Get Started With the NVIDIA DeepStream SDK

Overview

NVIDIA DeepStream is a powerful SDK that lets you use GPU-accelerated technology to develop end-to-end vision AI pipelines. The latest release, DeepStream 7.0, is packed with innovative features to accelerate the development of your next-generation applications.

Release Highlights

Release notes

DeepStream 7.0 Highlights:

• Libraries that simplify development with GStreamer using intuitive Python APIs
• DeepStream Service Maker to easily build C++ object-oriented applications
• Enhancements to Single-View 3D Tracking (SV3DT) tracker
• BEVFusion for robust sensor fusion, including lidar and radar
• Support for Windows Subsystem for Linux (WSL2) to easily develop applications in a Windows environment
• PipeTuner 1.0 to optimize tuning of AI parameters and maximize accuracy
• Support for accelerating development of generative AI applications

GXF 4.0 includes Python and C++ and support for event-based scheduling. For full details, check the new NGC Collection page and the DeepStream 7 Release Notes.

Containers

DeepStream is available in three different flavors of containers:

• Triton: Single container for both x86 and Jetson. (Multi-arch)
• Samples: Single container for both x86 and Jetson.(Multi-arch)
• Development: x86-only container, including Graph Composer

All dockerfiles are also available on GitHub. For full details, check the new NGC Collection page.

Product Advisory

If you’re planning to bring models that use an older version of NVIDIA® TensorRT™ (8.6.1), make sure you regenerate the INT8 calibration cache before using them with the latest release of DeepStream.

You can find details about regenerating the cache in the Read Me First section of the documentation. For new DeepStream developers or those not reusing old models, this step can be omitted.