Artificial intelligence labeling and training integrated system and method

Abstract

The invention discloses an artificial intelligence labeling and training integrated system and method, belongs to the technical field of deep learning, and aims to solve the technical problems of data complexity, training flow splitting, environment incompatibility and user friendliness in a deep learning model training process. Comprising a data management module, a labeling module, a model management module, an environment support module, a model evaluation optimization module and a model export deployment module, supports a user to select a model and customize hyper-parameters of model training, builds a training environment by calling a preset Docker mirror image, calls an analysis tool to perform model evaluation on a trained model, and exports and deploys the model. And exporting the optimized model to a target format, and publishing the exported model to a target environment through a built-in deployment tool.

Description

Technical Field

[0001] The present invention relates to the technical field of deep learning, and specifically to an artificial intelligence annotation and training integrated system and method. Background Art

[0002] With the rapid development of artificial intelligence technology, deep learning models have been widely applied in fields such as computer vision, natural language processing, and speech recognition. However, the model training process usually faces the following problems:

[0003] 1. Complexity of data management: Model training highly depends on the quality and diversity of data. However, current data annotation tools and training tools are usually separated, and the data format conversion and cleaning processes are time-consuming and laborious, increasing the development difficulty;

[0004] 2. Fragmentation of the training process: In the traditional development process, data annotation, model training, evaluation and optimization, and deployment are usually completed by different tools, and developers need to frequently switch between multiple platforms, resulting in low efficiency;

[0005] 3. Environment compatibility issues: The process of setting up the training environment is complex and error-prone, and the compatibility between different platforms is insufficient, making it difficult to reproduce the model results;

[0006] 4. High technical threshold: Current tools are not user-friendly enough for non-technical background users, and the complexity of artificial intelligence development makes it difficult for small and medium-sized enterprises to quickly carry out related projects.

[0007] The problems of complex data, fragmented training process, environment incompatibility, and user unfriendliness in the deep learning model training process are technical problems that need to be solved. Summary of the Invention

[0008] The technical task of the present invention is to provide an artificial intelligence annotation and training integrated system and method to solve the technical problems of complex data, fragmented training process, environment incompatibility, and user unfriendliness in the deep learning model training process in view of the above deficiencies.

[0009] In a first aspect, an artificial intelligence annotation and training integrated system of the present invention includes a data management module, an annotation module, a model management module, an environment support module, a model evaluation and optimization module, and a model export and deployment module;

[0010] The data management module is used to support users to upload various types of data, perform data preprocessing on the data, and generate a statistical report of the preprocessed data for users to verify;

[0011] A variety of annotation tools are configured in the annotation module, which are used to support users in selecting annotation types according to project requirements, allocating annotation tasks for data annotation to team members, supporting team members to collaborate and execute annotation tasks through the annotation tools, checking the annotated data and outputting the annotated data in a predetermined format;

[0012] A variety of models are configured in the model management module, which are used to support users in selecting models and customizing hyperparameters for model training;

[0013] A variety of pre-configured Docker images are integrated in the environment support module, covering mainstream deep learning frameworks, and supporting users to upload custom images, which are used to build a training environment by calling the preset Docker images, support distributed training and multi-GPU parallel computing, and record and display the logs and performance curves during the model training process;

[0014] A variety of analysis tools are configured in the model evaluation and optimization module, which are used to call the analysis tools to evaluate the trained model, and support users to perform model optimization operations on the trained model according to the model evaluation results;

[0015] The model export and deployment module is used to export the optimized model to the target format and publish the exported model to the target environment through the built-in deployment tool.

[0016] Preferably, the data management module supports users to upload data by means of dragging, API interfaces or file import, supports users to upload data in batches, the supported data types include images, texts, audios and videos, and supports operations such as format conversion, data cleaning, data deduplication and data augmentation of the data to obtain preprocessed data.

[0017] Preferably, the annotation tools deployed in the annotation module include tools for performing object detection box annotation, classification annotation and semantic segmentation annotation, the supported annotation types selected by users include image classification and object detection, and the supported target formats for exporting the annotated data are COCO and VOC.

[0018] Preferably, the model management module is configured with models built based on YOLO, ResNet and Transformer algorithms, and supports users to adjust hyperparameters including learning rate and batch size.

[0019] Preferably, the deep learning frameworks covered in the environment support module include TensorFlow, PyTorch and MXNet.

[0020] Preferably, the analysis tools configured in the model evaluation and optimization module include a confusion matrix and an ROC curve, and evaluation metrics including accuracy, recall, and F1 value are configured to call the analysis tools and perform model evaluation on the trained model based on the evaluation metrics. The supported model optimization operations include hyperparameter tuning, model pruning, and model quantization.

