This greatly reduces the amount of data that needs to be updated, thereby reducing the occupation of system resources during the update process. At the same time, the processing process of the first target ETL pipeline can well adapt to the storage structure characteristics of different databases. Through the first target ETL pipeline, data in the form of files can be converted into a structure suitable for storage in data tables, or through the second target ETL pipeline, the data in the data tables can be converted back into the form of files, realizing a seamless conversion between the two storage structures, giving full play to their respective advantages, and improving the flexibility of data storage and management. In addition, storing the tasks to be annotated in the data tables of the target database provides a structured and clear annotation environment for annotators. The form of the data tables is convenient for annotators to understand and operate, which can improve the efficiency and accuracy of annotation, and is also conducive to the management of the annotation process. In summary, the process based on the ETL pipeline strengthens the cooperation relationship between the source database and the target database, breaks the data silos, and realizes the flow and sharing of data between different systems. At the same time, based on the solution proposed in the embodiments of the present disclosure, it helps to build a more integrated and collaborative data ecosystem, improving the data processing ability and business value of the entire system.

[0031] In some embodiments, the steps of the first target ETL pipeline with visual configuration are roughly divided into four, namely, configuring basic information, configuring data destination, configuring annotation types, and configuring personnel management. These four steps will be described in detail later.

[0032] In some embodiments, based on the first target ETL pipeline that supports visual configuration, the set of data to be processed in the source database is processed to split out the tasks to be annotated that are suitable for storage in the target database, as Figure 3 shown, and it can be specifically implemented as follows: S301. In response to a first configuration operation on a set of data to be processed in a source database in a first visualization interface, obtain at least one source field in the set of data to be processed.

[0033] Among them, an exemplary first visualization interface is as Figure 4 shown, which is the configuration basic information in the foregoing steps. The task name can be set based on the actual situation. Exemplarily, as Figure 4 shown, it can be set as a test task. Figure 4 The task description in can be to configure the relevant content of the task. Exemplarily, it can be which contents need to be marked in the task, which type the marking task belongs to, etc. In addition, when the task description does not need to be filled in, it can also not be filled in. The data source represents the name of the set of data to be processed in the source database. In addition, the location information of the set of data to be processed can also be uploaded as the data source. After uploading the set of data to be processed, at least one source field can be parsed from the set of data to be processed. As Figure 4 shown in , the first row is at least one source field, which exemplarily includes designId (used to represent the name of the design), k2d_vis (used to represent the display method, coordinates, style and other details of the data to be marked in the two-dimensional space), etc.

[0034] S302. In response to a second configuration operation on a specified source field among at least one source field in a second visualization interface, determine the field attributes of at least one target field corresponding to the specified source field in a target database, and obtain the to-be-marked tasks of each target field.

[0035] In implementation, the second visualization interface includes the foregoing operation interface for configuring the data destination and the marking type. The user can configure the attributes corresponding to any source field in the target database in the second visualization interface, and then can generate the target field and the corresponding to-be-marked tasks based on these attributes.

[0036] In the embodiments of the present disclosure, the method of obtaining source fields through a visualization interface can obtain source fields in a set of data to be processed more flexibly and conveniently compared with the method of obtaining source fields using coding in the traditional method, and determine the target field attributes in the target database, providing a good foundation for subsequent splitting of marking tasks according to different requirements. In addition, in the case of changes in business requirements, using the coding method requires professionals to rewrite and debug the code again, which is time-consuming and laborious. The visualization configuration method can modify the corresponding configuration in the interface to quickly adjust the data processing flow and field mapping rules. Therefore, obtaining the to-be-marked tasks based on the visualization interface can reduce the technical threshold and improve flexibility and adaptability.

[0037] In some embodiments, in response to a second configuration operation on a specified source field in at least one source field in the second visualization interface, the field attributes of at least one target field corresponding to the specified source field in the target database are determined, and the annotation tasks for each target field are obtained. Specifically, it can be implemented as follows: Step A1, in response to a first sub-operation of creating a target table name in the second visualization interface, a target data table corresponding to the target table name is created in the target database.

