Core Concepts

Understanding these concepts will help you get the most out of Avala.

Datasets

A dataset is a collection of data items to be annotated. Datasets are the foundation of all annotation work in Avala.

Dataset Properties

Dataset Items

Each dataset contains items - individual data samples:

• For image datasets: each item is one image
• For video datasets: items are grouped into sequences (frames)
• For LiDAR datasets: items are point cloud scans

# List items in a dataset
items = requests.get(
    f"{BASE_URL}/datasets/{owner}/{slug}/items/",
    headers=headers
).json()

Sequences

Sequences group related items for temporal data:

• Video frames from the same recording
• LiDAR scans from a driving session
• Multi-camera captures at the same timestamp

Sequences enable:

• Frame-by-frame navigation
• Object tracking across frames
• Temporal consistency in annotations

Projects

A project defines an annotation workflow and connects datasets to annotation tasks.

Project Components

Project
├── Datasets (data sources)
├── Task Type (annotation method)
├── Label Taxonomy (object classes)
├── Quality Control (review stages)
└── Tasks (work units)

Project Status

Projects flow through states:

Task Types

Projects support different annotation methods:

Tasks

Tasks are individual work units within a project. Each task represents annotation work to be done on specific data items.

Task Lifecycle

PENDING → ACTIVE → COMPLETED
                ↓
            (Review)
                ↓
         ACCEPTED / REJECTED

Results

When an annotator completes a task, they submit a result:

• Contains the annotation data (boxes, polygons, etc.)
• Includes metadata (time spent, tool used)
• Goes through quality control review

# Submit annotation result
result = requests.post(
    f"{BASE_URL}/tasks/{task_uid}/results/",
    headers=headers,
    json={
        "data": {
            "objects": [
                {
                    "label": "car",
                    "bbox": [100, 150, 200, 300]
                }
            ]
        }
    }
)

Organizations

Organizations group users and resources for team collaboration.

Organization Structure

Organization
├── Members (users with roles)
├── Teams (workgroups)
├── Datasets (shared data)
├── Projects (shared workflows)
└── Settings (billing, permissions)

Member Roles

Teams

Teams organize members within an organization:

• Focus on specific projects or data types
• Have their own member lists
• Can be assigned to projects

Labels and Taxonomy

Predefined Labels

Projects use a label taxonomy - a set of predefined labels (classes):

{
  "labels": [
    {"name": "car", "color": "#FF0000"},
    {"name": "pedestrian", "color": "#00FF00"},
    {"name": "cyclist", "color": "#0000FF"}
  ]
}

Object Classification

For complex taxonomies, use classification configs:

• Hierarchical categories
• Attributes (color, size, occlusion)
• Conditional attributes based on class

Quality Control

Avala provides built-in quality assurance:

Reviews

Annotations go through review before acceptance:

1. Annotator submits result
2. Reviewer examines annotation
3. Reviewer accepts or rejects
4. Rejected work returns for correction

Annotation Issues

Track problems with annotation issues:

• Flag specific problems
• Assign to team members
• Track resolution status

Metrics

Monitor quality with built-in metrics:

• Acceptance rate
• Annotation time
• Inter-annotator agreement
• Issue frequency

Next Steps

• Set up Mission Control for your team
• Learn about API Authentication
• Configure Organizations