[0021] Preferably, the model export and deployment module supports exporting model formats including ONNX and TensorFlow SavedModel, and supports deploying the exported model to the cloud, edge devices, or local servers.

[0022] In a second aspect, an artificial intelligence annotation training method of the present invention realizes data annotation and model training through an artificial intelligence annotation training system according to any one of the first aspects, including the following steps:

[0023] Data management: The user uploads various types of data, performs data preprocessing on the data, and generates a statistical report for the user to verify the preprocessed data.

[0024] Annotation: The user selects an annotation type according to the project requirements, assigns the annotation task of annotating the data to team members, and the team members collaborate and execute the annotation task through an annotation tool, check the annotated data, and output the annotated data in a predetermined format.

[0025] Model management: The user selects a model and customizes the hyperparameters for model training.

[0026] Environment support: Build a training environment by calling a preset Docker image, support distributed training and multi-GPU parallel computing, record the logs and performance curves during model training, and display the logs and performance curves.

[0027] Model evaluation and optimization: Call the analysis tool to perform model evaluation on the trained model, and the user performs model optimization operations on the trained model according to the model evaluation results.

[0028] Model export and deployment: Export the optimized model to the target format, and publish the exported model to the target environment through a built-in deployment tool.

[0029] The artificial intelligence annotation training integrated system and method of the present invention have the following advantages:

[0030] 1. Full-process integration: A one-stop solution from data import to model deployment, reducing the time and complexity for users to switch between multiple tools;

[0031] 2. High efficiency: Support for distributed training and multi-GPU acceleration, greatly improving the model training speed and meeting the needs of large-scale data processing;

[0032] 3. Compatibility and Reproducibility: By uniformly managing the training environment through Docker images, the problems of complex traditional environment configuration and poor reproducibility are solved;

[0033] 4. Task Collaboration Ability: The annotation module supports multi-person collaborative task allocation, improving the team work efficiency and ensuring the high quality of the annotated data;

[0034] 5. Support for Multi-task Scenarios: Covers algorithms and application scenarios in multiple fields such as computer vision, natural language processing, and speech recognition;

[0035] 6. Low Technical Threshold: Provides an intuitive graphical interface and automated tools, reducing the usage threshold for non-professionals and being suitable for small and medium-sized enterprises to quickly start artificial intelligence projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0037] The present invention will be further described below with reference to the drawings.

[0038] Figure 1 It is a working process block diagram of an artificial intelligence annotation and training integrated system for Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The present invention will be further described below with reference to the drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it. However, the embodiments given are not intended to limit the present invention. Without conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0040] The embodiments of the present invention provide an artificial intelligence annotation and training integrated system and method for solving the technical problems of complex data, fragmented training process, environment incompatibility, and user unfriendliness in the training process of deep learning models.

[0041] Embodiment 1:

[0042] An artificial intelligence annotation and training integrated system of the present invention includes a data management module, an annotation module, a model management module, an environment support module, a model evaluation and optimization module, and a model export and deployment module.

[0043] The data management module is used to support users in uploading various types of data, preprocess the data, and generate a statistical report from the preprocessed data for users to verify.

[0044] As a specific implementation of the data management module, this data management module supports users in uploading data through drag-and-drop, API interfaces, or file import, supports users in uploading data in batches, supports data types for upload including images, texts, audios, and videos, and supports operations such as format conversion, data cleaning, data deduplication, and data augmentation on the data to obtain the preprocessed data.

[0045] The data management module in this embodiment supports data import and preprocessing, supports batch import of multiple data types (such as pictures, texts, audios, videos, etc.), and provides functions of automatic format conversion, data cleaning, deduplication, and data augmentation. During the application process, users upload data to the platform through the drag-and-drop upload, API interface, or file import function, and the platform automatically completes the format conversion, denoising processing, and data augmentation of the data, and generates a statistical report for users to verify.

[0046] Multiple annotation tools are configured in the annotation module, which are used to support users in selecting the annotation type according to project requirements, assigning the annotation tasks of annotating the data to team members, supporting team members to collaborate and execute the annotation tasks through the annotation tools, and used to check the annotated data and output the annotated data in a predetermined format.

[0047] As a specific implementation in the annotation module, the annotation tools deployed in this module include tools for performing object detection box annotation, classification annotation, and semantic segmentation annotation. The annotation types supported by users include image classification and object detection, and support exporting annotated data in target formats of COCO and VOC.