[0038] The first sub-operation of creating a target table name in the second visualization interface, the target table name refers to the name of the table used to store specific data created in the target database, which is the unique identifier of the target data table, used to distinguish different data tables and facilitate the management and operation of data. The target data table is a structured table that actually stores data in the target database, consisting of multiple fields (columns), and each field has a specific data type and attribute. The target data table is used to store the set of data to be processed extracted from the source database, providing a data basis for subsequent annotation tasks. As Figure 5 shown, it is the operation of configuring the data destination in the foregoing steps. At the Figure 5 target table position in, the location information used to store the set of data to be processed can be input. The data source is similar to the foregoing, which is equivalent to the location information of the data to be processed in the source database.

[0039] Step A2, in response to a second sub-operation of configuring the field attributes of the target field for the specified source field in the second visualization interface, map the field attributes of the specified source field to the target data table to obtain the annotation tasks for each target field; Among them, the field attributes of the target field include: target field index, target field name, target field type, target field value, annotation type, and display type; The display type includes any one of the following: picture type, web page type, 3D home improvement model type.

[0040] Among them, the picture type means that the annotation task is displayed to the user in the form of pictures (such as a thumbnail list). In the annotation task, the user can refer to the pictures for annotation. Such as operations such as drawing bounding boxes and marking key points, and can also annotate the style of the home and the home model in the pictures.

[0041] The web page type means that the annotation task is displayed to the user in the form of a web page. This type is usually used to display information such as text and tables. In the annotation task, the web page type can be used to correctly display the data to be annotated.

[0042] The 3D home improvement model type means that the annotation task is presented to the user in the form of a 3D model, which is usually used to display 3D spatial information such as interior design and architectural design. Based on this type, the 3D structure and spatial relationship of the home model can be accurately and intuitively displayed. In the annotation task, furniture, decorations, etc. in the 3D home improvement model can be annotated, such as position, size, material, etc.

[0043] Among them, the target field index, target field name, and target field type included in the field attributes of the target field can be executed based on Figure 5 the control of "Add Target Field" in, and then a pop-up window will appear as Figure 6 where the target field index, target field name, and target field type can be defined. Among them, the target field index is a number used to uniquely identify the target field in the target database. The target field name is the name of the target field in the target database, which is used to identify and distinguish this field from other fields. The target field type is the type of data that can be stored in the target field. If the data to be processed is text, the target field type can be TEXT.

[0044] In addition, the fields in the source database may include fields that are irrelevant to the annotation task or have poor data quality. The annotation task usually only needs to focus on specific parts of them, rather than all fields. Different types of annotation tasks have different data requirements. According to the specific task requirements, only mapping the necessary fields can customize the construction of the target data table to make it more suitable for the task requirements and reduce the storage of excessive redundant data. Therefore, only the specified part of the source field names in the source database can be mapped. Of course, if the task requirements require mapping all the source field names, all the source field names can be mapped to the target fields.

[0045] It should be noted that at the target field index, the primary key needs to be defined. If none of the target field indexes are defined as the primary key, then any configured target field will be selected as the primary key to ensure the integrity of the source data.

[0046] The target field value, annotation type, and display type included in the field attributes of the target field can be defined based on Figure 7 where the target field value is any of the previously configured target field values. The display type is the form in which the source data is displayed, which can be determined based on the actual situation. The annotation type is the type of the annotation task. Some annotation types can be created in advance, and here you can select from the names of the annotation types corresponding to the pre-created annotation tasks.

[0047] It can be based on Figure 8In the visualization interface, annotation tasks can be created. Among them, the name of the annotation type, the type of annotation operation, and the content of the annotation operation need to be defined. The name of the annotation type represents the name of the annotation task and is used to distinguish it from other annotation tasks. The type of annotation operation indicates whether the annotation operation is a normal annotation (i.e., annotation in text form) or a picture partition annotation (dividing the picture into multiple regions and annotating each region separately). For the content of the annotation operation, it is necessary to specify whether it is required, the annotation item identifier, the annotation item name, the annotation item type, the annotation component type, the default value, the range and format of restricted input values, etc. "Whether it is required" is used to indicate whether the annotator must fill it in. Usually, it is represented by "Yes" or "No". The "annotation item identifier" is used to represent the unique identifier of the annotation item, that is, it is used for internal identification and management of the annotation item. Usually, it is a short string, such as 1, 2, 3, or it can also be key1, key2, key3, etc. The "annotation item name" is used to represent the readable name of the annotation item and is used to display it to the annotator in the annotation interface. It should be concise and clear, and describe the content of the annotation item clearly. The "annotation item type" is used to represent the data type of the annotation item, such as text, number, enumeration, etc. The "annotation component type" is used to represent the input component type used by the annotation item in the annotation interface, such as text box, drop-down menu, checkbox, etc. The "default value" is used to represent the default input when there is no manual input for the annotation item. It can improve the annotation efficiency and reduce the workload of the annotator. The "range and format of restricted input values" is used to restrict the input value of the annotation item and specify the allowed value range or format. It is set based on the actual situation. Exemplarily, it can be the annotation of the style of indoor furniture and the indoor dimensions, etc.

