Data processing method and device applied to artificial intelligence, equipment and medium

Abstract

The invention provides a data processing method and device applied to artificial intelligence, equipment and a medium, relates to the technical field of artificial intelligence, can be applied to scenes such as cloud technology and artificial intelligence, and comprises the following steps: in response to a data query request carrying data query information sent by a terminal, sending an associated data table corresponding to the data query information to the terminal; receiving a data request sent by the terminal for target pre-stored data in the associated data table, obtaining the target pre-stored data and sending the target pre-stored data to the terminal; receiving annotation data generated by the terminal for the target pre-stored data, and associatively storing the annotation data, the target pre-stored data and data information of the target pre-stored data; and in response to a training data request carrying the training data filtering information sent by the terminal, sending target data information matched with the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end. The data management efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular, to a data processing method, device, equipment, and medium applied to artificial intelligence. Background Art

[0002] With the development of artificial intelligence technology, intelligent models are widely used in various fields, such as for data classification, data generation, etc. Especially in recent years, intelligent models such as AIGC (Artificial Intelligence Generated Content) based on large models, high computing power, and multi-modalities have subverted the public's perception of artificial intelligence technically. When the amount of data reaches a certain level, AI shows emergent capabilities, that is, there is logical judgment ability and it can make judgments on things that have not been trained. Therefore, the management requirements of large models such as AIGC for huge training data sets are becoming more and more urgent. When the data reaches a certain scale, the storage and management complexity of training data will increase exponentially. Currently, the processing and storage of training data generally adopt manual classification management, which is prone to problems such as data storage fragmentation and classification errors, resulting in low data management efficiency and high costs. Summary of the Invention

[0003] This application provides a data processing method, device, equipment, and medium applied to artificial intelligence, which can significantly improve the efficiency and convenience of data processing applied to artificial intelligence.

[0004] On the one hand, this application provides a data processing method applied to artificial intelligence, and the method includes:

[0005] In response to a data query request carrying data query information sent by a terminal, send an associated data table corresponding to the data query information to the terminal, so that the terminal displays the associated data table;

[0006] Receive a data request for target pre-stored data in the associated data table sent by the terminal, obtain the target pre-stored data and send it to the terminal, so that the terminal displays the target pre-stored data;

[0007] Receive the annotation data generated by the terminal for the target pre-stored data, and store the annotation data, the target pre-stored data, and the data information of the target pre-stored data in an associated manner. The data information includes model screening information, and the model screening information is used to indicate a neural network model related to the pre-stored data;

[0008] In response to a training data request carrying training data filtering information sent by the terminal, send target data information that matches the model screening information and the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end, and the target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

[0009] On the other hand, a data processing device applied to artificial intelligence is provided. The device includes:

[0010] A data list module: configured to respond to a data query request carrying data query information sent by the terminal, and send an associated data table corresponding to the data query information to the terminal, so that the terminal displays the associated data table;

[0011] A data acquisition module: configured to receive a data request for target pre-stored data in the associated data table sent by the terminal, acquire the target pre-stored data and send it to the terminal, so that the terminal displays the target pre-stored data;

[0012] A data storage module: configured to receive annotation data generated by the terminal for the target pre-stored data, and store the annotation data, the target pre-stored data, and the data information of the target pre-stored data in an associated manner. The data information includes model screening information, and the model screening information is used to indicate a neural network model related to the pre-stored data;

[0013] A training data screening module: configured to respond to a training data request carrying training data filtering information sent by the terminal, and send target data information that matches the model screening information and the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end, and the target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

[0014] On the other hand, a computer device is provided. The device includes a processor and a memory. At least one instruction or at least one segment of program is stored in the memory, and the at least one instruction or the at least one segment of program is loaded and executed by the processor to implement the data processing method applied to artificial intelligence as described above.

[0015] On the other hand, a computer-readable storage medium is provided. At least one instruction or at least one segment of program is stored in the storage medium, and the at least one instruction or the at least one segment of program is loaded and executed by a processor to implement the data processing method applied to artificial intelligence as described above.

[0016] On the other hand, a server is provided, which includes a processor and a memory. At least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement the data processing method applied to artificial intelligence as described above.

[0017] On the other hand, a terminal is provided, which includes a processor and a memory. At least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement the data processing method applied to artificial intelligence as described above.

[0018] On the other hand, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and when the computer instructions are executed by a processor, the data processing method applied to artificial intelligence as described above is implemented.

[0019] The data processing method, device, equipment, storage medium, server, terminal, computer program and computer program product applied to artificial intelligence provided by this application have the following technical effects:

[0020] The technical solution of this application can respond to a data query request carrying data query information sent by a terminal, automatically generate a corresponding associated data table, so that the terminal can display and submit a corresponding data request, and then send corresponding pre-stored data, so as to facilitate relevant personnel of the terminal to perform operations such as data annotation on the pre-stored data. After receiving the annotated data, it is associated and stored with the corresponding pre-stored data and the data information of the pre-stored data. While storing it as training data, when receiving a training data request, it can match the corresponding data information based on the training data filtering information and send it to the terminal, and seamlessly connect to the model training end, so that the model training end can automatically obtain the corresponding training data, realizing the centralization of the background management of training data. It can be applied to the automatic storage, processing and distribution of multi-model training data, improving the storage, processing and sorting efficiency, classification accuracy of pre-training data, as well as the distribution efficiency and accuracy of training data. It can be applied to the business lines of multiple models to optimize the business process processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1It is a schematic diagram of an application environment provided by an embodiment of the present application;

[0023] Figure 2 It is a schematic flowchart of a data processing method applied to artificial intelligence provided by an embodiment of the present application;

[0024] Figure 3 It is a schematic framework diagram of a training data management platform provided by an embodiment of the present application;

[0025] Figures 4 - 6 It is a schematic interface diagram of a client of a training data management platform provided by an embodiment of the present application;

[0026] Figure 7 It is a schematic diagram of the principle of another data encoding provided by an embodiment of the present application;

[0027] Figure 8 It is a schematic flowchart of another data processing method applied to artificial intelligence provided by an embodiment of the present application;

[0028] Figure 9 It is a schematic flowchart of another data processing method applied to artificial intelligence provided by an embodiment of the present application;

[0029] Figure 10 It is a schematic flowchart of another data processing method applied to artificial intelligence provided by an embodiment of the present application;

[0030] Figure 11 It is a schematic framework diagram of a data processing device applied to artificial intelligence provided by an embodiment of the present application;

[0031] Figure 12 It is a hardware structure block diagram of an electronic device that executes a data processing method applied to artificial intelligence provided by an embodiment of the present application. Detailed implementation manners

[0032] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0033] It should be noted that in the description and claims of this application and the above-mentioned drawings, terms such as "first" and "second" are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or sub-modules does not necessarily have to be limited to those steps or sub-modules clearly listed, but may include other steps or sub-modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0034] In the embodiments of this application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the function of that module or unit.

