Academics

Image Synthesis for Training Better Deep Detection and Recognition Models

Published:2019-11-22 

Speaker: Dr. Shijian

Time and Date: 11:30 - 13:00, November 25, 2019

Place: Room 530 of Scientific Building, Handan Campus, Fudan University

 

 

Abstract:

  The availability of a large amounts of annotated training images is critical to effective and efficient training of deep neural networks (DNN) for different computer vision problems. The current practice relies heavily on manual annotation of images which is usually expensive, time-consuming and difficult to scale across tasks and domains. This talk will share our recent works on realistic image synthesis which aims for generating annotated images automatically that are effective in training robust and accurate deep network models. Our approach is through image composition, and the target is to place foreground objects of interest into the background images as realistic as possible. Extensive evaluations on the challenging scene text detection and recognition tasks demonstrate the effectiveness of our composition based image synthesis approach.

 

Biography:

  Shijian received his PhD in electrical and computer engineering from the National University of Singapore. He is an Assistant Professor with School of Computer Science and Engineering, the Nanyang Technological University, Singapore. His major research interests include image and video analytics, visual intelligence, and machine learning. He has published more than 100 international refereed journal and conference papers and co-authored up to 10 patents in these research areas. He is currently an Associate Editor for the journal Pattern Recognition (PR). He has also served in the program committee of a number of international conferences, e.g. the Senior Program Committee of the International Joint Conferences on Artificial Intelligence (IJCAI) 2018 and 2019, the Senior Program Committee of the Association for the Advance of Artificial Intelligence (AAAI) 2019, the General Chair of the International Workshop on Document Analysis Systems (DAS) 2020, etc.

Copyrights 2017 © The School of Information Science and Technology, Fudan University