[0048] Each annotation task can be created based on Figure 8 to obtain the name of the annotation task and the annotation type. Each time based on Figure 8 a single annotation task can be created. In the case of creating N annotation tasks, during the execution of Figure 7 , the name of the annotation type corresponding to the target annotation task needed can be selected from the N annotation tasks as the annotation type of the target field, where N is a positive integer greater than 0.

[0049] In addition, the preview thumbnail column name and the preview title name can also be set according to the actual situation. Since generally multiple tasks to be annotated need to be processed, during the annotation process, the unannotated tasks can be displayed in the form of thumbnails or lists, and then it is necessary to specify the preview thumbnail column name, that is, determine which target field needs to be displayed, and the preview title name used to display this field.

[0050] In the embodiment of the present disclosure, by mapping the field attributes of the specified source field to the target data table, the data in the source database can be stored in the target database. At the same time, the field attributes of the target field are configured to obtain clear and specific tasks to be labeled, which lays the foundation for the subsequent generation of labeling tasks and improves the accuracy of generating tasks to be labeled.

[0051] In some embodiments, many users are unfamiliar with which display types are suitable for different data types. Leaving them to figure this out on their own would waste considerable time and effort. Therefore, to ensure that each piece of data to be processed is displayed in an appropriate manner, the following steps can be implemented: analyzing the target field value corresponding to the target field to obtain a recommended list of display types that are suitable for the target field value; and displaying the recommended list corresponding to the target field in the second visualization interface.

[0052] During implementation, if the source data is in a format such as jpg or png, it can be determined that the data to be processed is likely an image, and therefore the recommended display type list may be image. If the source data is http.www.com, it can be determined that the data to be processed is likely a webpage, and the recommended display type list may be webpage. Other types can be determined based on actual circumstances and are not limited in this embodiment.

[0053] In the disclosed embodiments, this approach eliminates the need to waste time selecting a display type; instead, the most appropriate display type can be directly selected from a recommended list. This allows for better adaptation to the characteristics of the data being processed and the annotation requirements, thereby reducing display errors caused by improperly selected display types. The resulting tasks to be annotated can then be displayed in a reasonable manner, improving annotation accuracy and efficiency.

[0054] In some embodiments, the annotation task scope of the annotation personnel may also be determined in response to a third configuration operation on the annotation personnel in the third visualization interface.

[0055] During implementation, the third visual interface is exemplarily as follows Figure 9 As shown in , it includes the annotator and the filtering conditions. The annotator is the person who performs the task to be annotated, and the filtering conditions are which tasks in the task to be annotated can be seen by the annotator. Figure 9 As shown, the target field value of the task to be labeled (i.e., the designld set above) can be used for division, GT means greater than, Figure 9 The annotation range in indicates that the tasks to be annotated greater than the 1000th task in the target field value are visible to the annotators.

[0056] In addition, conditions or groups of conditions can be added based on the actual situation. Exemplarily, based on the control for "adding conditions", another condition is added, and designld less than 2000 is also added to this group of conditions. When the relationship within this group is AND, the annotation tasks from the 1000th to the 2000th in the target field value can be made visible to the personnel to be annotated. In addition, when the relationship between the two conditions within this group is OR, either of the two conditions being met can be visible to the annotator. Exemplarily, when designld less than 2000 and designld greater than 1000 are created, then any of the annotation tasks within the first 2000 tasks less than 2000 can be visible to the annotator.

[0057] In addition, another group of conditions can be created based on the control for "adding a group of conditions". Exemplarily, it can be used to restrict another target field name.

[0058] In the embodiments of the present disclosure, the annotation task scope is reasonably allocated according to the capabilities, experience, workload, etc. of the annotators, and the annotation task scope of the annotators is reasonably determined, realizing refined management of the annotation work, enabling each annotator to clearly know the annotation content they are responsible for, and reducing problems such as duplication, omission, or confusion of annotation tasks.

