Faculty Candidate Seminar: Multi-perspective computational
imaging and its applications in 3D reconstruction
February 3, 2017
Dr. Jinwei Ye, Senior Scientist, Canon USA. Inc.
January 30, 2017 3:00 pm to 4:00 pm, 145/149 EE Building
Refreshments starting at 2:45 pm
Traditional single perspective camera has long served as the workhorse in compute vision and graphics.
However, the essential depth information of a scene is discarded in its image due to perspective flattening.
In contrast, a multi-perspective camera combines light rays collected from different viewpoints. Such
capability can benefit a broad class of applications, ranging from scene understanding to 3D
reconstruction. In this talk, I will present the designing, modeling, and constructing of multi-perspective
computational imaging systems and showcase their applications in depth recovery and 3D reconstruction.
First, I will show a special type of multi-perspective camera, the crossed-slit camera, and demonstrate its
advantages in acquiring 3D information. Then, I will present several multi-perspective illumination systems
and show their applications in recovering challenging invisible objects (e.g., mirror-like objects, transparent
fluids and gas flows). Finally, I will end my talk with a brief discussion on key challenges and future
opportunities in multi-perspective computational imaging.
Dr. Jinwei Ye is a senior scientist in the Innovation Center of Canon, USA. Prior to that, she was a post-
doctoral researcher in the US Army Research Laboratory (ARL). She received her Ph.D. in Computer
Science from the University of Delaware in 2014 and B.E. in Electrical Engineering from Huazhong
University of Science and Technology, China in 2009. Dr. Ye's research spans the fields of computer vision,
computational photography and computer graphics. In particular, she is interested in the co-design of
cameras, illuminations and computational algorithms for efficient 3D visual information acquisition and
processing. Jinwei's works have been published in premier conferences such as CVPR, ECCV and ICCV.