Institute of Information Science, Academia Sinica

Abstract: Recognizing object classes and their 3D viewpoints is an important problem in computer vision. Based on a part-based probabilistic representation of our CVPR 2009 work, we propose a new 3D object class model that is capable of recognizing unseen views by pose estimation and synthesis. We achieve this by using a dense, multiview representation of the viewing sphere parametrized by a triangular mesh of viewpoints. The model is trained incrementally using Internet images labelled with object bounding boxes and categorical labels. Our method achieves good result on the Savarese et al. 2007 and PASCAL datasets in object detection. We further explore the possibility of recognizing 3D objects on a robotic platform. We use the depth sensor on the robot to train a 3d-encoded object model and show that detection performance improves even 3d information is given only in training. Short Bio: Min Sun graduated from National Chiao Tung University in 2003 with an Electrical Engineering degree. He received a MS degree from Stanford University in Electrical Engineering department in 2007. He was a member of Vision Lab at the Princeton University from 2008 to 2009. Now he is a PhD student in the Vision Lab at the University of Michigan at Ann Arbor. His research interests include object recognition, image understanding, and machine learning. He was a recipient of W. Michael Blumenthal Family Fund Fellowship and has won the best paper awards in 3DRR of ICC`'07.