[0059] In some embodiments, in the case of obtaining the annotation tasks stored in the target database as described above, a visualization representation json is configured to visualize the annotation tasks on the terminal device, so as to facilitate subsequent annotation of the annotation tasks by the annotators.

[0060] In some embodiments, based on the target database, the annotation tasks are annotated to obtain the annotated data, as Figure 10 shown, including: S1001, when the display type of the annotation task is a three-dimensional home improvement model type, parse the annotation task from the target database to obtain the three-dimensional home improvement model in the annotation task, each construction unit in the three-dimensional home improvement model, and the annotation status of each construction unit.

[0061] Among them, the three-dimensional home improvement model is the entire house structure. Among them, the construction unit can exemplarily be any space such as a bedroom, a living room, a kitchen, etc. It can also be any component in the three-dimensional home improvement model, thus supporting component-level annotation. For example, cabinet doors, handles, decorative lights, etc. in the kitchen.

[0062] S1002, output the annotation status of each building unit, the 3D home improvement model, and the annotation task parameters of the 3D home improvement model to the annotation interface; among them, the annotation status is marked or unmarked; among them, the annotation status of each building unit is used to be displayed in the first display area of the annotation interface; the 3D home improvement model is used to be displayed in the second display area of the annotation interface; the annotation task parameters are used to be displayed in the third display area of the annotation interface; the second display area supports preset operations on each building unit.

[0063] During implementation, as Figure 11 shown, from left to right are the first display area, the second display area, and the third display area.

[0064] The first display area is to display based on the target field value preset in the preview thumbnail column name and the title name preset in the preview title name. The display type corresponding to the target field value in the preview thumbnail column name is the image type, and in the case where the preview title name is preset as a picture list, it is displayed as a picture list in this area.

[0065] The second display area is to display the 3D home improvement model, and the annotator can perform preset operations on it, such as rotation, scaling, translation, etc., which is convenient for the annotator to observe and annotate the model from different angles. In addition, in the second display area, any space in the 3D home improvement model (such as the living room) can be clicked to display the space, and at the same time, preset operations can also be performed on the space, such as rotation, scaling, translation, etc.

[0066] The second display area also includes display settings, building management, building information, etc. Among them, display settings refer to a series of adjustable options provided during the display of the 3D home improvement model, which are used to control and optimize the visualization effect of the model so that the annotator can perform annotation work more efficiently and accurately. Exemplarily, as Figure 11 shown, it includes hiding the house type, measuring angles, measuring areas, measuring side lengths, etc. Among them, the function of hiding the house type is used to set part or the entire house type model to be invisible, only retaining the specific area or building unit that needs to be annotated. The function of measuring angles is used to measure and display the included angle between the specified sides or faces in the model. The function of measuring areas is used to measure and display the area of the specified area or face in the model. The function of measuring side lengths is used to measure and display the length of the specified side in the model. In addition, the measurement surface, measurement length, measurement volume, etc. can also be set based on the actual situation.

[0067] Building management is used to manage each building unit in the 3D home improvement model. Exemplarily, when the option box corresponding to the floor slab is checked, the floor slab is visible, and when the option box corresponding to the floor slab is not checked, the floor slab is invisible. In addition to the floor slab, the visibility of walls, windows, doors, furniture, etc. can also be set.

[0068] Construction information, used to display the detailed information of each construction. Such as the name, type (walls, floors, furniture, etc.), dimensions, materials, colors and other attribute information of the construction. In addition, it can also be the intersection position of two constructions. Exemplarily, such as the intersection of two walls.

[0069] The third display area is for displaying the aforementioned preset annotation types, and the annotator needs to perform annotations in this area.

[0070] S1003, in response to the annotation operation on any construction unit, obtain the annotation result of any construction unit.

[0071] During implementation, the annotator fills in based on each rule in the third display area. Exemplarily, the dimension range can be in the form of a digital input box or a slider.

[0072] S1004, based on the annotation result of any construction unit, update the to-be-annotated task in the target database.

[0073] During implementation, when the annotator finishes annotating the target construction in the third display area, then click the annotation completion control, and it can be classified into the annotated part in the first display area.

[0074] S1005, when it is determined that all construction units of the 3D home improvement model have been annotated, obtain the annotated data of the 3D home improvement model.

[0075] During implementation, when it is determined that all target constructions in the first display area have been annotated (that is, the to-be-annotated task is 0), the annotated data of the 3D home improvement model can be obtained, and then the annotated data of the 3D home improvement model can be stored in the source database.

