Learning to understand dynamic scenes
Date:
Tuesday, December 12, 2017 15:00 - 16:00
Speaker:
Laura Leal-Taixe (TUM)
Location:
Mondi Seminar Room 2, Central Building
Series:
CS Talk Series
Host:
Christoph Lampert
Contact:
BONVENTRE-DARTHE Astrid
Dynamic scene understanding englobes many of the classic problems in Computer Vision. Semantically labelling every pixel in a video sequence is a big step towards understanding the world around us, and is key to applications such as autonomous driving or human-robot interaction. In this talk, I will present several works that bring us closer to solving the problem of dynamic scene understanding, focusing especially on the role of Deep Learning for video analysis. I will present our recent works on multiple object tracking and video object segmentation and discuss the importance of big data and our efforts towards building a benchmark for multiple object tracking.Finally, I will briefly discuss future research plans and the recently accepted project socialMaps.