There’s a tremendous opportunity to bring efficiency in our cities, in retail operations, manufacturing lines, shipping and routing in warehouses. The groundwork has already been laid out with billions of sensors and cameras installed worldwide, that are rich sources of data.
Yet the ability to extract insights from this information has been challenging, and today’s solutions are siloed for specific platforms, making it difficult to deploy AI technology at scale.
The NVIDIA DeepStream SDK changes that. The 4.0 version is packed with powerful features that let developers do more in less time. Deploy seamless streaming pipeline with DeepStream using NVIDIA T4 servers or NVIDIA Jetson products including Jetson Nano.
DeepStream 4.0 delivers a unified code base for all NVIDIA GPUs, quick integration with IoT services, and container deployment, which dramatically enhances the delivery and maintenance of applications at scale. A unified codebase enables code portability, which provides developers the flexibility to build on a single platform and deploy on multiple platforms. The new communication plugins offer turnkey integration with Azure Edge IoT, MQTT and Kafka message brokers, enabling developers to build applications and services to leverage the power of the cloud. A complete list of supported plug-ins can be found here. These features make it easier than ever for developers to create smarter applications for retail, optical inspection, parking, traffic management and more.
A new reference tracker design offers robustness for object tracking and is GPU-accelerated for greater accuracy and robustness. There’s also added support for multiple heterogeneous camera inputs and camera types within a single application — which are important for robotics and drones.
Bringing Streaming Analytics to Multiple Industries
DeepStream SDK 4.0 is purpose-built to enable the development of AI applications, including:
- Smart retail — support for segmentation and multi-object tracking to build end-to-end applications, which can generate better customer insights such as heat maps, create automated checkout systems, improve loss prevention, and more.
- Industrial inspection — hardware-accelerated JPEG decode and encode along with networks such as YOLO and U-Net to create applications that can automatically inspect manufacturing defects at a rate faster than manual detection.
- Smart transportation — IoT integration to transmit sensor data for smart parking and traffic applications to improve congestion, optimize parking experiences for drivers and provide occupancy statistics.
- Logistics and operations — multiple cameras along with new network topologies running the latest version of TensorRT can be used to create applications to sort and direct packages in warehouses.
Members of the NVIDIA Developer Program can get free access to download DeepStream 4.0. DeepStream 4.0 is also available as a container image from the NGC registry for GPU-optimized deep learning frameworks, machine learning algorithms, and pre-trained AI models for smart cities. If you plan on running DeepStream in Docker or on top of Kubernetes, NGC provides the simplest deployment alternative.
Sign up for our upcoming technical webinar with Q&A session:
DeepStream SDK- Accelerating Real-Time AI Based Video And Image Analytics