[0048] The annotation module in this embodiment provides multiple annotation tools, including object detection box annotation, classification annotation, semantic segmentation annotation, etc., supports multi-person collaborative annotation tasks, and ensures that the annotation results conform to mainstream data formats (such as COCO, VOC, YOLO). Based on this module, an annotation task is created. Users select the annotation type (such as image classification, object detection) according to project requirements and assign the task to team members. After annotation is completed, the system automatically checks the integrity of the annotated data and stores it in standard formats such as COCO and VOC.

[0049] Multiple models are configured in the model management module, which are used to support users in selecting models and customizing the hyperparameters of model training.

[0050] As a specific implementation of the model management module, this module is configured with models built based on the YOLO, ResNet, and Transformer algorithms, and supports users to adjust hyperparameters including learning rate and batch size.

[0051] In this embodiment, the model management module incorporates multiple mainstream algorithms (such as YOLO, ResNet, Transformer), supports users to select models and adjust hyperparameters (such as learning rate, batch size, etc.) through the interface. During application, users select a preset model in the algorithm library (such as YOLO for object detection), and can also customize training parameters (such as number of iterations, learning rate). The platform automatically recommends algorithms according to the task and provides performance prediction.

[0052] The environment support module integrates multiple pre-configured Docker images, covering mainstream deep learning frameworks, and supports users to upload custom images to build a training environment by calling the preset Docker images, supports distributed training and multi-GPU parallel computing, and is used to record the logs and performance curves during model training and display the logs and performance curves.

[0053] Among them, the deep learning frameworks covered by the environment support module include TensorFlow, PyTorch, and MXNet.

[0054] In this embodiment, the environment support module integrates multiple pre-configured Docker images, covering mainstream deep learning frameworks (such as TensorFlow, PyTorch, MXNet), and users can also upload custom images to quickly build the required environment. During application, the training environment is quickly built by calling the preset Docker images, and distributed training and multi-GPU parallel computing are supported. The logs and performance curves during the training process will be visualized in real time, facilitating users to track the progress.

[0055] The model evaluation and optimization module is configured with multiple analysis tools, used to call the analysis tools to evaluate the trained model, and supports users to perform model optimization operations on the trained model according to the model evaluation results.

[0056] As a specific implementation of the model evaluation and optimization module, the analysis tools configured in this module include confusion matrix and ROC curve, and evaluation metrics including accuracy, recall, and F1 value are configured, used to call the analysis tools and evaluate the trained model based on the evaluation metrics. The supported model optimization operations include hyperparameter tuning, model pruning, and model quantization.

[0057] In this embodiment, the model evaluation and optimization module provides comprehensive evaluation metrics (such as accuracy, recall, F1 value, etc.), supports automated hyperparameter tuning, model pruning, and quantization functions to improve model performance. When applied, after training is completed, the model is subjected to a detailed performance evaluation through the model evaluation and optimization module, including analysis tools such as confusion matrices and ROC curves. Users can select hyperparameter optimization, model pruning, or quantization operations based on the evaluation results to further improve model efficiency.

[0058] The model export and deployment module is used to export the optimized model to the target format and publish the exported model to the target environment through the built-in deployment tool.

[0059] In this embodiment, the model formats supported by the model export and deployment module for export include ONNX and TensorFlow SavedModel, and it supports deploying the exported model to the cloud, edge devices, or local servers.

[0060] In this embodiment, the model export and deployment module supports exporting the optimized model to multiple formats (such as ONNX, TensorFlow SavedModel) and supports directly deploying it to the cloud, edge devices, or local servers. When applied, the optimized model can be exported to the target format with one click and directly published to the target environment through the built-in deployment tool. The platform supports the rapid deployment of models on edge devices, cloud APIs, and local servers.

[0061] Embodiment 2:

[0062] An artificial intelligence annotation training method of the present invention realizes data annotation and model training through the system disclosed in Embodiment 1, including the following steps:

[0063] Step S100 Data Management: The user uploads various types of data, performs data preprocessing on the data, and generates a statistical report of the preprocessed data for the user to verify.

[0064] Step S200 Annotation: The user selects the annotation type according to the project requirements, assigns the annotation task of annotating the data to team members, and the team members collaborate and execute the annotation task through the annotation tool, check the annotated data, and output the annotated data in a predetermined format.

[0065] Step S200 Model Management: The user selects a model and customizes the hyperparameters for model training.

[0066] Step S400 Environment Support: The training environment is built by calling a preset Docker image, supporting distributed training and multi-GPU parallel computing, recording the logs and performance curves during model training, and displaying the logs and performance curves.