[0035] It should be noted that in the specific implementation of this application, for related data such as login requests, data query requests, data requests, training data requests, target databases, etc., when the embodiments in this application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards of relevant countries and regions.

[0036] Before further elaborating on the embodiments of this application, the nouns and terms involved in the embodiments of this application are described. The nouns and terms involved in the embodiments of this application are subject to the following explanations.

[0037] Artificial Intelligence (AI) is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science. It attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling machines to have the functions of perception, reasoning, and decision-making.

[0038] Artificial intelligence technology is an interdisciplinary subject that involves a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-training model technology, operation / interaction systems, and mechatronics. Among them, the pre-training model, also known as the large model or the foundation model, can be widely applied to downstream tasks in various directions of artificial intelligence after fine-tuning. Artificial intelligence software technologies mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0039] The pre-training model (PTM), also known as the foundation model or the large model, refers to a deep neural network (DNN) with a large number of parameters. It is trained on a large amount of unlabeled data, and uses the function approximation ability of the large-parameter DNN to extract common features from the data. Through techniques such as fine-tuning, parameter-efficient fine-tuning (PEFT), and prompt-tuning, it is suitable for downstream tasks. Therefore, the pre-training model can achieve ideal results in few-shot or zero-shot scenarios. PTM can be classified into language models, vision models, speech models, multi-modal models, etc. according to the data modalities it processes. Among them, the multi-modal model refers to a model that establishes feature representations of two or more data modalities. The pre-training model is an important tool for outputting artificial intelligence-generated content (AIGC), and can also be used as a general interface connecting multiple specific task models.

[0040] With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless, autonomous driving, drones, digital twins, virtual humans, robots, artificial intelligence-generated content (AIGC), conversational interaction, intelligent healthcare, intelligent customer service, game AI, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0041] Please refer to Figure 1 , Figure 1 which is a schematic diagram of an application environment provided by an embodiment of this application. As Figure 1 shown, this application environment may at least include a terminal 01, a data server 02, and a model training end 03. In practical applications, the terminal 01, the data server 02, and the model training end 03 can be directly or indirectly connected through wired or wireless communication methods, and this application does not make any restrictions here.

[0042] In the embodiments of the present application, the data server 02 and the model training server 03 can run on a server. The server can be an independent physical server, or a server cluster or a 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.

[0043] Specifically, cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or a local area network to achieve data computing, storage, processing, and sharing. Cloud technology can be applied to various fields, such as medical cloud, cloud IoT, cloud security, cloud education, cloud conferencing, artificial intelligence cloud services, cloud applications, cloud calls, and cloud social networking. Cloud technology is based on the cloud computing business model. It distributes computing tasks on a resource pool composed of a large number of computing devices, enabling various application systems to obtain computing power, storage space, and information services as needed. The network that provides resources is called the "cloud". The resources in the "cloud" seem to be infinitely expandable to users, and can be obtained at any time, used on demand, expanded at any time, and paid according to usage. As a basic capability provider of cloud computing, a cloud computing resource pool (referred to as a cloud platform, generally called IaaS (Infrastructure as a Service)) platform will be established, and various types of virtual resources will be deployed in the resource pool for external customers to choose and use. The cloud computing resource pool mainly includes: computing devices (virtual machines, including operating systems), storage devices, and network devices.

[0044] According to the logical function division, the PaaS (Platform as a Service) layer can be deployed on the IaaS layer, and the SaaS (Software as a Service) layer can be deployed on top of the PaaS layer. The SaaS can also be directly deployed on the IaaS. PaaS is a platform for software operation, such as databases, web containers, etc. SaaS is various business software, such as web portals, SMS mass senders, etc. Generally speaking, SaaS and PaaS are upper layers relative to IaaS.

[0045] Specifically, the above-mentioned server can include physical devices, which can specifically include a network communication sub-module, a processor, a memory, etc., or can also include software running on the physical devices, which can specifically include application programs, etc.

[0046] Specifically, the terminal 01 may include physical devices such as smart phones, desktop computers, tablet computers, laptop computers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, intelligent voice interaction devices, smart home appliances, smart wearable devices, in-vehicle terminal devices, etc., or may also include software running on the physical devices, such as application programs, etc.

[0047] In the embodiments of the present application, the data server 02 may run a training data management system, the model training end 03 is used to perform the training of one or more neural network models based on the training data, and the terminal 01 can run the client corresponding to the training data management system to be able to display a front-end page on the terminal and perform data display, data screening interface display, data annotation interface display, etc. on the front-end page, so that the terminal generates a data query request, a data request, annotated data, or a training data request carrying training data filtering information, etc. carrying data query information based on user operations and sends them to the data server 02; the data server 02 can provide corresponding data acquisition and forwarding services through the training data management system to feedback the corresponding information and data to the terminal 01, so that the terminal 01 performs result display on the front-end page, and sends target data information matching the training data filtering information to the model training end 03; the model training end 03 can obtain corresponding pre-stored data and its associated annotated data from the data server 02 based on the target data information to use as training data to train the corresponding neural network model. The training data management system and the client form a training data management platform.

[0048] In addition, it can be understood that Figure 1 What is shown is only an application environment of a data processing method applied to artificial intelligence. This application environment may include more or fewer nodes, and the present application does not limit this here.

[0049] The application environment involved in the embodiments of the present application, or the terminal 01 and the server, etc. in the application environment may be a distributed system formed by connecting a client and multiple nodes (any form of computing device accessing the network, such as a server, a terminal) in the form of network communication. The distributed system may be a blockchain system, and the blockchain system may provide the above-mentioned data processing services, model training services, data storage services, etc. applied to artificial intelligence.

[0050] The data processing method for artificial intelligence applied in this application can be applied to the above-mentioned training data management platform, which can be used to manage multi-modal training data, and specifically can be used to manage multi-modal training data for 3D AIGC model training. In one embodiment, the 3D AIGC model is a virtual object generation model, used for generating virtual characters, props or virtual environments in game services or virtual reality services, etc. The multi-modal training data can include but is not limited to three-dimensional modal data, text, and point cloud data, etc. The three-dimensional modal data can be 3D model files, etc.