[0076] In the embodiments of the present disclosure, after parsing the to-be-annotated task, the annotation status, 3D home improvement model and annotation task parameters of each construction unit are accurately pushed to the annotation interface. The annotator does not need to screen the required content from a large amount of information and can directly work on the unannotated construction units, saving the search time. The second display area supports preset operations on each construction unit, such as rotation, scaling, translation, etc., which is convenient for the annotator to observe and annotate the model from different angles, reducing the time waste caused by inconvenient operations and further improving the annotation efficiency. In addition, the annotation interface is divided into different display areas to display the annotation status, 3D home improvement model and annotation task parameters respectively. The overall layout is clear and intuitive, which is convenient for the annotator to quickly obtain the required information, reducing operation errors and unnecessary operation steps caused by interface chaos and improving the user experience.

[0077] In some embodiments, processing the labeled data based on the second target ETL pipeline and storing the labeled data in the source database includes: Step B1, in response to a third sub-operation of creating a target partition in the fourth visualization interface, creating a target partition in the source database for storing the labeled data.

[0078] During implementation, as Figure 12 shown, inputting the directory information where the target partition is located in the target partition can complete the operation of creating the partition.

[0079] Step B2, in response to a fourth sub-operation of configuring the field attributes of the source field for the target field in the fourth visualization interface, mapping the field attributes of the target field into the target partition to obtain at least one piece of labeled data suitable for storage in the source database.

[0080] Since there may be differences in the field names used between the target database and the source database, it is necessary to convert the fields so that they can be stored in the source database.

[0081] Step B3, merging at least one piece of labeled data and storing it in the target partition in the source database.

[0082] Exemplarily, the labeling tasks of each target in the foregoing 3D home improvement model are individual pieces of labeled data, and all the labeled data corresponding to the 3D home improvement model can be stored in one target partition.

[0083] In the embodiments of the present disclosure, by creating a target partition in the source database and mapping the field attributes, the refined organization and management of the labeled data are realized. Each piece of labeled data has a clear storage location and corresponding relationship, which is convenient for subsequent data query, statistics, and analysis.

[0084] In the embodiments of the present disclosure, a schematic diagram of a data processing method for training an artificial intelligence model is provided. As Figure 13 shown, it includes a source database 1301, a target database 1302, a terminal device 1303, a first target ETL pipeline 1304, and a second target ETL pipeline 1305. Among them, in the source database 1301, the data to be processed is processed into a labeling task to be sent to the data table in the target database 1302 for storage through the first target ETL pipeline 1304. The target database 1302 visually displays the labeling task to be labeled on the terminal device 1303 based on json technology. After the labeling personnel complete the labeling of the labeling task to be labeled, labeled data is obtained. After storing the labeled data in the target database 1302, it is stored in the source database 1301 through the second target ETL pipeline 1305. Among them, Figure 13The solid line part shown in the figure is the flow of the data to be processed in the source database, and the dashed line part is the flow of the labeled data stored back to the source database.

[0085] Based on the same technical concept, an embodiment of the present disclosure also proposes a data processing apparatus 1400 for training an artificial intelligence model, as Figure 14 shown, including: A first processing module 1401, configured to process a set of data to be processed in a source database based on a first target ETL pipeline supporting visual configuration, so as to split out a labeling task applicable to storage in a target database from the source database; the data in the source database is stored in the form of files; the target database stores data in the form of data tables; A labeling module 1402, configured to label the labeling task based on the target database to obtain labeled data; A second processing module 1403, configured to process the labeled data based on a second target ETL pipeline and store the labeled data in the source database.

[0086] In some embodiments, the first processing module includes: A first configuration unit, configured to obtain at least one source field in the set of data to be processed in response to a first configuration operation on the set of data to be processed in the source database in a first visual interface; A second configuration unit, configured to determine the field attributes of at least one target field corresponding to a specified source field in the target database in response to a second configuration operation on the specified source field in at least one source field in a second visual interface, and obtain a labeling task for each target field.

[0087] In some embodiments, the second configuration unit includes: A creation subunit, configured to create a target data table corresponding to the target table name in the target database in response to a first sub-operation of creating a target table name in the second visual interface; A mapping subunit, configured to map the field attributes of the specified source field to the target data table in response to a second sub-operation of configuring the field attributes of the target field for the specified source field in the second visual interface, and obtain a labeling task for each target field; wherein, the field attributes of the target field include: target field index, target field name, target field type, target field value, labeling type, and display type; The display type includes any one of the following: picture type, web page type, and three-dimensional home improvement model type.

