Getting Started

Welcome to CV-CUDA! This guide will help you get up and running with GPU-accelerated computer vision.

What is CV-CUDA?

CV-CUDA is a library of GPU-accelerated computer vision operators optimized for AI workflows. It enables you to:

  • Accelerate image pre- and post-processing for CV AI models on NVIDIA GPUs

  • Keep data on GPU throughout your entire pipeline (zero CPU-GPU copies)

  • Batch operations efficiently for maximum throughput

  • Integrate seamlessly with PyTorch, TensorRT, and other GPU libraries

Prerequisites

Before diving into CV-CUDA, make sure you have the necessary hardware and software.

See the Prerequisites Information for complete hardware and software requirements.

Quick Start (5 Minutes)

1. Install Dependencies

For CUDA 12:

python3 -m venv venv_samples
source venv_samples/bin/activate
python3 -m pip install -r samples/requirements_hello_world_cu12.txt

For CUDA 13:

python3 -m venv venv_samples
source venv_samples/bin/activate
python3 -m pip install -r samples/requirements_hello_world_cu13.txt

This installs minimal dependencies (CV-CUDA, NumPy, nvImageCodec) needed for the hello_world sample.

2. Run Your First Sample

python3 samples/applications/hello_world.py

3. See Results

Check cvcuda/.cache/cat_hw.jpg - you just processed an image entirely on GPU!

Note

The requirements_hello_world_cu12.txt and requirements_hello_world_cu13.txt files are minimal (only 4 packages) for quick testing. For other samples (operators, applications, interoperability), use the full installation script:

cd samples
./install_samples_dependencies.sh

What’s Next? Continue below to learn the prerequisites and explore more samples.

Samples

The samples are the best way to learn CV-CUDA. They demonstrate everything from basic operations to complete deep learning pipelines.

What’s in the Samples:

  • Hello World - Your introduction to CV-CUDA (load, resize, blur, save)

  • Operators - Learn individual CV-CUDA operations (resize, blur, reformat, etc.)

  • Applications - Complete pipelines (classification, detection, segmentation)

View the Samples Documentation:

See the Samples Documentation for a guided tour of all available examples.

Interoperability

See the Interoperability for information on how to use CV-CUDA with other libraries.

Advanced Topics

Once you’re comfortable with the basics, explore advanced features:

Additional Resources

Need Help?