[0051] In the training data management platform, the training data management system includes a backend service and a data service. A target database runs in the data service. The data service can be used for but is not limited to data upload, database update, data deletion, data download, data format conversion, etc. In one embodiment, referring to Figure 3 , the client (front end) can be built using a JavaScript framework to process information and display data, including data display, data filtering, and data annotation, etc. For example, the responsive Vue.js framework is used for data display such as text, charts, pictures, and 3D data on the front-end page. The 3D data display can be rendered in real time and can be based on the Three.js framework for data rendering. The backend can use a web framework, such as the python and Django frameworks, etc., where python is beneficial for the linkage with neural network models. The database model of the target database can be built based on MySQL (My Structured Query Language) or MongoDB (Humongous Databas), etc., to store data information and associated pre-stored data; the front-end of multiple users sends requests to the backend service or receives feedback information to obtain associated data tables and target data information, etc., from the backend service. Then the front-end obtains pre-stored data, etc., through the data service based on the foregoing feedback information to implement real-time update functions such as display, annotation, and modification of pre-stored data. The front-end and the backend service can communicate through Socket. The front-end is also used to send target data information to the training server, so that the training server can obtain corresponding pre-stored data and annotation data from the data service according to the target data information for training relevant neural network models as training data. The front-end is also used to send pre-stored data, etc., to the backend to import data through the backend service to store new data in the target database of the data service.

[0052] Exemplarily, the training data management platform is used to process training data applied to the game AIGC model, and the database model file can be but is not limited to the database model defined in Table 1 below.

[0053] Table 1

[0054]

[0055]

[0056] The above table is stored in binary format within the training data management system and contains data files (such as files in the formats of.frm,.ibd,.myd, etc.) and log files (such as files in the formats of.log,.binlog, etc.). Among them, the model data name, model file path, model style, data source, data web address, model game (the name of the game to which it belongs), the number of model patches, whether there is texture, whether manual annotation is required, screening time, annotation time, data status, whether it has been annotated, whether it has been screened, etc. belong to basic information. Among them, information such as screening time, annotation time, data status, whether it has been annotated, whether it has been screened, etc. update the information values based on operations such as annotation, modification, and screening of the pre-stored data; the model file path, the model nine-square grid (thumbnail) path, the model rendering result path, the model text path, the model point cloud path, the model encoding path, etc. belong to storage address information.

[0057] Exemplarily, the front end can use the Vue framework to build the target page, adopting the development mode of virtual DOM and componentization, making it simple and efficient to build complex application programs. Refer to Figure 4 , Figure 4 shows a paging interface of a front-end page built using the Vue framework in this application, which can include but is not limited to interfaces such as the dashboard, data, annotation, and 3D display. Among them, the dashboard interface is used to display data statistical information, which is obtained by conditionally screening and counting the data in the target database according to the combination of filtering condition information. Different combinations of filtering condition information can be switched to different data statistical information. The filtering condition information can include but is not limited to being annotated, screened, model style, model game, data source, screening time, annotation time, the number of model patches, having texture, data status, etc., or it can also be obtained by performing data retrieval based on the uploaded data to be queried. As Figure 4 shown, based on the filtering conditions, the data category statistics of each pre-stored data in the current target database and the trainable data statistics of the currently trainable data that meet the filtering conditions are displayed. Further, the data interface is mainly used to display the specific information of the currently filtered or retrieved data, which is specifically consistent with the information of the database model. The annotation interface is mainly used for operations such as screening and modifying each data one by one (refer to Figure 5 ). The 3D interface is mainly for 3D display or annotation, which can be but is not limited to Figure 6As shown, a three-dimensional interface can be used to display three-dimensional data (such as 3D models) after real-time rendering, and can also display data information, operation options, etc., to receive data annotation and modification, etc. In addition, the training data management platform is also used to provide login authentication based on a secure interface and a corresponding login authentication interface, facilitating different users to perform various data processing based on Web login.

[0058] The above-mentioned interfaces can be built through the component system in the Vue framework, which includes interfaces such as directives, watchers, transitions and animations, event handling, routing management, and state management. Different interfaces can be used for encapsulation and layout construction based on the requirements of each interface.

[0059] Exemplarily, the backend service can be built using Python and Django REST Framework. Django REST Framework is a framework for building Web APIs, built on top of Django, providing a variety of tools and functions that make building and managing RESTful APIs simple and efficient. It provides a variety of functions, including serialization, authentication, permission control, view sets, routing, etc. These functions enable developers to build APIs that conform to RESTful design principles and provide flexible data serialization and deserialization. The interfaces of the backend service mainly include login, registration, obtaining user information, paging to obtain a list, obtaining default filtering values, json files, querying according to filtering conditions, obtaining the data address sent by the data service, and updating annotations, etc. The above interfaces can all be encapsulated through the Django framework. Among them, login, registration, and obtaining user information are secure login interfaces. The backend service can, in response to a login request from the front end, trigger operations to call interfaces such as login, registration, and obtaining user information based on a secure login event. The information in the user table can include, but is not limited to, username, number, password, email, avatar, etc. It should be noted that any user-related information in this application is obtained under the premise of user authorization and permission, for data communication and processing under the authorized permissions.

[0060] Based on the above training data management platform with a complete front-end and back-end architecture, it can solve a series of processing of large-scale multi-modal training data, such as screening, annotation, statistics, filtering, 3D web display, and information storage. The front-end mainly uses the Vue framework for cross-platform communication and processing, and the back-end service uses Python and the Django framework. This combination can be combined with a neural network model based on the Python language to maximize the use of the capabilities of existing neural network models. The database can use MySQL, etc., to comprehensively solve the information storage, web access, data screening, annotation, and other post-processing requirements of multi-modal data, and meet the various types of data classification and processing requirements needed during the training process of the 3D AIGC model.

[0061] The following introduces the technical solution of this application based on the above application environment and training data management platform. The embodiments of this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, intelligent transportation, assisted driving, etc. Please refer to Figure 2 , Figure 2 FIG. is a schematic flowchart of a data processing method applied to artificial intelligence provided by an embodiment of this application. This specification provides method operation steps such as embodiments or flowcharts, but based on routine or non-creative labor, it may include more or fewer operation steps. The step order listed in the embodiments is only one of the execution orders of numerous steps and does not represent the only execution order. When the actual system or server product executes, it can be executed in the method order shown in the embodiments or the drawings or executed in parallel (such as in an environment of parallel processors or multi-threaded processing). Specifically, as Figure 2 shown, the method may include the following steps S201-S207:

[0062] S201: In response to a data query request carrying data query information sent by a terminal, send an associated data table corresponding to the data query information to the terminal, so that the terminal can display the associated data table.

[0063] Specifically, the training data management platform displays the target page of the client on the terminal to receive operations on relevant options for data query on the target page (such as "query data upload" or as Figure 4 the "filter condition" in), and then generates a data query request carrying data query information based on the operation data, and the terminal sends the data query request to the back-end service, so that the back-end service can obtain the associated data table corresponding to the data query information through the data service in response to the data query request.

