Management multiplexing system of computer vision model

Abstract

The invention relates to a management multiplexing system of a computer vision model, and relates to the technical field of artificial intelligence. The system comprises an asset management subsystem used for carrying out storage management on existing sample data and an existing computer vision model, and the sample data represents data used for training the computer vision model; and the operation management subsystem is used for carrying out target identification processing on the obtained video to be monitored by calling an existing computer vision model and generating identification information, and the identification information is information describing an identified target to be identified. Through the system, existing artificial intelligence related models and data assets can be integrated, continuous accumulation of the sample data and the models is realized, and reuse values of the existing sample data and the models can be further exerted so as to meet requirements of business applications.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a management and reuse system for computer vision models. Background Art

[0002] After more than half a century of development and evolution, AI technology has entered a new stage, driven by new theories and technologies such as mobile Internet, big data, cloud computing, supercomputing, and brain science, as well as strong demands for economic and social development. At present, AI technology has the conditions for implementation in various fields.

[0003] In power transmission and transformation projects, AI-related businesses are gradually increasing, especially in computer vision, such as detecting safety risks by training algorithm models. However, some problems have gradually been exposed in the management of related projects: for current AI-related businesses, there is a lack of effective management tools and means for sample data and algorithm models in the process; sample data and model assets in the project have not been "precipitated", and the reuse value of sample data and model algorithms has not been fully utilized. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a management and reuse system for computer vision models to solve the above technical problems.

[0005] The technical solution of the present invention to solve the above technical problem is as follows: A management and reuse system for computer vision models, comprising:

[0006] An asset management subsystem, used for storing and managing existing sample data and existing computer vision models, where the sample data represents data used for training the computer vision model;

[0007] The operation management subsystem is used to perform target recognition processing on the acquired video to be monitored by calling the existing computer vision model, and generate recognition information, wherein the recognition information is information describing the recognized target to be recognized.

[0008] The beneficial effects of the present invention are as follows: the system forms an asset management subsystem, which realizes the storage management of existing sample data and computer vision models. It also forms an operation management subsystem to provide support for the user side, serve specific landing applications, and realize the reuse of existing computer vision models. Through this system, it is possible to integrate existing artificial intelligence-related models and data assets, realize the continuous accumulation of sample data and models, and further give play to the reuse value of existing sample data and models to meet the needs of business applications.

[0009] Based on the above technical solution, the present invention can also be improved as follows.

[0010] Furthermore, the asset management subsystem includes:

[0011] The sample library management module is used to store and manage existing sample data and provide sample annotation functions;

[0012] The model library management module is used to store and manage existing computer vision models and to provide model training functions.

[0013] Furthermore, the sample data includes original sample data and annotated sample data, and the sample library management module includes:

[0014] An original material management unit, used for storing and managing the original sample data;

[0015] A labeling data management unit, used for storing and managing the labeling sample data;

[0016] The online annotation unit is used to provide sample annotation function.

[0017] Furthermore, the online annotation unit includes:

[0018] An online annotation subunit, which is used to provide image classification annotation tools and object detection annotation tools;

[0019] The multi-person annotation subunit is used to provide annotation team management functions and annotation task management functions.

[0020] Furthermore, the model library management module includes:

[0021] A model management unit, used to store and manage existing computer vision models;

[0022] The model training unit is used to train the existing computer vision model based on the existing sample data.

[0023] Furthermore, the model library management module also includes a model evaluation unit, which is used to test the existing computer vision model and generate performance evaluation information, wherein the performance evaluation information is information describing the performance of the existing computer vision model.

[0024] Further, the model evaluation unit comprises:

[0025] An online testing subunit, used to obtain test data, and test the existing computer vision model according to the test data to generate test results;

[0026] The performance evaluation subunit is used to obtain user evaluation information and generate performance evaluation information according to the user evaluation information and the test result, wherein the user evaluation information is information describing the user's evaluation of the existing computer vision model.

[0027] Furthermore, the asset management subsystem also includes a system management module, and the system management module is used to manage the basic information of users and the permissions of users.

[0028] Furthermore, the operation management subsystem includes:

[0029] A video access module, used to obtain the video to be monitored;

[0030] A model access module, used to obtain an existing computer vision model from the asset management subsystem;

[0031] A video processing module, used to perform target recognition processing on the video to be monitored according to an existing computer vision model and generate recognition information;

[0032] The video alarm module is used to provide identification information query function.

