Large-Scale Learning for Video Understanding

Jia Deng, University of Michigan

Usage Details

Jia Deng, Jason Corso, Yu-Wei Chao, Alejandro Newell, Vikas Dhiman, Chenliang Xu, Parker Hill, Yixin Jin, Jiaxuan Wang, Dawei Yang, Weifeng Chen, Kaiyu Yang, Zhao Fu, Hei Law, Jonathan Stroud, Lanlan Liu, Zhefan Ye, Yunfan Liu, Dominic Calabrese, MeiXing Dong

Video understanding, endowing computers with the ability to interpret videos as humans do, is one of the fundamental challenges of computer vision and artificial intelligence. Video data is ubiquitous. Yet our ability to perform automated analysis on such data is still primitive. A major barrier is computational: processing videos is extremely computationally intensive. For computers to handle the complexity of video data and understand a large variety of objects, actions, and events, it is necessary to perform machine learning on massive datasets of videos. This project aims to use the Blue Waters system to tackle core problems in video understanding. The goals are (1) to build upon existing software to develop a general framework for scalable, distributed visual learning, (2) to explore new algorithms and models that better recognize objects, actions, and events in videos, and (3) to provide a set of pre-computed features and pre-trained models that can be readily re-used by the research community at large.

http://web.eecs.umich.edu/~jiadeng/