[0064] Specifically, the data query information is used to match with the data information of the pre-stored data in the target database to obtain the associated data information, and then the backend service generates an associated data table and sends it to the terminal. The data information is used to represent the information of the pre-stored data. As exemplified in Table 1 above, multiple data information can be stored in a database model file. After receiving the associated data table, the terminal displays it on the interface. For example, it can be displayed in the "dashboard" interface through a statistical chart, and clicking on the statistical chart can display the associated data table in tabular form.

[0065] S203: Receive a data request for the target pre-stored data in the associated data table sent by the terminal, obtain the target pre-stored data, and send it to the terminal so that the terminal can display the target pre-stored data.

[0066] Specifically, the target pre-stored data refers to at least one pre-stored data corresponding to the selected data information in the associated data table. The terminal can receive a selection operation for each data information in the associated data table based on the dashboard interface or the data interface, so as to generate a data request carrying the data selection information and send it to the backend service. In response to the data request, the backend service obtains the corresponding target pre-stored data from the target database through the data service based on the data information corresponding to the data selection information, and feeds it back to the terminal. Alternatively, the data request is sent to the data service, and the data service sends the target pre-stored data to the terminal in response to the data request.

[0067] Specifically, after receiving the target pre-stored data, the terminal can display it. In some embodiments, the pre-stored data is multi-modal data, including at least one of text, pictures, point cloud information, and three-dimensional modal data. The client can obtain the corresponding data through the storage address information in the data information (such as the model file path, model text path, model point cloud path, etc. in Table 1). For three-dimensional modal data, it can be rendered in real time and displayed on the "3D" interface.

[0068] S205: Receive the annotation data generated by the terminal for the target pre-stored data, and store the annotation data, the target pre-stored data, and the data information of the target pre-stored data in an associated manner.

[0069] Specifically, based on the displayed target pre-stored data, the user submits annotation operations such as acceptance, rejection, repair, or data annotation for pictures or three-dimensional modal data on the interface, so that the client generates annotation data based on the annotation operations and sends it to the backend service through the corresponding interface (such as the Vue interface). The backend service updates the above annotation data to the target database through the data interface service to store it in an associated manner with the target pre-stored data and its data information, and generates the storage address information of the annotation data and updates it to the data information.

[0070] S207: In response to a training data request carrying training data filtering information sent by a terminal, send target data information that matches the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end. The target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

[0071] Specifically, the training data filtering information can be set based on the specific requirements of the neural network model. For example, it can include information items such as requiring manual annotation, being annotated, model style, having texture, model game, etc. The client generates a training data request carrying the training data filtering information based on the received information items and sends it to the back-end service. The back-end service matches the training data filtering information with the data information of each pre-stored data in the target database, determines the target data information, and feeds it back to the client. The client sends it to the model training end. The model training end directly obtains the pre-stored data and annotation data corresponding to the target data information stored in the target database from the data service according to the target data information for model training. In some embodiments, the client sends the target data information to the model training end in a first data format, such as the json format. The model training end can directly obtain the training data from the data service according to the information in the json format.

[0072] In some embodiments, the data information can further include model screening information. The model screening information is used to indicate the neural network model related to the pre-stored data, that is, to indicate the neural network model that the pre-stored data can train. Correspondingly, the training data filtering information can include the network model information of the pointed neural network model. Then, after receiving the training data request, the back-end service directly matches the network model information with the model screening information of each data information, determines the data information to which the matched model screening information belongs as the target data information, and forwards it to the model training end through the terminal to improve the training data classification and acquisition efficiency, and at the same time improve the accuracy of data classification management. It can be understood that the same data information can correspond to multiple neural network models, that is, the model screening information can point to multiple neural network models. The neural network model can be an AIGC model, such as an AIGC large model applied to virtual object generation.

[0073] By adopting the above technical solution, it is possible to respond to a data query request carrying data query information sent by a terminal, automatically generate a corresponding associated data table, enable the terminal to display and submit a corresponding data request, and then send corresponding pre-stored data, so as to facilitate operations such as data annotation by relevant personnel of the terminal. After receiving the annotation data, it is associated and stored with the corresponding pre-stored data and the data information of the pre-stored data. While storing it as training data, when receiving a training data request, it can match the corresponding data information based on the training data filtering information and send it to the terminal, seamlessly connecting to the model training end, enabling the model training end to automatically obtain the corresponding training data, realizing the centralization of the background management of training data, and being applicable to the automatic storage, processing, and distribution of multi-model training data, improving the storage, processing, and sorting efficiency, classification accuracy of pre-training data, as well as the distribution efficiency and accuracy of training data, being applicable to the business lines of multiple models, and optimizing the business process processing.

[0074] In the traditional game development business line, the creation of virtual objects such as game characters requires a complete set of pipeline design, such as including mesh modeling, texturing, bone binding, and model driving, etc., which involves a large amount of manual operations. Based on the training data management system and data processing method of the present application, it is possible to centrally manage the multi-modal training data required by different models, directly train multiple AIGC models (such as 3D AIGC models), replace the traditional character model creation process, greatly improve the generation efficiency of virtual objects, and then realize multi-modal model design, so as to create richer and more flexible scenarios, characters, and props, etc., making the gameplay of the game more diverse and interesting.

[0075] It can be understood that the terminal can import or export data of the training data management system in the background through the client. Correspondingly, in some embodiments, the method may further include the data import process shown in S301 - S307:

[0076] S301: Receive the pre-stored data sent by the terminal, where the pre-stored data includes three-dimensional modal data, rendering data of the three-dimensional modal data, and additional data of the three-dimensional modal data.

[0077] Specifically, the three-dimensional modal data can be, for example, a 3D model file. Taking the game service as an example, the three-dimensional modal data can be the 3D model file of virtual objects (such as characters, tools, or scenes) in the game. The rendering data is obtained by rendering the three-dimensional modal data, and can include but is not limited to multiple model pictures obtained by rendering (such as a nine-grid picture), the 2D picture of the front of the model, the model rendering result, etc.; the additional data includes at least one other modal data for describing the three-dimensional modal data, including but not limited to data text (such as model text) and point cloud data (such as model point cloud), etc. Among them, the data text can be multiple text descriptions for the multi-modal data, which can be obtained by text extraction from the two-dimensional conversion picture of the three-dimensional modal data by a text generation from image model, or can also be an artificial description text.

[0078] S303: Perform data encoding on the three-dimensional modal data and the additional data to obtain a pre-stored encoding.

[0079] Specifically, the pre-stored encoding is obtained by encoding and converting the data features obtained after feature extraction of the three-dimensional modal data or the additional data. For data of different modalities, different encoding models can be used for feature encoding. Correspondingly, in some embodiments, the generation methods of the pre-stored encoding corresponding to the three-dimensional modal data include S3031 - S3033:

[0080] S3031: Render the three-dimensional modal data based on multiple preset perspectives to obtain perspective images corresponding to each of the multiple preset perspectives;

[0081] S3032: Input each perspective image into a data encoding model to obtain the image features of each perspective image;

[0082] S3033: Perform encoding conversion on each image feature based on the encoding algorithm corresponding to a preset encoding library to obtain the pre-stored encoding corresponding to the three-dimensional modal data.