[0033] Furthermore, the video processing module includes:

[0034] A parameter acquisition unit, used to acquire monitoring parameters, wherein the monitoring parameters include monitoring frequency and monitoring time;

[0035] The polling scheduling unit is used to perform target recognition processing on the video to be monitored through an existing computer vision model according to the monitoring parameters and generate recognition information. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic diagram of a management and reuse system for a computer vision model according to the present invention. DETAILED DESCRIPTION

[0037] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0038] like Figure 1 As shown, this embodiment provides a management and reuse system for a computer vision model, including:

[0039] An asset management subsystem, used for storing and managing existing sample data and existing computer vision models, where the sample data represents data used for training the computer vision model;

[0040] The operation management subsystem is used to perform target recognition processing on the acquired video to be monitored by calling the existing computer vision model, and generate recognition information, wherein the recognition information is information describing the recognized target to be recognized.

[0041] This system forms an asset management subsystem, which realizes the storage management of existing sample data and computer vision models. It also forms an operation management subsystem to provide support for the user side, serve specific landing applications, and realize the reuse of existing computer vision models. Through this system, it is possible to integrate existing artificial intelligence-related models and data assets, realize the continuous accumulation of sample data and models, and further give play to the reuse value of existing sample data and models to meet the needs of business applications.

[0042] Optionally, in an embodiment, the asset management subsystem includes:

[0043] The sample library management module is used to store and manage existing sample data and provide sample annotation functions;

[0044] The model library management module is used to store and manage existing computer vision models and to provide model training functions.

[0045] In the asset management subsystem, the sample library management module solves the storage management and production problems of sample data. The model library management module solves the information management and application problems of computer vision models.

[0046] Optionally, in an embodiment, the sample data includes original sample data and labeled sample data. The original sample data represents the original image material that has not been labeled and is used to train the computer vision model. The labeled sample data represents the labeled image material that is used to train the computer vision model.

[0047] The sample library management module includes:

[0048] An original material management unit, used for storing and managing the original sample data;

[0049] A labeling data management unit, used for storing and managing the labeling sample data;

[0050] The online annotation unit is used to provide sample annotation function.

[0051] Specifically, the original material management unit includes an external data subunit and a material library. The external data subunit is used to connect to an external data source, manage and view the external data source, and transmit data between the local and external data sources. The material library is used to store original sample data, and can import local data and data from external data sources into the material library and store them as original sample data in the material library. The material library can also export the original sample data in the material library to local and external data sources.

[0052] The annotation data management unit mainly provides storage and management functions for annotation sample data. It can create different annotation data sets for tasks such as helmet detection and seat belt detection, manage the annotation sample data in the form of data sets, and users can view the specific data in each data set. Through the annotation data management unit, users can also edit the data in each data set, and the annotation data management unit records the modified content and modified version. In addition, through the annotation data management unit, users can also export the data set to the local computer, or generate a shared link for sharing.

[0053] Optionally, in an embodiment, the online labeling unit includes:

[0054] An online annotation subunit, which is used to provide image classification annotation tools and object detection annotation tools;

[0055] The multi-person annotation subunit is used to provide annotation team management functions and annotation task management functions.

[0056] The online annotation unit mainly provides online annotation and multi-person annotation functions for sample data. It can perform online annotation according to different training tasks and generate standardized annotation files for further model training or export.

[0057] The image classification annotation tool is used for online image classification annotation. Users can use the image classification annotation tool to annotate the image with the category label. The target detection annotation tool is used to select and annotate the target, such as a rectangular box. Users can use the target detection annotation tool to select the target and add the corresponding label.

[0058] The annotation team management function can manage and edit the annotation team, team members and other information. The annotation task management function can manage annotation tasks, distribute tasks, and review tasks. For task review, including batch review and sheet-by-sheet review, reviewers can accept or reject the annotation results submitted by the annotators, and the reviewed data will be saved to the corresponding data set for storage.

[0059] Optionally, in an embodiment, the model library management module includes:

[0060] A model management unit, used to store and manage existing computer vision models;

[0061] The model training unit is used to train the existing computer vision model based on the existing sample data.

