Frequently Asked Questions

​

General Questions

​

What is CVAT?

​

Is CVAT free?

• Use cvat.ai for free with limitations (10 tasks, 500MB storage)
• Self-host CVAT for unlimited usage
• Subscribe to CVAT Cloud for additional features
• Get Enterprise support for self-hosted deployments

​

What annotation types does CVAT support?

• Bounding boxes (rectangles)
• Rotated bounding boxes
• Polygons
• Polylines
• Points
• Ellipses
• Cuboids (3D boxes)
• Skeletons (keypoints)
• Masks (segmentation)
• Tags (image-level labels)

​

What file formats does CVAT support?

• CVAT (XML)
• COCO (JSON)
• YOLO / Ultralytics YOLO
• Pascal VOC
• KITTI
• Cityscapes
• LabelMe
• Datumaro
• TFRecord
• And many more!

​

Installation

​

What are the system requirements?

• RAM: 8GB (16GB recommended)
• CPU: 4 cores (8 cores recommended)
• Storage: 20GB+ depending on data size
• Docker: 20.10.0+
• Docker Compose: 1.29.0+

• RAM: 32GB+
• CPU: 16+ cores
• Storage: SSD with sufficient space for your datasets

​

How do I install CVAT?

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

​

Can I use CVAT without Docker?

• CVAT has many dependencies (PostgreSQL, Redis, ClickHouse, etc.)
• Docker ensures consistent environments
• Manual setup is complex and error-prone

​

How do I update CVAT?

cd cvat
git pull origin develop
docker compose down
docker compose pull
docker compose up -d

​

Can I install CVAT on Windows?

• WSL2 (Windows Subsystem for Linux) - Recommended
• Docker Desktop for Windows

​

Usage

​

How do I create my first annotation task?

1. Go to Tasks → Create new task
2. Enter a task name
3. Add labels (e.g., “car”, “person”, “dog”)
4. Upload images or videos
5. Click Submit
6. Wait for processing to complete
7. Click Open to start annotating

​

What’s the difference between a Task, Job, and Project?

• Project: Container for related tasks with shared labels
• Task: A set of media files to annotate
• Job: A subset of a task’s frames, assigned to an annotator

​

How do I import existing annotations?

1. Open your task
2. Go to Actions → Upload annotations
3. Select format (COCO, YOLO, etc.)
4. Upload your annotation file
5. Click Submit

​

How do I export annotations?

1. Open your task
2. Go to Actions → Export task dataset
3. Select export format
4. Choose whether to save images
5. Click Export
6. Download when ready

​

Can I use CVAT for video annotation?

• Track objects across frames
• Interpolation between keyframes
• Video chapter navigation
• Frame-by-frame control
• High-performance frame caching

​

How does tracking work?

1. Create a Track annotation (not a shape)
2. Draw the object on the first frame
3. Navigate to a later frame
4. Adjust the annotation position
5. CVAT automatically interpolates between frames

​

Can multiple people annotate the same task?

1. When creating a task, set Job overlap and Segment size
2. CVAT creates multiple jobs automatically
3. Assign different jobs to different annotators
4. Each annotator works on their assigned frames

​

What are serverless functions?

• Auto-annotation: Automatic annotation using AI models
• Interactive annotation: AI-assisted annotation tools (e.g., Segment Anything)
• Tracking: Object tracking across frames

• Segment Anything (SAM)
• YOLOv7
• Mask R-CNN
• Custom models via Nuclio

​

Troubleshooting

​

CVAT is slow or unresponsive

• Insufficient RAM: Increase Docker memory limit
• Large images: Use smaller images or enable compression
• Too many objects: Split task into smaller jobs
• Slow storage: Use SSD instead of HDD

• Enable frame caching
• Use compressed chunks
• Reduce chunk size
• Close unused browser tabs

​

I can’t upload large videos

• Increase Traefik’s upload size limit
• Use cloud storage integration (AWS S3, Azure, Google Cloud)
• Split video into smaller segments
• Use mounted file shares

traefik:
  command:
    - --entrypoints.web.http.middlewares=cvat-request-body-limit

​

Annotations disappeared or weren’t saved

• Click Save (Ctrl+S) regularly
• Check browser console for errors
• Verify permissions (can you write to the job?)
• Check server logs: docker logs cvat_server

• Check Actions → Export job dataset to see if data exists
• Contact admin if data is critical

​

Docker containers fail to start

docker ps -a  # Check running containers
sudo lsof -i :8080  # Check port usage

• Increase Docker memory limit (Docker Desktop → Settings → Resources)
• Kill unnecessary processes

sudo usermod -aG docker $USER  # Add user to docker group
# Log out and back in

​

How do I check logs?