[0083] Specifically, the preset encoding library is used to store the pre-stored encodings of each pre-stored data in the target database. The three-dimensional modal data can be rendered from multiple preset perspectives based on the backend service to obtain multiple two-dimensional perspective images, or multiple perspective images obtained by rendering the three-dimensional modal data by the client can be directly received. Refer to Figure 7, the backend service invokes the data encoding model to extract image features, so as to obtain multiple image features, and the multiple image features can be feature vectors of a fixed length. The data encoding model can be, but is not limited to, a pre-trained large model of the language-image type, which can be obtained by training with a large amount of text-image data and can perform zero-shot transfer using the provided text description corresponding to the concepts on the graph. Then, the encoding algorithms corresponding to the preset encoding library are respectively invoked to perform encoding conversion on each image feature to obtain the pre-stored encoding of each perspective image (such as Figure 7 the I in i , i ∈ [1, N], and N is the number of perspective images), so as to serve as the pre-stored encoding corresponding to the three-dimensional modal data. Exemplarily, the preset encoding library can be the faiss library, and then its encoding algorithm is used to convert the image features into index encodings for storage on the server. In this way, the data encoding alignment of the three-dimensional modal data into two-dimensional images is performed to facilitate subsequent picture retrieval, improve the retrieval efficiency and accuracy, and further optimize the training data screening efficiency.

[0084] In some embodiments, the additional data includes at least one of the data text corresponding to the three-dimensional modal data and the point cloud data; the generation method of the pre-stored encoding corresponding to the additional data includes S3034 - S3036:

[0085] S3034: Input the data text into the data encoding model for text feature encoding to obtain text features;

[0086] S3035: Input the point cloud data into the point cloud encoding model for feature encoding of aligning the point cloud features with the image text features to obtain point cloud features;

[0087] S3036: Based on the encoding algorithms corresponding to the preset encoding library, perform encoding conversion on the text features and the point cloud features to obtain the pre-stored encoding corresponding to the text features and the pre-stored encoding corresponding to the point cloud features.

[0088] Specifically, the data encoding model includes a text encoding network and a visual encoding network. The visual encoding network can be used to encode the above-mentioned perspective images, and the text encoding network can encode the input data text, and the output text features and image features are aligned. The backend service calls the data encoding model to extract features from two-dimensional image data and data text, realizes the alignment of text and image features, and further facilitates text and image multi-modal matching based on image search, as well as text and image multi-modal matching based on text search, so as to improve the accuracy and comprehensiveness of data screening. For point cloud data, a point cloud encoding model is used for feature encoding. The point cloud features output by the point cloud encoding model can be aligned with the text features and image features output by the data encoding model in terms of feature dimension and semantics, further realizing the matching accuracy and comprehensiveness of multi-modal search and optimizing the data screening results. Exemplarily, the point cloud encoding model can be constructed based on existing 3D point cloud feature encoding algorithms, and the present application does not make specific limitations.

[0089] S305: Store the three-dimensional modal data, rendering data, additional data, and pre-stored encoding in the target database, and feedback the storage address information of the three-dimensional modal data, rendering data, additional data, and pre-stored encoding to the terminal, so that the terminal generates initial data information in the first data format based on the basic information and storage address information of the pre-stored data.

[0090] Specifically, the backend service uploads relatively large-capacity data such as three-dimensional modal data, rendering data (such as model nine-grid pictures, model front 2D pictures, model rendering results, etc.), additional data (such as model text or model point cloud), and pre-stored encoding to the data storage end of the target database to obtain the storage address information of each data (such as the aforementioned model file path, model nine-grid (thumbnail) path, model rendering result path, model text path, model point cloud path, model encoding path, etc.), and sends it to the terminal, so that the client of the terminal combines the storage address information with the basic information to generate the data information corresponding to the pre-stored data, and the data information is generated in the form of a database model. Exemplarily, the basic information may include but is not limited to model data name, model style, model source, data web address, model game, number of patches, whether there is texture, etc. The first data format can be the json format. Correspondingly, the terminal encapsulates the data information of the pre-stored data into a json file and sends it to the backend service.

[0091] S307: Receive the initial data information sent by the terminal, and convert the initial data information into data information in the second data format and store it in the target database.

[0092] Specifically, the second data format is the data format adopted by the target database, such as a MySQL database file, etc.; the backend service converts the format of the initial data information and then realizes the associated storage of the data information and the pre-stored data. If there is historical data information corresponding to the data information in the target database, the current data information is merged with the historical data information. If not, a new storage entry is added. Exemplarily, the terminal transmits the generated json file from the front end to the backend service through the Vue interface, obtains a MySQL database file through the Django conversion interface, and this database file will be incorporated according to the existing historical database file. If there is historical data with the same name, the data is updated. If not, a new entry is added to realize the distributed storage of multi-modal training data. In this way, during the data import process, data encoding, classified storage, and information association of each item of data are realized to synchronously generate search information for the training data, and then the distributed storage and centralized management of the training data are realized.

[0093] In some embodiments, the front end can receive the data to be queried or filtering condition information to generate a data query request carrying data query information, and then transmit it to the backend service through a corresponding interface (such as a Vue web interface) for data filtering or retrieval to obtain an associated data table, and send it to the front end in the first data format through a corresponding interface (such as a Django interface) so that the user can download the corresponding associated data table file to realize data export.

[0094] It can be understood that different data query information is generated based on different relevant options of the data query. In some cases, the data query information includes the above-mentioned data to be queried, and the data to be queried can be images, texts, or point cloud information, etc. submitted for a target page, for retrieving similar data based on the image, text, or point cloud information; the data information includes the pre-stored coding information of the pre-stored data, and the pre-stored coding corresponding to the pre-stored coding information is obtained by encoding the data based on the coding model corresponding to the pre-stored data; correspondingly, refer to Figure 8 , the generation method of the associated data table includes S401 - S405:

[0095] S401: Encode the data to be queried to obtain a query code;

[0096] S403: Perform a similarity-based coding search in a preset coding library according to the query code to obtain a pre-stored coding similar to the query code, and the preset coding library is used to store the pre-stored codings of multiple pre-stored data;

[0097] S405: Generate an associated data table based on the data information corresponding to the similar pre-stored codings.

[0098] Specifically, the data to be queried may include query images, query texts, or query point cloud data. Based on the query encoding of the data to be queried, a preset encoding library is retrieved to obtain pre-stored encodings whose similarity meets the preset similarity conditions, and an associated data table is generated based on the data information to which the pre-stored encoding information of the obtained pre-stored encodings belongs. The associated data table may include at least part of the retrieved data information. In this way, query data retrieval is realized based on the construction of data encoding, and efficient screening of similar training data is achieved.

