Quickstart guide

Get started with CVAT in minutes - create your first project, annotate data, and export results

This guide will walk you through creating your first annotation project in CVAT, from initial setup to exporting your labeled data.

Choose your deployment

Before you start, decide how you want to use CVAT:

CVAT Cloud (recommended)

The fastest way to get started - no installation required.

1. Sign up for CVAT Cloud

   1. Visit app.cvat.ai
   2. Click “Sign Up” and create your account
   3. Verify your email address
   4. Log in to start annotating

   The free tier includes up to 10 tasks and 500MB of data storage.

2. Access the platform

   Once logged in, you’ll see the CVAT dashboard where you can create projects and tasks.

Self-hosted (Docker)

Install CVAT locally for complete control over your data.

1. Prerequisites

   Ensure you have:

   • Docker Engine (19.03+)
   • Docker Compose (1.29+)
   • 4GB+ RAM available
   • Linux, macOS, or Windows with WSL2
2. Clone and start CVAT

   git clone https://github.com/cvat-ai/cvat
   cd cvat
   docker-compose up -d
   

   Wait for all containers to start (this may take several minutes on first run).

3. Create superuser account

   docker exec -it cvat_server bash -ic 'python3 manage.py createsuperuser'
   

   Follow the prompts to set up your admin account.

4. Access CVAT

   Open your browser and navigate to http://localhost:8080

Create your first project

Projects help organize related annotation tasks and define shared labels.

1. Navigate to Projects

   From the CVAT dashboard, click the Projects tab in the top navigation.

2. Create new project

   Click the + Create Project button.

   Fill in the project details:

   • Name: Choose a descriptive name (e.g., “Car Detection Dataset”)
   • Labels: Define the object classes you want to annotate

   Example: Object detection labels

   [
     {"name": "car", "color": "#ff0000"},
     {"name": "person", "color": "#00ff00"},
     {"name": "bicycle", "color": "#0000ff"}
   ]
   
3. Configure advanced settings (optional)

   • Set a bug tracker URL for issue reporting
   • Configure source and target storage (cloud storage)
   • Assign project owner and team members
4. Save the project

   Click Submit to create your project.

You can also create tasks without a project, but projects make it easier to manage multiple related tasks with consistent labels.

Create and upload data to a task

Tasks contain the actual data (images or videos) to be annotated.

1. Create a new task

   From your project page or the Tasks tab, click + Create Task.

   Configure the task:

   • Name: Descriptive task name (e.g., “Street scenes - batch 1”)
   • Project: Select the project you created (optional)
   • Labels: If not using a project, define labels here
2. Upload your data

   Choose your data source:

   Local files

   • Click Select files and choose images or videos from your computer
   • Supports: JPG, PNG, BMP, MP4, AVI, MOV, and more
   • You can select multiple files at once

   Remote sources

   Add URLs to images or videos:

   https://example.com/image1.jpg
   https://example.com/image2.jpg
   

   Cloud storage

   • Configure cloud storage connection (AWS S3, Azure, GCS)
   • Select files from your connected storage

   Share

   • Place files in the mounted share directory
   • Specify paths relative to the share root
3. Configure advanced options (optional)

   • Chunk size: Frames per segment (for large videos)
   • Overlap: Frame overlap between jobs for consistency
   • Segment size: Frames per job for distributing work
   • Image quality: Compression level (1-100)
   • Frame filter: Skip frames (e.g., step=10 for every 10th frame)
4. Submit and wait for processing

   Click Submit. CVAT will:

   1. Upload your files
   2. Process and compress images/videos
   3. Create jobs for annotation

   Processing time depends on data size. Large videos may take several minutes.

Annotate your data

Now you’re ready to start labeling!

1. Open a job

   From the task page, click on a job to open the annotation interface.

   The interface has three main areas:

   • Canvas: The image/video display area
   • Objects sidebar: List of annotations
   • Controls toolbar: Drawing and navigation tools
2. Create annotations

   Select a shape type from the left toolbar:

   Rectangle (bounding box)

   1. Click the Rectangle tool
   2. Select a label
   3. Click and drag on the canvas to draw a box
   4. Release to complete

   Shortcut: N key

   Polygon

   1. Click the Polygon tool
   2. Select a label
   3. Click points around the object perimeter
   4. Double-click or press N to close the polygon

   Shortcut: Shift+N

   Points

   1. Click the Points tool
   2. Select a label
   3. Click to place individual points
   4. Press N when done

   Polyline

   1. Click the Polyline tool
   2. Select a label
   3. Click points along the line
   4. Double-click to finish
3. Edit and refine annotations

   • Move: Click and drag annotations
   • Resize: Drag corner/edge handles
   • Delete: Select annotation and press Delete
   • Copy: Ctrl+C, paste with Ctrl+V
   • Attributes: Set in the right sidebar after selecting an annotation
4. Track objects in videos (interpolation)

   For video annotation, use Track mode:

   1. Switch to Track mode in the toolbar
   2. Create a shape on the first frame
   3. Move to another frame
   4. Adjust the shape position
   5. CVAT automatically interpolates between keyframes

   Use automatic tracking with AI models to follow objects across frames automatically.

5. Save your work

   Click Save (or Ctrl+S) to save annotations.

   Save frequently to avoid losing work. Auto-save is enabled by default but manual saves are recommended.

Export annotations

Once you’ve completed annotation, export your dataset in the format you need.

1. Open export menu

   From the task page, click Actions → Export task dataset.

2. Choose export format

   Select from 40+ supported formats:

   COCO

   COCO 1.0
   

   Popular format for object detection, segmentation, and keypoint detection.

   YOLO

   YOLO 1.1
   YOLOv8 Detection 1.0
   YOLOv8 Segmentation 1.0
   

   For training YOLO models.

   Pascal VOC

   Pascal VOC 1.1
   

   Classic format with XML annotations.

   CVAT for images

   CVAT for images 1.1
   

   CVAT’s native format, best for re-importing.

3. Download the dataset

   Click OK and wait for processing. A download link will appear when ready.

   The export includes:

   • Images or video frames
   • Annotation files in chosen format
   • Label mapping and metadata

Automate with CLI or SDK

For larger workflows, use the CVAT Python SDK or CLI:

Python SDK

from cvat_sdk import Client

# Connect to CVAT
client = Client(url='https://app.cvat.ai')
client.login(('username', 'password'))

# Create a task
task = client.tasks.create_from_data(
    spec={
        'name': 'My Task',
        'labels': [
            {'name': 'car'},
            {'name': 'person'}
        ]
    },
    resource_type='local',
    resources=['image1.jpg', 'image2.jpg']
)

print(f"Created task ID: {task.id}")

# Export annotations
task.export_dataset('YOLO 1.1', 'dataset.zip')

Install with:

pip install cvat-sdk

CLI

# Create a task
cvat-cli --auth user:password task create \
  --labels '[{"name": "car"}, {"name": "person"}]' \
  "My Task" \
  local \
  image1.jpg image2.jpg

# List tasks
cvat-cli --auth user:password task ls

# Export dataset
cvat-cli --auth user:password task export-dataset \
  --format "YOLO 1.1" \
  123 \
  output.zip

Install with:

pip install cvat-cli

Next steps

You’ve completed your first annotation workflow! Here’s what to explore next:

Key concepts

Learn about projects, tasks, jobs, labels, and CVAT’s architecture

Annotation guide

Explore advanced annotation features and tools

Quality control

Set up quality control workflows for your team

Dataset formats

Learn about 30+ supported annotation formats

API reference

Explore the full API, SDK, and CLI documentation

Join our Discord community to ask questions, share tips, and connect with other CVAT users!