Skip to main content
The CVAT Python SDK provides a comprehensive client library for interacting with CVAT servers. This page covers installation methods and available extras.

​
Basic Installation

Install the SDK using pip: 
pip install cvat-sdk
This installs the core SDK with the following dependencies:
  • attrs - For data classes
  • packaging - For version handling
  • Pillow - For image processing
  • platformdirs - For cross-platform cache directories
  • tqdm - For progress bars
  • typing_extensions - For enhanced type hints
  • python_dateutil - For date/time handling
  • urllib3 - For HTTP requests

​
Optional Dependencies

The SDK provides optional extras for additional functionality:

​
Masks Extra

For working with mask annotations, install the masks extra: 
pip install "cvat-sdk[masks]"
This adds:
  • numpy>=2 - For efficient mask data manipulation
Use this when working with segmentation tasks that use mask annotations.

​
PyTorch Extra

For PyTorch integration and built-in auto-annotation functions, install the pytorch extra: 
pip install "cvat-sdk[pytorch]"
This adds:
  • torch - PyTorch deep learning framework
  • torchvision - Computer vision models and utilities
  • scikit-image>=0.24 - Image processing
  • numpy>=2 - Array operations (via masks extra)
Use this when:
  • Using the PyTorch dataset adapters (TaskVisionDataset)
  • Training models with CVAT data
  • Using built-in auto-annotation functions

​
Installing Multiple Extras

You can install multiple extras at once: 
pip install "cvat-sdk[masks,pytorch]"

​
Requirements

  • Python: 3.10 or higher
  • Operating System: Cross-platform (Linux, macOS, Windows)

​
Development Installation

For development from source, see the SDK Developer Guide.

​
Verifying Installation

After installation, verify the SDK is available: 
import cvat_sdk
print(cvat_sdk.__version__)

​
Next Steps

Build docs developers (and LLMs) love

Get started for freeTalk to us