[0099] In some embodiments, S401 may include S4011 - S4012:

[0100] S4011: Input the data to be queried into the encoding model corresponding to the data modality of the data to be queried for feature encoding to obtain query data features;

[0101] S4012: Based on the encoding algorithm corresponding to the preset encoding library, perform encoding conversion on the query data features to obtain query encodings.

[0102] Similar to the foregoing S303, the data encoding model can be called to perform feature encoding on the query image and query text, and the point cloud encoding model can be called to perform feature encoding on the query point cloud to obtain the corresponding query data features, and then convert them into query encodings matching the preset encoding library, so as to facilitate pre-stored encoding retrieval and improve retrieval accuracy, comprehensiveness, and retrieval efficiency.

[0103] Exemplarily, the client sends query images, query texts, or query point clouds to the backend service through the web interface service; the backend service calls the corresponding encoding model to perform feature encoding on the query images, query texts, or query point clouds, and then converts the output result into a query encoding corresponding to the faiss library; based on the obtained query encoding, the preset encoding library is retrieved, and a fast similarity calculation is performed on the pre-stored encoding and the query encoding, such as performing a faiss nearest neighbor search on the index query encoding and the pre-stored encoding in the faiss library to obtain data information with high similarity, and then generating an associated data table.

[0104] In some other cases, the data query information includes at least one filtering condition information; the at least one filtering condition information may be generated based on the filtering option operation submitted for the target page, and the filtering condition information may include labeled, model style, model game, data source, presence of texture, etc. Correspondingly, referring to Figure 9 , the generation method of the associated data table includes S501 - S503:

[0105] S501: Query the data information that matches the filtering condition information in the target database, where the target database is used to associatively store the pre-stored data and the data information of the pre-stored data;

[0106] S503: Generate an associated data table based on the data information matched with the filtering condition information.

[0107] Specifically, match at least one piece of filtering condition information with the data information of each pre-stored data to screen out the data information whose information values of each item are consistent with each piece of filtering condition information, and then generate an associated data table and send it to the terminal for display. The terminal can display each data information in the associated data table, or implement statistical data display in different dimensions based on the associated data table. In this way, the data information is matched through the filtering conditions to facilitate the accurate and efficient filtering of the training data.

[0108] It can be understood that the back-end service can concurrently connect to multiple terminals to receive multiple data query requests sent concurrently by multiple terminals, so as to meet the operation requirements such as data query, data annotation, and training data acquisition on multiple terminals. Correspondingly, in the case of receiving data query requests sent concurrently by multiple terminals, refer to Figure 10 , the associated data table sent to the terminal in S201 includes S601 - S603:

[0109] S601: For each data query request, determine the initial associated data set in the target database based on the data query information;

[0110] S603: Screen out some data information from the initial associated data set to generate an associated data table corresponding to the current data query request, and the data information in the associated data tables corresponding to each data query request is mutually exclusive.

[0111] Specifically, the initial associated data set includes multiple data information that matches the data query information. There may be the same data information among the initial associated data sets corresponding to multiple concurrently sent data query requests. Therefore, after generating the associated data table for the current data query request, mark the data information therein to lock the data information, such as modifying the information value of the "whether screened" item to yes, so as to avoid adding it to the associated data table of another data query request, realizing the mutual exclusion of the associated data tables corresponding to each data query request, so that different terminals obtain associated data tables containing different data information. Each terminal directly obtains the specific pre-stored data through the data service based on the storage address information of each data information in the associated data table, and then displays it on the target page to facilitate operations such as receiving data annotation and revision, avoiding repeated screening and processing of the pre-stored data, and improving the system robustness and data processing efficiency.

[0112] In summary, by adopting the data processing method of the present application, it is possible to simply and efficiently implement steps such as statistical display, retrieval, screening, annotation, modification, and rendering of multi-modal training data. For example, the dashboard interface for data statistics can display the current data distribution according to the submitted filtering condition information, or only by submitting a query text, query image, or query point cloud, it is possible to retrieve the required pre-stored data, and it is also possible to export the data list; in addition, based on the construction of the training data management system, it is also possible to seamlessly connect model training with data screening, realizing efficient training data collection and training distribution; when performing multi-user parallel data queries, through the mutex table setting, the data screening of each user is independent of each other, avoiding data duplication; greatly improving the data processing and sorting efficiency before multi-modal data model training and optimizing the overall business process. Taking the game business as an example, the training of AIGC 3D characters requires thousands of character data, and the generation of each character data has undergone a lot of steps of processing such as data source, data screening, annotation, modification, and rendering. The data itself contains various attributes, such as style, specific game, number of patches, data source, and screening time, etc. Based on the above data processing method, the data generated in many steps is processed in a web-based distributed manner, and then packaged into a database file for storage, and seamlessly connected with the corresponding character model training, significantly improving the ability of multi-modal data management and meeting various diverse data requirements for model training. The present application can be applied to the management of multi-modal training data with a data volume of tens of millions, realizing concurrent data visualization annotation, modification, and rendering, etc., realizing data management centralization, without relying on manual sorting, and optimizing the model training business.

[0113] The embodiment of the present application also provides a data processing device 700 applied to artificial intelligence, as Figure 11 shown Figure 11 shows a schematic structural diagram of a data processing device applied to artificial intelligence provided by an embodiment of the present application. The device may include the following modules:

[0114] Data list module 10: configured to respond to a data query request carrying data query information sent by a terminal, and send an associated data table corresponding to the data query information to the terminal, so that the terminal displays the associated data table;

[0115] Data acquisition module 20: configured to receive a data request for target pre-stored data in the associated data table sent by the terminal, acquire the target pre-stored data, and send it to the terminal, so that the terminal displays the target pre-stored data;

[0116] Data storage module 30: configured to receive annotation data generated by the terminal for the target pre-stored data, and store the annotation data, the target pre-stored data, and the data information of the target pre-stored data in an associated manner;

[0117] Training data screening module 40: For responding to a training data request carrying training data filtering information sent by a terminal, sending target data information that matches the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end, and the target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

[0118] In some embodiments, the data query information includes the data to be queried, the data information includes the pre-stored coding information of the pre-stored data, and the data list module 10 includes:

[0119] Query coding sub-module: For performing data coding on the data to be queried to obtain a query code;

[0120] Coding search sub-module: For performing similarity-based coding search in a preset coding library according to the query code to obtain a pre-stored code similar to the query code. The preset coding library is used to store the pre-stored codes of multiple pre-stored data, and the pre-stored code corresponding to the pre-stored coding information is obtained by performing data coding based on the coding model corresponding to the pre-stored data;

[0121] First table generation sub-module: For generating an association data table based on the data information corresponding to the similar pre-stored codes.

