Holoscan Reference Applications

This is a central repository for the NVIDIA Holoscan AI sensor processing community to share reference applications, operators, tutorials and benchmarks. We invite users and developers of the Holoscan platform to reuse and contribute to this repository.

Table of Contents

• Overview
• Prerequisites
• Building
• Running
• Contributing
• Glossary
• More Information

Overview

This repository is a collection of applications and extensions created by the Holoscan AI sensor processing community. The following directories make up the core of this repo:

• Example applications: Visit applications to explore an evolving collection of example applications built on the NVIDIA Holoscan platform. Examples are available from NVIDIA, partners, and community collaborators.
• Community components: Visit operators and gxf_extensions to explore reusable Holoscan modules.
• Tutorials: Visit tutorials for extended walkthroughs and tips for the Holoscan platform.

Visit the Holoscan SDK User Guide to learn more about the NVIDIA Holoscan AI sensor processing platform. You can also chat with the Holoscan-GPT Large Language Model to learn about using Holoscan SDK, ask questions, and get code help. Holoscan-GPT requires an OpenAI account.

Prerequisites

Supported Platforms

You will need a platform supported by NVIDIA Holoscan SDK. Refer to the Holoscan SDK User Guide for the latest requirements. In general, Holoscan supported platforms include:

• An x64 PC with an Ubuntu operating system and an NVIDIA GPU; or
• A supported NVIDIA ARM development kit.

Individual examples and operators in this repo may have additional platform requirements. For instance, some examples may support only ARM platforms.

Build Environment

You may choose to build applications and operators in a containerized development environment or in your native environment.

We strongly recommend new users follow our Container Build instructions to set up a container for development.

If you prefer to build locally without docker, take a look at our Native Build instructions.

Once your development environment is configured you may move on to Building Applications.

Container Build (Recommended)

Software Prerequisites

To build and run in a containerized environment you will need:

• the NVIDIA Container Toolkit (v1.12.2 or later)
• Docker, including the buildx plugin (docker-buildx-plugin)
• git version control

You will also need to set up your NVIDIA NGC credentials at ngc.nvidia.com.

Fetch

Clone the repository to your local system:

$ git clone https://www.github.com/nvidia-holoscan/holohub.git
$ cd holohub

Alternatively, download sources as a ZIP archive from the GitHub homepage.

Build

Run the following command to build the default development container. The build may take a few minutes.

$ ./dev_container build

Check to verify that the image is created:

$ docker images
REPOSITORY      TAG               IMAGE ID       CREATED         SIZE
...
holohub         ngc-v2.1.0-dgpu   17e3aa51f129   13 days ago     13.2GB
...

Note: The development container script dev_container will by default detect if the system is using an iGPU (integrated GPU) or a dGPU (discrete GPU) and use NGC's Holoscan SDK container v2.1 for the Container build. See Advanced Container Build Options if you would like to use an older version of the SDK as a custom base image.

See the Developer Reference document for additional options.

Launch

Launch the Docker container environment:

$ ./dev_container launch

You are now ready to run applications! You may jump to the Running Applications section to get started.

Note The launch option will use the default development container built using Holoscan SDK's container from NGC for the local GPU. The script will also inspect for available video devices (V4L2, AJA capture boards, Deltacast capture boards) and the presence of Deltacast's Videomaster SDK and map it into the development container.

See also: Advanced Launch Options

Platform Notes (Container)

The development container has been tested on the following platforms:

• x86_x64 workstation with multiple RTX GPUs
• Clara AGX Dev Kit (dGPU mode)
• IGX Orin Dev Kit (dGPU and iGPU mode)
• AGX Orin Dev Kit (iGPU)

Notes for AGX Orin Dev Kit:

(1) On AGX Orin Dev Kit the launch script will add --privileged and --group-add video to the docker run command for the reference applications to work. Please also make sure that the current user is member of the group video.

(2) When building Holoscan SDK on AGX Orin Dev Kit from source please add the option --cudaarchs all to the ./run build command to include support for AGX Orin's iGPU.

Building Applications

Make sure you have installed the prerequisites before attempting to build reference applications.

Reference applications based on the Holoscan platform may be found under the applications directory. Reference applications aim to provide a demonstration of the SDK capabilities.

This repository provides a convenient run script to build and run applications in your development environment. To build an application:

  ./run build <application>

When the build is successful you can run applications.

Running Applications

To list all available applications you can run the following command:

  ./run list

Then you can run the application using the command:

  ./run launch <application>

For example, to run the tool tracking endoscopy application:

  ./run launch endoscopy_tool_tracking

Several applications are implemented in both C++ and Python programming languages. You can request a specific implementation as a trailing argument to the ./run launch command or omit the argument to use the default language. For instance, the following command will run the Python implementation of the tool tracking endoscopy application:

  ./run launch endoscopy_tool_tracking python

The run script reads the "run" command from the metadata.json file for a given application and runs from the "workdir" directory. Make sure you build the application (if applicable) before running it.

Cleanup

You can run the command below to reset your build directory:

./run clear_cache

In some cases you may also want to clear out datasets downloaded by applications to the data folder:

rm -rf ./data

Note that many applications supply custom container environments with build and runtime dependencies. Failing to clean the build cache between different applications may result in unexpected behavior where build tools or libraries appear to be broken or missing. Clearing the build cache is a good first check to address those issues.

Contributing

The goal of this repository is to allow engineering teams to easily contribute and share new functionalities and to demonstrate applications. Please review the Contributing Guidelines for more information.

Glossary

Many applications use the following keyword definitions in their README descriptions:

• <HOLOHUB_SOURCE_DIR> : Path to the source directory
• <HOLOHUB_BUILD_DIR> : Path to the build directory
• <HOLOSCAN_INSTALL_DIR> : Path to the installation directory of Holoscan SDK
• <DATA_DIR> : Path to the top level directory containing the datasets for the reference applications
• <MODEL_DIR> : Path to the directory containing the inference model(s)

More Information

Refer to additional documentation:

• Contributing Guide
• Developer Guide
• Release Discussion

You can find additional information on Holoscan SDK at:

• Holoscan GitHub organization
• Holoscan SDK repository
• Holoscan-GPT (requires an OpenAI account)
• Holoscan Support Forum