Kitware’s Senior Director of Computer Vision, Anthony Hoogs, Ph.D., was a member of the organizing panel for the National Academies of Sciences, Engineering, and Medicine Workshop on Challenges in Machine Generation of Analytic Products from Multi-Source Data. Sponsored by the Office of the Director of National Intelligence (ODNI), this invitation-only event was held in Washington, D.C. August 9-10. The workshop brought together internationally-known professors and researchers with government managers to discuss the potential impact and research directions for machine learning and artificial intelligence applied to intelligence analytics.
Dr. Hoogs was selected as a member of the 6-person organizing panel, which was responsible for creating the program and selecting the participants for this 2-day workshop. In addition, he was invited to speak on “Learning About Human Activities From Images and Videos.” He provided an overview of action recognition challenges and current methods, emphasizing that deep learning is playing a prominent role in video research as it explores temporal or sequential data. He recommended that the Intelligence Community and DoD should invest in the unique challenges they face that are not being addressed by industry, which is investing very heavily in video content extraction for consumer videos. The IC and DoD should also support open source frameworks and data sharing. As an example of this, he discussed the IARPA Deep Intermodal Video Analytics (DIVA) program, a current 5-year program focused on ground surveillance video analytics, on which Kitware is the test and evaluation support contractor. Check out his talk on Vimeo.
On the second day of this event, Anthony led the panel on the Evaluation of Machine-Generated Products, which included Jason Duncan (MITRE), Jonathan Fiscus (NIST), and Rob Fergus (NYU and Facebook). This panel discussion opened up with an overview by each participant, followed by details on how to successfully couple evaluation with annotation for maximum performance. Each individual described their own relevant experience in different methods to annotate and evaluate the massive amounts of data that are available to this community. Dr. Hoogs described Kitware’s work on DIVA, specifically collecting and annotating large amounts of surveillance video and developing an open source video analytics framework. Be sure to check out this panel in Vimeo.