Skip to Content
Getting StartedIntroduction to Avala

Introduction to Avala

Avala is a professional data annotation platform designed for machine learning teams building computer vision and perception models. We help you create high-quality labeled datasets efficiently.

Who Uses Avala?

  • Autonomous Vehicle Teams: Label camera images, LiDAR point clouds, and multi-sensor data
  • Robotics Companies: Annotate perception data for robot navigation and manipulation
  • AI/ML Teams: Create training datasets for object detection, segmentation, and classification
  • Research Labs: Build labeled datasets for computer vision research

Platform Capabilities

Data Types

Avala supports multiple data modalities:

TypeDescriptionUse Cases
ImagesSingle images in common formatsObject detection, classification
VideosFrame sequences with temporal trackingAction recognition, tracking
Point Clouds3D LiDAR scans3D object detection, segmentation
Multi-CameraSynchronized camera arraysSurround-view perception
MCAP/ROSROS bag files with sensor dataRobotics, AV development

Annotation Tools

Professional-grade annotation tools for every use case:

  • Bounding Boxes: 2D rectangular regions for object detection
  • Polygons: Arbitrary shapes for precise object boundaries
  • Semantic Segmentation: Pixel-level classification masks
  • 3D Cuboids: 3D bounding boxes in point cloud data
  • Polylines: Path and edge annotations
  • Keypoints: Landmark and pose annotations

Quality Control

Built-in quality assurance workflows:

  • Review Pipeline: Multi-stage review with accept/reject decisions
  • Issue Tracking: Flag and resolve annotation problems
  • Metrics Dashboard: Track annotation quality and throughput
  • Consensus: Compare annotations from multiple labelers

Team Collaboration

Enterprise-ready collaboration features:

  • Organizations: Group users and resources
  • Teams: Create focused workgroups within organizations
  • Roles: Owner, admin, and member permission levels
  • Work Assignment: Distribute tasks to annotators

Architecture Overview

┌─────────────────────────────────────────────────────────────┐
│                     Mission Control                          │
│              (Web UI - Flutter WASM)                        │
├─────────────────────────────────────────────────────────────┤
│                        REST API                              │
│                  (Django REST Framework)                     │
├──────────────┬──────────────┬──────────────┬───────────────┤
│   Datasets   │   Projects   │    Tasks     │ Organizations │
│   & Items    │  & Workflows │  & Results   │   & Teams     │
└──────────────┴──────────────┴──────────────┴───────────────┘

Key Concepts

Datasets

A dataset is a collection of data items (images, video frames, point clouds) to be annotated. Datasets can be:

  • Public or private
  • Owned by users or organizations
  • Organized into sequences (for video/temporal data)

Projects

A project defines an annotation workflow:

  • Which datasets to annotate
  • What annotation tools to use
  • Label taxonomy (object classes, attributes)
  • Quality control stages

Tasks

Tasks are individual annotation work units:

  • Each task corresponds to one or more data items
  • Tasks flow through states: pending → active → completed
  • Results are submitted by annotators and reviewed for quality

Organizations

Organizations provide team-level resource management:

  • Centralized billing and settings
  • Member invitations and role management
  • Team creation for workgroup organization

Next Steps

Last updated on
ChangelogCore Concepts