High performance deep learning inference for video analytics

NVIDIA DeepStream simplifies the development of scalable intelligent video analytics (IVA) applications powered by deep learning for smart cities and hyperscale datacenters. It brings together NVIDIA TensorRT for inference, Video Codec SDK for transcode and all of the required pre-processing and data curation into a single optimized API. Developers can now use this to quickly build new applications to transform video into valuable insight.

Applications for the DeepStream SDK include image classification, scene understanding, video categorization, and content filtering.

Developers can use the SDK to efficiently process and understand video frames at scale to meet the most demanding throughput and latency requirements. Applications developed with the DeepStream SDK can then be deployed to the NVIDIA Tesla accelerated computing platform.

The SDK includes sample code and pre-trained deep learning models as examples to perform classification and object detection on video streams. For applications requiring high channel density, the SDK includes examples for smart decode.

Developers can download the DeepStream SDK today to get started quickly.


DeepStream SDK is now available on Jetson. Learn more

Key Features

  • Deploy widely-used neural network models such as GoogLeNet and ResNet for real-time video content analysis
  • Support for common video formats such as H.264, HEVC/H.265, MPEG-2, MPEG-4, VP9, and VC1
  • Integrate into existing workflows with a simple C++ API
  • Deploy neural networks in full (FP32) or optimized precision (INT8, FP16)
  • Optimized for NVIDIA® Tesla® P4 GPU accelerators