Seminar: Seeing the Unseen: "Invisible" Object Reconstruction with Computational Imaging Approaches

Jinwei Ye headshot



Jinwei Ye

Assistant Professor, LSU Division of Computer Science and Engineering

Friday March 20, 2020

11:00 am

Location: online-only


Specular or transparent objects are considered “invisible” as they borrow appearances from nearby objects. The problem of reconstructing and modeling the “invisible” object has attracted much attention in recent years. Successful solutions can benefit numerous applications in oceanology, fluid mechanism and computer graphics as well as lead to new insights towards 3D scanning. The problem, however, is inherently difficult for several reasons. First, these objects do not have their own image. Second, determining the light paths within these objects for shape reconstruction is non-trivial since reflections or refractions are highly non-linear. Finally, transparent dynamics (such as fluid) even exhibit temporally varying distortions. In this talk, I present several computational imaging techniques for 3D reconstruction of specular or transparent object First, I present a novel polarization-encoded imaging system for mirror surface reconstruction. Our system uses a specialized liquid crystal display to modulate the polarization state of illumination rays. Through polarization analysis, we able to determine the reflection light paths and recover the 3D surface by ray-ray triangulation. Real experiments show that our approach is able to recover specular surfaces with high accuracy. Next, I present a learning-based approach for reconstructing the dynamic fluid surfaces. We design a deep neural network that takes a single refraction image as input and generalize distortion features for 3D fluid surface reconstruction. Finally, I present a PIV-based solution for 3D fluid flow reconstruction. Specifically, we use light filed imaging to resolve particles injected into the fluid body to estimate dense fluid motion field.


Jinwei Ye is an Assistant Professor in the Division of Computer Science and Engineering at Louisiana State University. 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. Before joining LSU, she was a senior scientist at the Innovation Center of Canon USA from 2015 to 2017 and a postdoc at the US Army Research Laboratory from 2014 to 2015. She leads the Imaging and Vision (IV) lab at LSU. Her research interests lie in the areas of computer vision, computer graphics and computational photography. She receives the LSU LIFT2 award and the NSF CISE CRII award. She will serve as an area chair for ICPR 2020 and CVPR 2021, and a local chair for CVPR 2022.