Streamline Deep Learning for Video Analytics with DeepStream SDK 2.0

Discover how to leverage AI-based video analytics applications using DeepStream SDK 2.0 for Tesla to transform video into valuable insights for smart cities. Explore the modular plugin architecture and scalable framework for application development, featuring frequently used plugins for multi-stream decoding/encoding, scaling, color space conversion, and tracking. Dive into topics such as IVA in smart cities, DeepStream software stack, metadata structure, pipeline architecture, and NVIDIA-accelerated plugins. Learn about efficient memory management, reference applications for vehicle detection and tracking, configuration files, and performance best practices. Gain insights on implementing custom TensorRT plugin layers, integrating object detection models, and creating modular applications with DeepStream. Download the NVIDIA DeepStream SDK and accompanying webinar slides to start developing powerful video analytics solutions.

Intro
IVA IN SMART CITIES
DEEPSTREAM SOFTWARE STACK
DEEPSTREAM 2.0
DEEPSTREAM BUILDING BLOCK Gstreamer Plugin
METADATA IN DEEPSTREAM
METADATA STRUCTURE (1/2)
DEEPSTREAM PIPELINE ARCHITECTURE
NVIDIA-ACCELERATED PLUGINS Some of the most commonly used
DECODER PLUGIN Name nvdec_h264, nvdec_H265
MULTI-STREAM BATCHING
VIDEO AGGREGRATOR
INFERENCE Name nvinfer
FORMAT CONVERSION & SCALING PLUGIN Name nvvidconv
OBJECT TRACKER
SCREEN TILER
ON SCREEN DISPLAY
AND MANY MORE PLUGINS
MEMORY MANAGEMENT Efficient Memory Management
MEMORY MANAGEMENT GPU to CPU copy
DEEPSTREAM REFERENCE APPLICATION Vehicle detection, tracking, classification
NVINFER CONFIGURATION FILE
DEEPSTREAM SINGLE STREAM OUTPUT
DEEPSTREAM REFERENCE APP OUTPUT
DEEPSTREAM REFERENCE APPLICATION System Configuration & Performance for 25x 720p streams
KPIs FOR PERFORMANCE
MOTIVATION
METHODOLOGY FOR PERFORMANCE ANALYSIS
THROUGHPUT MEASUREMENT
LATENCY MEASUREMENT USING GST-LOGS
PERFORMANCE BEST PRACTICES
CUSTOM PLUGIN FOR OBJECT DETECTION Implement custom TensorRT plugin layers for your network topology Integrate your TensorRT based object detection model in DeepStream
YOLO V2 OVERVIEW
DARKNET TO TENSORRT
TENSORRT PLUGIN FACTORY
CREATING APPLICATIONS WITH DEEPSTREAM Object detection and counting
DEEPSTREAM MODULAR APPLICATION
START DEVELOPING WITH DEEPSTREAM