[0088] In some embodiments, it further includes an analysis module for: Analyze the target field value corresponding to the target field to obtain a recommended list of display types adapted to the target field value; Display the recommended list corresponding to the target field in the second visualization interface.

[0089] In some embodiments, it further includes a range configuration module for: Respond to the third configuration operation of the annotator in the third visualization interface to determine the annotation task range of the annotator.

[0090] In some embodiments, the annotation module includes: A parsing unit for, when the display type of the task to be annotated is a three-dimensional home improvement model type, parsing the task to be annotated from the target database to obtain the three-dimensional home improvement model in the task to be annotated, each building unit in the three-dimensional home improvement model, and the annotation status of each building unit; An output unit for outputting the annotation status of each building unit, the three-dimensional home improvement model, and the annotation task parameters of the three-dimensional home improvement model to the annotation interface; wherein, the annotation status is annotated or unannotated; wherein, the annotation status of each building unit is used for display in the first display area of the annotation interface; the three-dimensional home improvement model is used for display in the second display area of the annotation interface; the annotation task parameters are used for display in the third display area of the annotation interface; the second display area supports preset operations on each building unit; An annotation unit for obtaining the annotation result of any building unit in response to the annotation operation on any building unit; An update unit for updating the task to be annotated in the target database based on the annotation result of any building unit; A determination unit for obtaining the annotated data of the three-dimensional home improvement model when it is determined that all building units of the three-dimensional home improvement model have been annotated.

[0091] In some embodiments, the second processing module includes: A creation unit for creating a target partition for storing annotated data in the source database in response to the third sub-operation of creating a target partition in the fourth visualization interface; A mapping unit for mapping the field attributes of the target field to the target partition in response to the fourth sub-operation of configuring the source field for the target field in the fourth visualization interface to obtain at least one piece of annotated data suitable for storage in the source database; Merge at least one piece of annotated data and store it in the target partition in the source database.

[0092] For the specific functions and examples of each module and sub-module of the device in the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the above method embodiments, which will not be elaborated here.

[0093] In the technical solution of the present disclosure, the acquisition, storage, and application of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0094] Figure 15 It is a structural block diagram of an electronic device according to an embodiment of the present disclosure. As Figure 15 shown, the electronic device includes: a memory 1510 and a processor 1520. The memory 1510 stores a computer program that can run on the processor 1520. The number of the memory 1510 and the processor 1520 can be one or more. The memory 1510 can store one or more computer programs. When the one or more computer programs are executed by the electronic device, the electronic device executes the method provided in the above method embodiment. The electronic device may further include: a communication interface 1530, configured to communicate with external devices and perform data interaction and transmission.

[0095] If the memory 1510, the processor 1520, and the communication interface 1530 are implemented independently, the memory 1510, the processor 1520, and the communication interface 1530 can be connected to each other through a bus and complete communication with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 15 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0096] Optionally, in a specific implementation, if the memory 1510, the processor 1520, and the communication interface 1530 are integrated on a chip, the memory 1510, the processor 1520, and the communication interface 1530 can complete communication with each other through an internal interface.

[0097] It should be understood that the above-mentioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. It is worth noting that the processor can be a processor that supports the Advanced RISC Machines (ARM) architecture.

[0098] Furthermore, optionally, the above-mentioned memory can include a read-only memory and a random access memory, and can also include a non-volatile random access memory. The memory can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can include a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory. The volatile memory can include a Random Access Memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Date SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct RAMBUS RAM (DR RAM).

[0099] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present disclosure are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or wirelessly (such as infrared, Bluetooth, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a Digital Versatile Disc (DVD)), or a semiconductor medium (such as a Solid State Disk (SSD)), etc. It should be noted that the computer-readable storage medium mentioned in the present disclosure can be a non-volatile storage medium, in other words, it can be a non-transitory storage medium.

[0100] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium, and the storage medium mentioned above can be a read-only memory, a magnetic disk, an optical disc, or the like.

[0101] In the description of the embodiments of the present disclosure, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0102] In the description of the embodiments of the present disclosure, unless otherwise specified, " / " means "or". For example, A / B may mean A or B. The "and / or" herein is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B may mean: A exists alone, A and B exist simultaneously, and B exists alone.

