Skip to content

Open source annotation and labeling tool for image and video assets

License

Notifications You must be signed in to change notification settings

datagym-ai/datagym-core

Repository files navigation

DataGym.ai

DataGym.ai is a modern, web based workbench to label images and videos. It allows you to manage your projects and datasets, label data, control quality and build your own training data pipeline. With DataGym.ai´s API and Python SDK you can integrate it into your toolchain.

DataGym.ai Workspace

📒 Ressources

🧩 Features

  • Organize your data into different projects with tasks
    • Dashboard with useful statistics / overview
    • Tasks lifecycle with states (backlog, waiting, in progress, completed, skipped, reviewed)
    • Pagination, Filtering and Search
    • Integrated quality control / review process
  • Organize your media within datasets
    • Different storage types (direct upload, public url´s, aws s3 cloud storage)
    • Supported mime types: jpeg, png, mp4
    • Support of large high resolution images
  • Labeling features
    • Global classifications (image wide)
    • Image annotation
      • Variety of geometries: point, line, bounding box, polygons
      • Different classification types: text, checklists, option-box
      • Supports nested geometries (child-geometries)
    • Video annotation: Specialized editor for video labeling
      • Frame-by-frame navigation
      • Linear interpolation to track objects
      • Adjustable playback-speed
      • Analyze and extract video metadata (codec, framerate, duration, ...)
    • Image segmentation
      • Bitmap export
  • Feature-rich Workspace
    • Temporary screen manipulations: contrast, brightness, saturation
    • Hide unused geometry-groups for more clarity
    • Shortcut support
    • Panning and zooming, multi-select, moving, duplication
    • Supports transformation of the same geometry type
    • Context menu for geometries
  • Powerful REST API to build your own workflows
    • Python SDK Package
  • Data exporting- and importing (json)
    • Export your labeled data as json (works for images and videos)
    • Import your labeled data to refine your ml model
    • Export-/import your label configuration and use it in multiple projects

🎯 Quickstart

Running with docker-compose

The simplest way to run DataGym.ai locally is by using docker-compose.

  1. Download the docker-compose.yml from the projects root-directory
  1. Launch container using docker-compose up -d
  2. Wait until the initialization is done
  3. Navigate to localhost:8080

Local development, build manually

Build the whole project:

mvn clean install 

🗳️ Build with

  • Java / Spring Boot
  • Angular

👐 Contributing

We would love to receive contributions - please review our Contributing Guide for all relevant details.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details