Semantic Image Segmentation with Deep Learning: Conditional Random Fields as Recurrent Neural Networks

报告类型:研究组巡讲
报告嘉宾:Philip Torr(牛津大学  University of Oxford)
主持人:程明明(南开大学)
报告题目:Semantic Image Segmentation with Deep Learning: Conditional Random Fields as Recurrent Neural Networks [PPT & Video]
报告时间:2016年2月24日,19:00 (为了考虑英国时间,比平常时间提前一个小时) 
报告简介:

Image understanding requires not only object recognition, but also object delineation. This shape recovery task is challenging because of two reasons. First, the necessity of learning a good representation of the visual inputs. Second, the need to account for contextual information across the image, such as edges and appearance consistency. Deep Convolutional Neural Networks (CNNs) are successful at the former, but have limited capacity to delineate visual objects. We will present a framework that extends the capabilities of deep learning techniques to tackle this issue, obtaining cutting edge results in semantic image segmentation (i.e. detecting and delineating objects).

A live demo of the system we will be presenting is available at: http://crfasrnn.torr.vision.

.
报告人简介:Philip H. S. Torr received the PhD degree from Oxford University. After working for another three years at Oxford, he worked for six years as a research scientist for Microsoft Research, first in Redmond, then in Cambridge, founding the vision side of the Machine Learning and Perception Group. He is now a professor at Oxford University. He has won awards from several top vision conferences, including ICCV, CVPR, ECCV, NIPS and BMVC.
He is a Royal Society Wolfson Research Merit Award holder.

(Visited 1,095 times, 1 visits today)