# All services
docker compose logs

# Specific service
docker logs cvat_server
docker logs cvat_worker_import

# Follow logs
docker logs -f cvat_server

​

Database is corrupted

docker compose down -v  # Remove volumes
docker compose up -d    # Recreate

docker exec cvat_db pg_dump -U root cvat > backup.sql

​

CVAT shows “502 Bad Gateway”

docker logs cvat_server  # Check for errors
docker compose restart cvat_server  # Restart backend

• Database is running: docker ps | grep db
• Redis is running: docker ps | grep redis
• Migrations completed: docker exec cvat_server python manage.py showmigrations

​

Features

​

Can I use CVAT offline?

• No internet connection required for annotation
• Internet needed for:
  • Auto-annotation (if using cloud models)
  • Cloud storage integration
  • Updates and new models

​

Does CVAT support 3D annotation?

• Point cloud annotation (.pcd files)
• 3D cuboids with rotation
• Multiple camera views
• Related images (synchronized 2D images)

​

Can I customize CVAT?

• UI plugins: Add custom UI components
• Server plugins: Extend backend functionality
• Custom annotation formats: Add import/export formats
• Serverless functions: Deploy custom AI models
• Webhooks: Integrate with external systems

​

Does CVAT have an API?

• REST API: Full-featured HTTP API
• Python SDK: pip install cvat-sdk
• CLI: pip install cvat-cli
• OpenAPI schema: https://app.cvat.ai/api/docs

​

Can I integrate CVAT with my ML pipeline?

• FiftyOne: Dataset management and visualization
• Roboflow: Model training and deployment
• HuggingFace: Model integration
• Custom scripts: Using SDK/CLI
• Webhooks: Real-time notifications

1. Train model
2. Use model for auto-annotation in CVAT
3. Manually correct predictions
4. Export improved dataset
5. Retrain model
6. Repeat

​

What about annotation quality control?

• Quality reports: Compare annotations between annotators
• Consensus: Merge annotations from multiple annotators
• Honeypot frames: Ground truth frames for validation
• Review mode: Accept/reject annotations with comments
• Inter-annotator agreement: Calculate agreement metrics

​

Does CVAT support SSO/LDAP?

• LDAP/Active Directory
• SAML 2.0
• OAuth 2.0

​

Can I run CVAT in production?

• Use Kubernetes/Helm for orchestration
• Set up SSL/TLS (HTTPS)
• Configure backups (database and media)
• Use cloud storage (S3, Azure, GCS)
• Set up monitoring (Prometheus, Grafana)
• Consider Enterprise support

​

Getting Help

​

Where can I get help?

• Documentation: https://docs.cvat.ai/
• GitHub Issues: https://github.com/cvat-ai/cvat/issues
• Gitter Chat: https://gitter.im/opencv-cvat/public
• Discord: https://discord.gg/fNR3eXfk6C
• Stack Overflow: Tag questions with #cvat
• Email: [email protected]

​

How do I report a bug?

1. Check if already reported: GitHub Issues
2. Create new issue with:
   • CVAT version
   • Steps to reproduce
   • Expected vs actual behavior
   • Screenshots/videos if applicable
   • Browser console errors
   • Server logs if relevant

​

How do I request a feature?

1. Check existing requests
2. Create new issue:
   • Describe the feature
   • Explain use case
   • Provide examples if possible
3. Discuss with maintainers
4. Consider contributing!

​

Can I contribute to CVAT?

• Code: Fix bugs, add features
• Documentation: Improve guides and docs
• Testing: Report bugs, test new features
• Community: Answer questions, help others

​

More Resources

• Website: https://www.cvat.ai/
• Documentation: https://docs.cvat.ai/
• GitHub: https://github.com/cvat-ai/cvat
• YouTube: https://www.youtube.com/@cvat-ai
• LinkedIn: https://www.linkedin.com/company/cvat-ai/
• Blog: https://www.cvat.ai/post

​

Still Have Questions?

• Check the full documentation
• Ask on Gitter or Discord
• Open a GitHub Issue
• Contact us at [email protected]