[0122] In some embodiments, the query coding sub-module includes:

[0123] Feature coding unit: For inputting the data to be queried into the coding model corresponding to the data modality of the data to be queried for feature coding to obtain query data features;

[0124] Coding conversion unit: For performing coding conversion on the query data features according to the coding algorithm corresponding to the preset coding library to obtain a query code.

[0125] In some embodiments, the data query information includes at least one filtering condition information; the data list module 10 includes:

[0126] Data information matching sub-module: For querying data information that matches the filtering condition information in the target database, and the target database is used to associatively store the pre-stored data and the data information of the pre-stored data;

[0127] Second table generation sub-module: For generating an association data table based on the data information that matches the filtering condition information.

[0128] In some embodiments, in the case of receiving data query requests sent concurrently by multiple terminals, the data list module 10 includes:

[0129] Dataset determination sub-module: For each data query request, based on the data query information, determine the initial associated dataset in the target database, where the initial associated dataset includes multiple data information that matches the data query information;

[0130] Third table generation sub-module: Used to screen out some data information from the initial associated dataset to generate an associated data table corresponding to the current data query request, and the data information in the associated data tables corresponding to each data query request is mutually exclusive.

[0131] In some embodiments, the apparatus further includes:

[0132] Pre-stored data receiving module: Used to receive the pre-stored data sent by the terminal, where the pre-stored data includes three-dimensional modal data, rendering data of the three-dimensional modal data, and additional data of the three-dimensional modal data. The rendering data is obtained by rendering the three-dimensional modal data, and the additional data includes at least one other modal data for describing the three-dimensional modal data;

[0133] Encoding module: Used to perform data encoding on the three-dimensional modal data and the additional data to obtain pre-stored encoding;

[0134] The data storage module is further used to store the three-dimensional modal data, rendering data, additional data, and pre-stored encoding into the target database, and feedback the storage address information of the three-dimensional modal data, rendering data, additional data, and pre-stored encoding to the terminal, so that the terminal generates initial data information in the first data format based on the basic information and storage address information of the pre-stored data;

[0135] Format conversion module: Used to receive the initial data information sent by the terminal, and convert the initial data information into data information in the second data format and store it in the target database.

[0136] In some embodiments, the encoding module includes:

[0137] Rendering sub-module: Used to render the three-dimensional modal data based on multiple preset perspectives to obtain perspective images corresponding to each of the multiple preset perspectives;

[0138] Visual encoding sub-module: Used to input each perspective image into a data encoding model to obtain image features corresponding to each perspective image;

[0139] Encoding conversion sub-module: Used to perform encoding conversion on each image feature based on the encoding algorithm corresponding to the preset encoding library to obtain the pre-stored encoding corresponding to the three-dimensional modal data.

[0140] In some embodiments, the additional data includes at least one of the data text and point cloud data corresponding to the three-dimensional modal data; the encoding module includes:

[0141] The visual encoding sub-module is also used to input data text into a data encoding model for text feature encoding to obtain text features;

[0142] The point cloud encoding sub-module: is used to input point cloud data into a point cloud encoding model for feature encoding that aligns point cloud features with image text features to obtain point cloud features;

[0143] The encoding conversion sub-module is also used to perform encoding conversion on the text features and point cloud features based on the encoding algorithm corresponding to a preset encoding library to obtain the pre-stored encoding corresponding to the text features and the pre-stored encoding corresponding to the point cloud features.

[0144] It should be noted that the above device embodiments and method embodiments are based on the same implementation manner.

[0145] An embodiment of the present application provides a device, which can be a terminal or a server, including a processor and a memory. At least one instruction or at least one program segment is stored in the memory, and the at least one instruction or the at least one program segment is loaded and executed by the processor to implement the data processing method for artificial intelligence provided in the above method embodiment.

[0146] The memory can be used to store software programs and modules. The processor executes various functional applications and anomaly detections by running the software programs and modules stored in the memory. The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for functions, etc.; the data storage area can store data created according to the use of the device, etc. In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Correspondingly, the memory can also include a memory controller to provide the processor with access to the memory.

[0147] The method embodiments provided in the embodiments of the present application can be executed in electronic devices such as mobile terminals, computer terminals, servers, or similar computing devices. Figure 12 It is a hardware structure block diagram of an electronic device for a data processing method for artificial intelligence provided in an embodiment of the present application. As Figure 12As shown, the electronic device 900 can vary significantly due to different configurations or performances. It may include one or more Central Processing Units (CPUs) 910 (the processor 910 may include, but is not limited to, a processing device such as a microprocessor MCU or a Field Programmable Gate Array FPGA), a memory 930 for storing data, and one or more storage media 920 for storing application programs 923 or data 922 (such as one or more mass storage devices). Among them, the memory 930 and the storage media 920 can be transient storage or persistent storage. The program stored in the storage media 920 may include one or more modules, and each module may include a series of instruction operations on the electronic device. Further, the central processor 910 can be configured to communicate with the storage media 920 and execute a series of instruction operations in the storage media 920 on the electronic device 900. The electronic device 900 may also include one or more power supplies 960, one or more wired or wireless network interfaces 950, one or more input / output interfaces 940, and / or one or more operating systems 921, such as Windows Server TM , Mac OS X TM , Unix TM , LinuxTM, FreeBSDTM, and so on.

[0148] The input / output interface 940 can be used to receive or send data via a network. Specific examples of the above network may include a wireless network provided by the communication provider of the electronic device 900. In one example, the input / output interface 940 includes a Network Interface Controller (NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one example, the input / output interface 940 can be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0149] Those of ordinary skill in the art can understand that Figure 12 the structure shown is only schematic and does not limit the structure of the above electronic device. For example, the electronic device 900 may also include more or fewer components than those shown Figure 12 in the figure, or have a different configuration from that shown Figure 12 in the figure.

[0150] An embodiment of the present application further provides a computer-readable storage medium, which can be disposed in an electronic device to store at least one instruction or at least one segment of program related to an anomaly detection method in the method embodiment. The at least one instruction or the at least one segment of program is loaded and executed by the processor to implement the anomaly detection method provided in the above method embodiment.

[0151] Optionally, in this embodiment, the above storage medium may be located in at least one of multiple network servers in a computer network. Optionally, in this embodiment, the above storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disc.

[0152] According to one aspect of the present application, there is provided a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to enable the computer device to execute the methods provided in the above various optional implementation manners.

