Dive deep into NVIDIA's DeepStream SDK and its multi-object tracking capabilities in this 28-minute technical video. Explore the fundamentals of the new tracker unified architecture, including how to select from three object tracker alternatives (NvDCF, DeepSORT, or IOU) or integrate your own tracker for vision AI app development. Gain insights into the tracker's state machine and learn which parameters can be configured to optimize performance for specific applications. Examine NVIDIA's state-of-the-art NvDCF tracker, comparing different configuration parameters and understanding the results and tradeoffs of various optimization strategies. Learn how to handle errors from detection and explore visual tracking in NvDCF. Conclude with a case study on TrafficCamNet, applying the concepts to real-world scenarios. Access additional resources for DeepStream tracker comparisons, parameter tuning, plugin documentation, and getting started guides to further enhance your understanding and implementation of multi-object tracking in AI-based multi-sensor processing applications.
NVIDIA DeepStream Technical Deep Dive - Multi-Object Tracker
Overview
Syllabus
Intro
DEEPSTREAM SDK
MULTI-OBJECT TRACKERS IN DEEPSTREAM
MULTI-OBJECT TRACKING IN A NUTSHELL
HANDLING ERRORS FROM DETECTION
VISUAL TRACKING IN NVDCF
CASE STUDY: TRAFFICCAMNET
Taught by
NVIDIA Developer
Tags
Related Courses
Reviews
3.5 rating, based on 2 Class Central reviews
-
Rinu RejiThe good stuff first - this course really delivers on the "deep dive" promise. The instructor breaks down the tracking pipeline in a way that finally made several concepts click for me, especially around how the NvDCF tracker actually works under the hood. The sections on parameter tuning were particularly valuable - I was able to immediately apply these insights to optimize our production system's tracking performance. -
Alex BarnadasGood video gathering, lack of exam or exercises or any original content. But the information in the videos is good, comes from Nvidia YT channel.
