Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore the world of transparent open source AI video analytics in this 46-minute conference talk from linux.conf.au. Dive into the challenges of implementing AI-powered video analytics on edge devices and discover how the Panfrost open source driver for Mali GPUs is revolutionizing the field. Learn about building a pure open source stack that combines TensorFlow Lite and GStreamer to create a powerful AI solution for video analytics. Understand the importance of explainability in ML systems, particularly for applications impacting privacy like facial recognition. Follow the process of constructing an AI-driven multimedia pipeline using an entirely open source inference stack, including GPU driver, machine learning framework, and models. Gain insights into optimizing these models for resource-constrained hardware such as the Rockchip RK3399. Explore topics like object detection, AI on edge devices, explainable models, and layerwise relevance propagation. Discover how this open stack approach contributes to the development of ethical and trustworthy video analytics systems.
Syllabus
Intro
Welcome
Object Detection
AI on an Edge Device
TensorFlow Lite
Explainability
Why
Reason Codes
Lack of Open Source
Visualizing Data
Saliency Maps
Story
Explainable Models
Layerwise Relevance Propagation
Recommendations
Conclusion
Taught by
linux.conf.au