[0153] The data processing method, apparatus, device, storage medium, server, terminal, and program product provided by the present application can respond to a data query request carrying data query information sent by a terminal, automatically generate a corresponding associated data table, enable the terminal to display and submit a corresponding data request, and then send corresponding pre-stored data, so as to facilitate operations such as data annotation by relevant personnel of the terminal. After receiving the annotated data, it is associated and stored with the corresponding pre-stored data and the data information of the pre-stored data. While storing it as training data, when receiving a training data request, it can match the corresponding data information based on the training data filtering information and send it to the terminal, and seamlessly connect to the model training end, enabling the model training end to automatically obtain the corresponding training data, realizing the centralization of the background management of training data. It can be applied to the automatic storage, processing, and distribution of multi-model training data, improving the storage, processing, and sorting efficiency, classification accuracy of pre-training data, as well as the distribution efficiency and accuracy of training data. It can be applied to the business lines of multiple models to optimize the business process processing. It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above specific embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0154] Each embodiment in the present application is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the embodiments of the apparatus, device, and storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0155] Those of ordinary skill in the art can understand that all or part of the steps to implement 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 above-mentioned storage medium can be a read-only memory, a disk, or an optical disc, etc.

[0156] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A data processing method applied to artificial intelligence, characterized in that, The method includes: In response to a data query request carrying data query information sent by a terminal, sending an associated data table corresponding to the data query information to the terminal, so that the terminal displays the associated data table; Receiving a data request for target pre-stored data in the associated data table sent by the terminal, obtaining the target pre-stored data and sending it to the terminal, so that the terminal displays the target pre-stored data; Receiving annotation data generated by the terminal for the target pre-stored data, and associatively storing the annotation data, the target pre-stored data, and the data information of the target pre-stored data; In response to a training data request carrying training data filtering information sent by the terminal, sending target data information matching the training data filtering information to the terminal, so that the terminal sends the target data information to a model training end, and the target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

2. The method according to claim 1, characterized in that, The data query information includes data to be queried, the data information includes pre-stored coding information of the pre-stored data, and the generation method of the associated data table includes: Performing data coding on the data to be queried to obtain a query code; Performing similarity-based coding search in a preset coding library according to the query code to obtain pre-stored codes similar to the query code, where the preset coding library is used to store pre-stored codes of multiple pre-stored data, and the pre-stored code corresponding to the pre-stored coding information is obtained by performing data coding based on a coding model corresponding to the pre-stored data; Generating the associated data table based on the data information corresponding to the similar pre-stored codes.

3. The method according to claim 2, characterized in that, The performing data coding on the data to be queried to obtain a query code includes: Inputting the data to be queried into a coding model corresponding to the data modality of the data to be queried for feature coding to obtain query data features; Performing coding conversion on the query data features according to a coding algorithm corresponding to the preset coding library to obtain the query code.

4. The method according to claim 1, wherein The data query information includes at least one filtering condition information; the generation method of the associated data table includes: Querying data information matching the filtering condition information in a target database, where the target database is used to associatively store the pre-stored data and the data information of the pre-stored data; Generating the associated data table based on the data information matching the filtering condition information.

5. The method according to any one of claims 1-4, characterized in that, In the case of receiving data query requests concurrently sent by multiple terminals, the sending the associated data table corresponding to the data query information to the terminal includes: For each data query request, determining an initial associated data set in the target database based on the data query information, where the initial associated data set includes multiple data information matching the data query information; Selecting some of the data information from the initial associated data set to generate an associated data table corresponding to the current data query request, and the data information in the associated data tables corresponding to each data query request is mutually exclusive.

6. The method according to any one of claims 1-4, characterized in that, The method further includes: Receive the pre-stored data sent by the receiving terminal, where the pre-stored data includes three-dimensional modal data, rendering data of the three-dimensional modal data, and additional data of the three-dimensional modal data. The rendering data is obtained by rendering the three-dimensional modal data, and the additional data includes at least one other modal data for describing the three-dimensional modal data; Perform data encoding on the three-dimensional modal data and the additional data to obtain a pre-stored encoding; Store the three-dimensional modal data, the rendering data, the additional data, and the pre-stored encoding in a target database, and feedback the storage address information of the three-dimensional modal data, the rendering data, the additional data, and the pre-stored encoding to the terminal, so that the terminal generates initial data information in a first data format based on the basic information of the pre-stored data and the storage address information; Receive the initial data information sent by the terminal, and convert the initial data information into data information in a second data format and store it in the target database.

7. The method according to claim 6, characterized in that The generation method of the pre-stored encoding corresponding to the three-dimensional modal data includes: Render the three-dimensional modal data based on multiple preset perspectives to obtain perspective images corresponding to the multiple preset perspectives; Input each of the perspective images into a data encoding model to obtain image features corresponding to each of the perspective images; Perform encoding conversion on each of the image features based on the encoding algorithm corresponding to the preset encoding library to obtain the pre-stored encoding corresponding to the three-dimensional modal data.

8. The method according to claim 6, characterized in that, The additional data includes at least one of data text and point cloud data corresponding to the three-dimensional modal data; the generation method of the pre-stored encoding corresponding to the additional data includes: Input the data text into a data encoding model for text feature encoding to obtain text features; Input the point cloud data into a point cloud encoding model for feature encoding that aligns point cloud features with image text features to obtain point cloud features; Perform encoding conversion on the text features and the point cloud features based on the encoding algorithm corresponding to the preset encoding library to obtain the pre-stored encoding corresponding to the text features and the pre-stored encoding corresponding to the point cloud features.

9. A data processing device applied to artificial intelligence, characterized in that, The device includes: A data list module: used to respond to a data query request carrying data query information sent by the terminal, and send an associated data table corresponding to the data query information to the terminal, so that the terminal displays the associated data table; A data acquisition module: used to receive a data request for target pre-stored data in the associated data table sent by the terminal, acquire the target pre-stored data and send it to the terminal, so that the terminal displays the target pre-stored data; A data storage module: used to receive the annotation data generated by the terminal for the target pre-stored data, and store the annotation data, the target pre-stored data, and the data information of the target pre-stored data in an associated manner; Training data screening module: configured to respond to a training data request carrying training data filtering information sent by the terminal, and send target data information matching the training data filtering information to the terminal, so that the terminal sends the target data information to the model training end, and the target data information is used to instruct the model training end to obtain corresponding pre-stored data and annotation data.

10. A computer-readable storage medium, characterized in that, At least one instruction or at least one program segment is stored in the storage medium, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the data processing method applied to artificial intelligence as described in any one of claims 1-8.

11. A computer device, characterized in that, The device includes a processor and a memory, and at least one instruction or at least one program segment is stored in the memory. The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the data processing method applied to artificial intelligence as described in any one of claims 1-8.

12. A computer program product or a computer program, characterized in that, The computer program product or the computer program includes computer instructions, and when the computer instructions are executed by a processor, the data processing method applied to artificial intelligence as described in any one of claims 1-8 is implemented.

Images