Car Vision Labeller WebApp

Client: Protected By NDA
Project Type: Product
Industry: Automotive
Problem
The client faced challenges with the manual labeling of a vast number of images required for
training object detection systems. The reliance on on-premise data storage resulted in a
slow and inconsistent labeling process. To improve efficiency and speed, Infinity-tech was
tasked with upgrading the system. This involved migrating the solution to the cloud and
redesigning its architecture, enabling labeling teams to work remotely and process data
more rapidly.
Solution
The solution is a standalone web application composed of two modules: Admin Panel and
Workspace. Users are assigned specific roles—Super Admin, Coordinator, Observer,
Labeling Manager, and Labeler—each with distinct functionalities.
Labeler Capabilities:

• Annotate objects using geometric figure-shaped masks (triangle, circle, rectangle,
  trapezoid, etc.)
• Tag objects
• Categorize objects into relevant groups
  Super Admin Capabilities:
• Manage organizations, users, and labeling groups
• Upload videos
• Assign project roles
  Coordinator Responsibilities:
• Oversee project and task management
  Observer and Labeling Manager Responsibilities:
• Monitor labelers’ performance and ensure project quality
  A labeler receives instructions in .xml format detailing objects that need labeling (e.g.,
  vehicles, pedestrians, road markings, road signs). These files are stored in remote media
  storage. The labeler manually annotates objects not identified by Machine Learning
  algorithms.
  Currently, the app supports only low color depth videos, meaning imported video files
  require preprocessing. The solution uses convolutional neural networks-based image
  preprocessing algorithms to prepare video frames for manual labeling.
  Challenge
  The system and its media storages were originally hosted on local servers in Berlin, causing
  frequent connection issues and server response delays for remote labelers. To improve
  system reliability and reduce response time, the Infinity-tech team migrated the application
  and its media storages to the Azure Cloud.
  To enhance scalability and distribute resource-intensive operations, the team redesigned the
  system architecture. Functions such as caching, queue messaging, and containerization were
  outsourced to external services—Redis, RabbitMQ, and Docker, among others.