[0103] In the description of the embodiments of the present disclosure, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the embodiments of the present disclosure, unless otherwise specified, "a plurality of" means two or more.

[0104] The foregoing are only exemplary embodiments of the present disclosure and are not intended to limit the present disclosure. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A data processing method for training an artificial intelligence model, comprising: Processing a set of data to be processed in a source database based on a first target ETL pipeline supporting visual configuration, so as to split out annotation tasks applicable to storage in a target database from the source database; The source database stores data in the form of files; the target database stores data in the form of data tables; Based on the target database, annotating the annotation tasks to obtain annotated data; Processing the annotated data based on a second target ETL pipeline and storing the annotated data in the source database.

2. The method according to claim 1, wherein The processing of the set of data to be processed in the source database based on the first target ETL pipeline supporting visual configuration to split out annotation tasks applicable to storage in the target database from the source database includes: Responding to a first configuration operation on the set of data to be processed in the source database in a first visual interface to obtain at least one source field in the set of data to be processed; Responding to a second configuration operation on a specified source field among the at least one source field in a second visual interface to determine the field attributes of at least one target field corresponding to the specified source field in the target database, and obtaining annotation tasks for each target field.

3. The method according to claim 2, wherein The responding to the second configuration operation on the specified source field among the at least one source field in the second visual interface to determine the field attributes of at least one target field corresponding to the specified source field in the target database and obtaining annotation tasks for each target field includes: Responding to a first sub-operation of creating a target table name in the second visual interface to create a target data table corresponding to the target table name in the target database; Responding to a second sub-operation of configuring the field attributes of the target field for the specified source field in the second visual interface to map the field attributes of the specified source field into the target data table, and obtaining annotation tasks for each target field; Wherein, the field attributes of the target field include: target field index, target field name, target field type, target field value, annotation type, and display type; The display type includes any one of the following: picture type, web page type, 3D home decoration model type.

4. The method according to claim 3, further comprising: Analyzing the target field value corresponding to the target field to obtain a recommended list of display types adapted to the target field value; Displaying the recommended list corresponding to the target field in the second visual interface.

5. The method according to any one of claims 1-4, further comprising: Responding to a third configuration operation on an annotator in a third visual interface to determine the annotation task scope of the annotator.

6. The method according to claim 1, wherein The annotating the annotation tasks based on the target database to obtain annotated data includes: When the display type of the task to be labeled is a 3D home improvement model type, parse the task to be labeled from the target database to obtain the 3D home improvement model in the task to be labeled, each building unit in the 3D home improvement model, and the labeling status of each building unit; Output the labeling status of each building unit, the 3D home improvement model, and the labeling task parameters of the 3D home improvement model to the labeling interface; wherein, the labeling status is labeled or unlabeled; wherein, the labeling status of each building unit is used to be displayed in the first display area of the labeling interface; the 3D home improvement model is used to be displayed in the second display area of the labeling interface; the labeling task parameters are used to be displayed in the third display area of the labeling interface; the second display area supports preset operations on each building unit; In response to a labeling operation on any building unit, obtain the labeling result of the any building unit; Based on the labeling result of the any building unit, update the task to be labeled in the target database; When it is determined that all building units of the 3D home improvement model have been labeled, obtain the labeled data of the 3D home improvement model.

7. The method according to claim 3, wherein The processing of the labeled data based on the second target ETL pipeline and storing the labeled data in the source database includes: In response to a third sub-operation of creating a target partition in the fourth visualization interface, create a target partition in the source database for storing the labeled data; In response to a fourth sub-operation of configuring the field attributes of the source field for the target field in the fourth visualization interface, map the field attributes of the target field to the target partition to obtain at least one labeled data suitable for storing in the source database; Merge the at least one labeled data and store it in the target partition in the source database.

8. A data processing device for training an artificial intelligence model, comprising: A first processing module, configured to process a set of data to be processed in a source database based on a first target ETL pipeline that supports visual configuration, so as to split out a task to be labeled suitable for storage in a target database from the source database; The source database stores data in a file form; the target database stores data in a data table form; A labeling module, configured to label the task to be labeled based on the target database to obtain labeled data; A second processing module, configured to process the labeled data based on a second target ETL pipeline and store the labeled data in the source database.

9. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

11. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 - 7.

Images