Product Functions

• Integrated Management for Unified Image Tagging

  Intelligent One-stop Data Solution

  The traditional method separates data collection tagging from data modeling, which causes problems and potential risks, such as the unification of the format and interface, difficulty in controlling tagging quality in real time and delayed feedback from data to model. 4Paradigm provides a tagging platform through an integrated management for unified image tagging that covers the entire process from creating a tagging team, importing data to be tagged to tagging quality evaluation and management, to exporting tagging data. The product provides users with a full-process solution that includes intelligent image modeling.

  Multiple Image Application Scenarios

  4Paradigm’s tagging platform provides a comprehensive image tagging system, which includes image tagging benchmarks, such as image classification, object detection, text recognition and other labeling standards.

  Complete Tagging Format Specification

  Provides a unified tagging framework for standard tagging format and template, and image template basic toolkit that can assist users in the tagging process.

  Compatible With Common Data Formats

  Compatible with common image file formats, such as jpg, png, bmp, etc.

• Flexible Data Access, for the Convenience of Management

  Support Multiple Data Uploading Methods

  Supports local uploading, HDFS distributed file system uploading, FTP, tagging platform uploading and other data access methods.

  Visualized Data Management Interface

  Supports viewing, preview, and modification of tagging information for quick access to overall macroscopic and microscopic data, providing management from different dimensions, from managing individual images to collection of images.

• Professional Image Pre-processing

  Multiple Image Processing Functions

  Capable of more than ten image processing functions through the internal automation of the algorithm, including flipping, cropping, rotating, resizing, sharpening, super pixelation, gray processing and blurring, which can enrich image data to enhance the robustness of the model.

  Efficient Sample Enhancement Method

  Simply checking the option to complete pre-processing configuration. It also supports combined superpositioning, sample balancing, etc., to ensure the convenience of pre-processing functions, enhancing the samples efficiently, and improving the generalization ability of the machine learning models.

• Deep Computer Vision Model Engine

  Multiple Task Scenarios Template

  Deep model construction can be carried out for various scenarios (more scenarios are being added in the process), including image classification, object detection, text positioning, text recognition, and the results are comparable to the algorithmic benchmark in academia.

  Deep Learning Algorithm Library

  Built-in Fast R_CNN, ResNET, CTPN, DeepText and other deep learning models to meet the diverse needs of users in different task scenarios.

  Transfer Learning Technology

  4Paradigm’s founding team members are pioneers in transfer learning technology. Transfer learning is considered by the industry to be “the next generation of artificial intelligence technology.” It is a new machine learning method that uses existing knowledge to address issues of different but related fields. HyperCycle CV uses transfer learning technology to train models faster and better.

  Multiple Operating Modes Applicable to Different Professional Backgrounds

  HyperCycle CV provides operating modes at three levels of difficulty: intelligent recommendation mode, precision adjustment mode and expert mode, which corresponds to business personnel, data engineers and professional AI practitioners respectively, so as to achieve threshold controllability, and to be used by people with different professional backgrounds.

  Real-time Training Process Display

  HyperCycle CV provides real-time training process display, such as real-time model training indicators, real-time invalid pictures, real-time error pictures. Users can visually obtain the current model training information through the interface, and put timely measures in place when abnormal training occurs to avoid a waste in hash rate and time.

• Convenient and Complete Online Operation and Maintenance

  Visualized Resource Monitoring

  Provides visual display of the usage of GPU, CPU, RAM and memory, so that users can monitor resource usage in real time, which is an additional layer of protection for the system’s operation stability in a visual and convenient manner.

  Model One-click Release, Automatic Configuration

  There are two types of resource configuration for trained models, the intelligent resource configuration and the on-demand configuration, in order to flexibly configure the CPU/GPU, RAM and other resources during the launch of the model application. After the application has launched, the service API will be automatically generated to facilitate decoupling and carry out flexible adaptation.

  Convenient Model Management and Monitoring

  The visualized function supports model information viewing and management, as well as providing detailed viewing of the construction process logs, achieving a clear and transparent model construction application process.

  Multiple Application Monitoring Indicators

  The system has an in-built monitoring function, including resource usage, service access, and model effect evaluation indicators. It provides the load monitoring of CPU, GPU and RAM during the model training phase, as well as providing model training logs after the training is completed, so that the training process can be traced back at any time. During the application service phase, real-time CPU, GPU, RAM, request and other load conditions are provided, so that the administrator can obtain information on the current running state of the smart image application in real time.