[0062] The computer vision direction mainly refers to scenarios such as target detection and image classification. The computer vision model refers to models related to computer vision, such as convolutional neural network models, YOLO (You Only Look Once) models, and SSD (Single Shot Multibox Detector) models.

[0063] The model management unit is used to import, export, update, view, and store computer vision models. The model management unit also generates a model list so that users can view each model. It also records the version information of each model to facilitate model adjustment and update.

[0064] Through the model training unit, the sample data in the sample library management module can be used to train the existing basic computer vision model to generate the model required in practice. The trained model is imported into the model management unit for storage so that it can be directly reused in subsequent related work.

[0065] Optionally, in an embodiment, the model library management module further includes a model evaluation unit, and the model evaluation unit is used to test the existing computer vision model and generate performance evaluation information, wherein the performance evaluation information is information describing the performance of the existing computer vision model.

[0066] Optionally, in an embodiment, the model evaluation unit includes:

[0067] An online testing subunit, used to obtain test data, and test the existing computer vision model according to the test data to generate test results;

[0068] The performance evaluation subunit is used to obtain user evaluation information and generate performance evaluation information according to the user evaluation information and the test result, wherein the user evaluation information is information describing the user's evaluation of the existing computer vision model.

[0069] The test data can be the original sample data obtained from the original material management unit to test the computer vision model and generate test results. The test data is processed by manual annotation or other existing models, and the obtained results are compared with the test results generated by the online test subunit to calculate parameters such as accuracy, providing a reference for subsequent model performance evaluation.

[0070] User evaluation information includes model calculation speed, accuracy, stability, etc. The performance evaluation subunit can record the performance of each model, which facilitates the selection of models in subsequent use.

[0071] The model library management module also includes a container management unit and a model publishing unit. Due to the different frameworks and dependent environments used by different algorithm models, the configuration during deployment and application is very troublesome. In order to ensure that the model can be independently deployed and applied, the system uses docker virtual containers to deploy all algorithm services in different docker containers according to the dependent environments. By managing virtual containers, it is convenient for the rapid deployment and migration of algorithm models. The container management unit mainly provides management functions for virtual container information deployed for model training. According to the environmental requirements of different models, different virtual containers are customized to meet the migration and rapid deployment requirements of reasoning analysis services. The container management unit can be used to import, export, update, and view containers.

[0072] The model publishing unit mainly provides the platform public cloud deployment and service publishing functions for the trained models, and also supports the management of external reasoning service information through third-party registration.

[0073] Optionally, in an embodiment, the asset management subsystem further includes a system management module, and the system management module is used to manage the user's basic information and the user's permissions.

[0074] The system management module includes a user management unit, a permission management unit and a login management unit. The user management unit includes: a user registration subunit for users to register with the system; a user information maintenance subunit for users to maintain basic information. The permission management unit includes: a role management subunit for users to manage roles; a role permission management subunit for users to manage role permissions. The login management unit provides users with a login function.

[0075] The asset management subsystem also includes an application management module and an open service module. The application management module mainly provides management of application and service permission authentication, including: application management unit, which is used for application information list and addition, deletion, modification and query; permission management unit, which is used for application and service permission configuration management. The open service module includes: API document unit, which is used to generate API online documents, providing interface description, request description, return description and historical version information; online trial unit, which is used for online trial of models on the Web side.

[0076] Optionally, in an embodiment, the operation management subsystem includes:

[0077] A video access module, used to obtain the video to be monitored;

[0078] A model access module, used to obtain an existing computer vision model from the asset management subsystem;

[0079] A video processing module, used to perform target recognition processing on the video to be monitored according to an existing computer vision model and generate recognition information;

[0080] The video alarm module is used to provide identification information query function.

[0081] Through the video access module, you can connect with the video acquisition device, import the video collected by the video acquisition device into the system, or directly upload the local video. In addition, through the video access module, you can also preview the real-time video screen.

[0082] The model access module can provide a list of computer vision models authorized by the asset management subsystem, and users can view, download, and perform other operations on each model.

[0083] Optionally, in an embodiment, the video processing module includes:

[0084] A parameter acquisition unit, used to acquire monitoring parameters, wherein the monitoring parameters include monitoring frequency and monitoring time;

[0085] The polling scheduling unit is used to perform target recognition processing on the video to be monitored through an existing computer vision model according to the monitoring parameters and generate recognition information.

