Segmentation-Aware Convolutional Nets

Thursday April 14th, 2016 at 2:15 p.m. Adam Harley will be presenting “Segmentation-Aware Convolutional Nets”.

Speaker: Adam Harley

Day & Time: Thursday, April 14th, 2016
2:15 p.m. – 3:15 p.m.

Location: Room ENG 288
Computer Science Department
George Vari Centre for Computing and Engineering
Ryerson University
245 Church St., Toronto, ON, M5B 2K3

Organizer: IEEE Magnetics Chapter, IEEE Instrumentation & Measurement Joint Chapter and Computer Science Department Ryerson University

Contact: Dr. Maryam Davoudpour

Abstract: In this talk, I will propose a new deep convolutional neural network (DCNN) architecture that learns pixel embeddings, such that pairwise distances between the embeddings can be used to infer whether or not the pixels lie on the same region. Experimental results show that when this embedding network is used in conjunction with a DCNN trained on semantic segmentation, there is a systematic improvement in per-pixel classification accuracy. The contributions of this work consist in straightforward modifications to convolution routines. As such, they can be exploited for any task involving convolution layers, including object recognition, image retrieval, and video understanding.

Biography: Adam Harley received a BA (Honours) degree in psychology from Ryerson University in 2012, and was awarded the Canadian Psychological Association’s Certificate of Academic Excellence for his undergraduate thesis. Subsequently he began a computer science undergraduate degree at Ryerson, where he was awarded the NSERC USRA. In 2014 he joined Ryerson’s MSc program in computer science. During the MSc he did research at INRIA in France, as part of a Mitacs-Globalink research award. He is a recipient of the Queen Elizabeth II Graduate Scholarship for 2015. His main areas of research interest are computer vision and artificial intelligence.