tags: - CV-CUDA - Computer Vision - CV title: CV-CUDA: Basic Interoperability (C++)

Simple CV-CUDA application ¶

Authors: Holoscan Team (NVIDIA)
Supported platforms: x86_64, aarch64
Last modified: March 18, 2025
Language: C++
Latest version: 1.0
Minimum Holoscan SDK version: 0.6.0
Tested Holoscan SDK versions: 0.6.0
Contribution metric: Level 1 - Highly Reliable

This application demonstrates seamless interoperability between Holoscan tensors and CV-CUDA tensors. The image processing pipeline is just a simple flip of the video orientation.

Note that the C++ version of this application currently requires extra code to handle conversion back and forth between CV-CUDA and Holoscan tensor types. On the Python side, the conversion is trivial due to the support for the DLPack Python specification in both CV-CUDA and Holoscan. We provide two operators to handle the interoperability between CVCUDA and Holoscan tensors.

Using the docker file¶

This application requires a compiled version of CV-CUDA. For simplicity a DockerFile is available. To generate the container run:

./dev_container build --docker_file ./applications/cvcuda_basic/Dockerfile

The C++ version of the application can then be built by launching this container and using the provided run script.

./dev_container launch
./run build cvcuda_basic

Running the Application¶

This application uses the endoscopy dataset as an example. The run build command above will automatically download it. This application is then run inside the container.

./dev_container launch

The Python version of the simple CV-CUDA pipeline example can be run via

python applications/cvcuda_basic/python/cvcuda_basic.py --data=/workspace/holohub/data/endoscopy

or using the run script

./run launch cvcuda_basic python

The C++ version of the simple CV-CUDA pipeline example can then be run via

./build/applications/cvcuda_basic/cpp/cvcuda_basic --data=/workspace/holohub/data/endoscopy

or using the run script

./run launch cvcuda_basic cpp

Dev Container¶

To start the the Dev Container, run the following command from the root directory of Holohub:

./dev_container vscode cvcuda_basic

VS Code Launch Profiles¶

C++¶

Use the **(gdb) cvcuda_basic/cpp** launch profile configured for this application to debug the application.

Python¶

There are two launch profiles configured for this Python application:

(debugpy) cvcuda_basic/python: Launch cvcuda_basic using a launch profile that enables debugging of Python code.
(pythoncpp) cvcuda_basic/python: Launch cvcuda_basic using a launch profile that enables debugging of Python and C++ code.

Simple CV-CUDA application¶