[0086] In some embodiments, multiple polling scheduling tasks can be set at the same time. Multiple videos to be monitored are accessed through the video access module, and the monitoring parameters corresponding to each video to be monitored are respectively obtained through the parameter acquisition unit. The monitoring parameters are data set by the user, and also include a computer vision model in the model access module corresponding to the current video to be monitored. Through the polling scheduling unit, each video to be monitored is processed separately. In the video alarm module, the identification information corresponding to each video to be monitored can be queried.

[0087] The operation management subsystem also includes a data overview module, which is used to view the basic information of the operation management subsystem, including statistical information such as access videos, deployment algorithms, task creation, detection count statistics, and hardware information such as CPU, GPU, memory, and network.

[0088] In summary, this system aims to realize the storage management, online processing and reuse of artificial intelligence-related assets in the direction of computer vision (mainly including sample data and algorithm models). An asset management subsystem with a sample library management module and a model library management module as the core has been formed. In addition to providing storage management capabilities for sample data and computer vision models, it also provides online sample annotation review and online training capabilities for algorithm models. An operation management subsystem has also been formed to provide support for the user side, serve specific landing applications, and realize the reuse of existing computer vision models. Through this system, it is possible to integrate existing artificial intelligence-related models and data assets, build an artificial intelligence model library and sample library, and continuously accumulate data assets, laying the foundation for further research and application in the future. It can further give play to the reuse value of existing sample data and algorithm models to meet the needs of internal business applications of the company.

[0089] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0090] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A management and reuse system for computer vision models, characterized in that: include: An asset management subsystem, used for storing and managing existing sample data and existing computer vision models, where the sample data represents data used for training the computer vision model; The operation management subsystem is used to perform target recognition processing on the acquired video to be monitored by calling the existing computer vision model, and generate recognition information, wherein the recognition information is information describing the recognized target to be recognized.

2. A computer vision model management and reuse system according to claim 1, characterized in that: The asset management subsystem includes: The sample library management module is used to store and manage existing sample data and provide sample annotation functions; The model library management module is used to store and manage existing computer vision models and to provide model training functions.

3. A computer vision model management and reuse system according to claim 2, characterized in that: The sample data includes original sample data and labeled sample data. The sample library management module includes: An original material management unit, used for storing and managing the original sample data; A labeling data management unit, used for storing and managing the labeling sample data; The online annotation unit is used to provide sample annotation function.

4. A computer vision model management and reuse system according to claim 3, characterized in that: The online annotation unit comprises: An online annotation subunit, which is used to provide image classification annotation tools and object detection annotation tools; The multi-person annotation subunit is used to provide annotation team management functions and annotation task management functions.

5. A computer vision model management and reuse system according to claim 2, characterized in that: The model library management module includes: A model management unit, used to store and manage existing computer vision models; The model training unit is used to train the existing computer vision model based on the existing sample data.

6. A computer vision model management and reuse system according to claim 5, characterized in that: The model library management module also includes a model evaluation unit, which is used to test the existing computer vision model and generate performance evaluation information, where the performance evaluation information is information describing the performance of the existing computer vision model.

7. A computer vision model management and reuse system according to claim 6, characterized in that: The model evaluation unit comprises: An online testing subunit, used to obtain test data, and test the existing computer vision model according to the test data to generate test results; The performance evaluation subunit is used to obtain user evaluation information and generate performance evaluation information according to the user evaluation information and the test result, wherein the user evaluation information is information describing the user's evaluation of the existing computer vision model.

8. A computer vision model management and reuse system according to claim 1, characterized in that: The asset management subsystem also includes a system management module, which is used to manage the basic information of users and the permissions of users.

9. A computer vision model management and reuse system according to claim 1, characterized in that: The operation management subsystem includes: A video access module, used to obtain the video to be monitored; A model access module, used to obtain an existing computer vision model from the asset management subsystem; A video processing module, used to perform target recognition processing on the video to be monitored according to an existing computer vision model and generate recognition information; The video alarm module is used to provide identification information query function.

10. A computer vision model management and reuse system according to claim 9, characterized in that: The video processing module comprises: A parameter acquisition unit, used to acquire monitoring parameters, wherein the monitoring parameters include monitoring frequency and monitoring time; The polling scheduling unit is used to perform target recognition processing on the video to be monitored through an existing computer vision model according to the monitoring parameters and generate recognition information.

Images