[0067] Step S500 Model Evaluation and Optimization: Call an analysis tool to evaluate the trained model, and the user performs model optimization operations on the trained model according to the model evaluation results;

[0068] Step S600 Model Export and Deployment: Export the optimized model to the target format, and publish the exported model to the target environment through the built-in deployment tool.

[0069] The method of this embodiment integrates functions of data import, annotation, model training, evaluation and optimization, and deployment, and supports the processing of various data types such as pictures, texts, audios, and videos. Through Docker image technology and distributed computing capabilities, the platform realizes the complete process from data annotation to model deployment. The platform is designed with high efficiency and ease of use as the core, supports real-time training monitoring, automatic optimization, and multi-format model output, and is applicable to various artificial intelligence task scenarios such as computer vision, natural language processing, and speech recognition.

[0070] The above has introduced in detail the artificial intelligence annotation and training integrated system and method provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. An artificial intelligence annotation and training integrated system, characterized in that: It includes data management module, annotation module, model management module, environment support module, model evaluation and optimization module and model export and deployment module; The data management module is used to support users to upload various types of data, perform data preprocessing on the data, and generate statistical reports from the preprocessed data for user verification; The annotation module is configured with a variety of annotation tools to support users to select annotation types according to project requirements, to assign annotation tasks for data annotation to team members, to support team members to collaborate and perform annotation tasks through annotation tools, and to check the annotation data and output the annotation data in a predetermined format; The model management module is configured with a variety of models to support users in selecting models and customizing hyperparameters for model training; The environment support module integrates a variety of pre-configured Docker images, covers mainstream deep learning frameworks, and supports users to upload custom images. It is used to build a training environment by calling preset Docker images, supports distributed training and multi-GPU parallel computing, and is used to record logs and performance curves during model training and display them. The model evaluation and optimization module is configured with a variety of analysis tools for calling the analysis tools to perform model evaluation on the trained model, and supports the user to perform model optimization operations on the trained model according to the model evaluation results; The model export and deployment module is used to export the optimized model to the target format and publish the exported model to the target environment through the built-in deployment tool.

2. The artificial intelligence annotation and training integrated system according to claim 1, characterized in that: The data management module supports users to upload data by dragging and dropping, API interface or file import, supports users to upload data in batches, supports the upload of data types including images, text, audio and video, supports data format conversion, data cleaning, data deduplication and data enhancement operations to obtain pre-processed data.

3. The artificial intelligence annotation and training integrated system according to claim 1 is characterized in that: The annotation tools deployed in the annotation module include tools for performing target detection box annotation, classification annotation, and semantic segmentation annotation. The annotation types supported by users include image classification and target detection, and the export of annotation data in the target formats of COCO and VOC is supported.

4. The artificial intelligence annotation and training integrated system according to claim 1, characterized in that: The model management module is configured with models built based on YOLO, ResNet and Transformer algorithms, supporting users to adjust hyperparameters including learning rate and batch size.

5. The artificial intelligence annotation and training integrated system according to claim 1, characterized in that: The deep learning frameworks covered in the environment support module include TensorFlow, PyTorch, and MXNet.

6. The artificial intelligence annotation and training integrated system according to claim 1, characterized in that: The analysis tools configured in the model evaluation and optimization module include confusion matrix and ROC curve, and are configured with evaluation indicators including accuracy, recall rate and F1 value, which are used to call the analysis tools and perform model evaluation on the trained model based on the evaluation indicators. The supported model optimization operations include hyperparameter tuning, model pruning and model quantization.

7. The artificial intelligence annotation and training integrated system according to claim 1, characterized in that: The model export and deployment module supports exporting model formats including ONNX and TensorFlow SavedModel, and supports deploying the exported model in the cloud, edge devices or local servers.

8. An artificial intelligence annotation training method, characterized in that: Data labeling and model training are implemented by an artificial intelligence labeling training system as described in any one of claims 1 to 7, comprising the following steps: Data management: Users upload various types of data, pre-process the data, and generate statistical reports for the pre-processed data for user verification; Annotation: Users select annotation types based on project requirements and assign the task of annotating data to team members. Team members collaborate and perform annotation tasks using annotation tools, check the annotated data, and output the annotated data in a predetermined format. Model management: users select models and customize hyperparameters for model training; Environment support: Build a training environment by calling the preset Docker image, support distributed training and multi-GPU parallel computing, record logs and performance curves during model training, and display them; Model evaluation and optimization: Call the analysis tool to evaluate the trained model, and the user can optimize the trained model based on the model evaluation results. Model export and deployment: Export the optimized model to the target format and publish the exported model to the target environment through the built-in deployment tool.

Images