GTC ON-DEMAND

 
SEARCH SESSIONS
SEARCH SESSIONS

Search All
 
Refine Results:
 
Year(s)

SOCIAL MEDIA

EMAIL SUBSCRIPTION

 
 

GTC ON-DEMAND

Cyber Security
Presentation
Media
Learning Semantic Video Captioning Using Data Generated with Grand Theft Auto
Abstract:
Learn an effective strategy for training deep video captioning and semantic search of overhead scenes, taking advantage of the photo-realism and ubiquitous access to high-level semantic information in a game like Grand Theft Auto. We train video captioning models and are surprised by how well their performance transfers to real drone, security camera, and even infrared footage in the face of poor video quality, compression artifacts, occlusion, and clutter. The speed of NVIDIA GPUs makes it possible to run the game engine with computationally heavy realism-augmenting mods, train video captioning models quickly, run captioning models faster than real-time video, and leverage commercially trained image models to quickly perform semantic search-by-example over large video datasets at scale.
 
Topics:
Cyber Security, Intelligent Video Analytics, Computer Vision
Type:
Talk
Event:
GTC Washington D.C.
Year:
2017
Session ID:
DC7237
Download:
Share: