Data Science

Build Better IVA Applications for Edge Devices with NVIDIA DeepStream SDK on Jetson

Register Now for Early Access Program
NVIDIA’s DeepStream SDK on Jetson makes it easy for developers to create robust, complex intelligent video analytics (IVA) capabilities for edge devices. Rapidly build end-to-end modular and scalable deep learning solutions — from intelligent cameras to appliances — with applications for smart cities, robotics, and industrial automation.
The SDK is built on top of JetPack, which includes L4T, multimedia APIs, CUDA, and TensorRT. DeepStream offers a rich collection of plugins and libraries for the GStreamer framework to build flexible applications that pull valuable insights from video.

Jetson software stack

Jetson’s unified memory architecture enables DeepStream to reduce management overhead and deliver a low-latency solution. With its advanced AI capabilities and a rich set of imaging and I/Os, developers can build highly integrated systems at the edge in small form factors for as little as 7.5W.
Easily Build Complex IVA Solutions for AI at the Edge
Creating IVA applications for complete scene understanding has traditionally been a challenging task. DeepStream, using GStreamer plugins, streamlines image capture, encode, decode, and inference using TensorRT, making the development of complex applications easier than ever.
AI-based intelligent video analytics application workflow

We’re providing developers with out-of-the-box samples that include an end-to-end AI-based video analytics application through multiple deep learning networks. The sample demonstrates the use of plugins for hardware-accelerated multimedia processing, color space conversion, tracking, and more. This also includes pre-trained DNNs to classify things such as cars and pedestrians along with extracting their attributes.
There’s also an application adaptation guide to modify the included samples and build custom plugins.
The DeepStream SDK on Jetson supports:

  • Video sources such as cameras, RTSP streams, and stored files from disks.
  • Widely used neural networks such as GoogLeNet and ResNet using TensorRT runtime.
  • Multiple output sinks that include rendering to display, metadata logging to file, and saving to disk.

A Comprehensive Edge-to-Cloud SDK
The DeepStream SDK on Jetson supports the Metropolis platform. This software package complements the existing DeepStream SDK on Tesla to enable developers to build applications from the edge to the cloud.
Sign up for the DeepStream on Jetson early access program today >>

Discuss (0)