Mark Hamilton shares a novel deep learning approach for creating a robust object detection network for use in an infrared, UAV-based poacher recognition system. The approach uses Microsoft AirSim to generate thousands of hours of simulated drone footage in the African Savannah and then uses deep domain adaptation to translate the simulation into a form that is adversarially indistinguishable from real infrared drone footage. This yields a programmable data generator that can be used to dramatically improve the accuracy of algorithms without requiring expensive human curated annotations. Mark also explains how this work has been extended to contribute a photorealism extension to AirSim, automating much of the domain-specific expertise needed for computer graphics work and enabling the generation of limitless quantities of photorealistic data for use in reinforcement learning and autonomous vehicles.
This session is sponsored by Microsoft.
Mark Hamilton is a software engineer in Microsoft’s Azure Machine Learning Group in Cambridge, MA. He works on integrating the deep learning framework CNTK with the distributed computing framework Spark. Mark studied physics, mathematics, and automated theorem proving at Yale University. His current academic research focuses mainly on deep learning, unsupervised learning, and NLP.
Anand Raman is the chief of staff for the AI CTO office at Microsoft. Previously, he was the chief of staff for the Microsoft Azure Data Group, covering data platforms and machine learning, and ran the company’s product management and the development teams for Azure Data Services and the Visual Studio and Windows Server user experience teams; he also worked several years as researcher before joining Microsoft. Anand holds a PhD in computational fluid mechanics.
©2018, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • email@example.com