[PAST EVENT] Human and machine inference of material properties: cloth and translucent materials

November 30, 2018
3pm - 4pm
McGlothlin-Street Hall, Room 020
251 Jamestown Rd
Williamsburg, VA 23185Map this location


Bei Xiao, American University


Human and machine inference of material properties: cloth and translucent materials


My talk mainly focuses on understanding perceptual inference of two types of common materials: cloth and translucent objects. First, I will discuss how humans estimate mechanical properties of fabrics in dynamic scenes. I will discuss the effect of spatiotemporal information on estimating the bending stiffness. I will present both human psychophysics results as well as machine learning algorithm on how dynamic information could be used to estimate stiffness of cloth. Second, I will discuss recent progress on the perception of translucent materials (e.g. skin, wax, jade). Physics-based rendering has developed sophisticated descriptions of material properties such as the bidirectional reflectance distribution function (BRDF) to describe surface reflectance. The parameter space of such physical models is very large but humans have a much more reduced representation of parameter space describing material appearances. Understanding perceptual representation is helpful in graphical simulation of novel materials, reducing computational cost, predicting contextual effects, and establishing computational models of material inference. I will present recent results regarding the perceptual dimensions of sub-surface scattering parameters as well as the interaction between lighting and 3D geometry on perception of translucent materials.


Bei Xiao is an Assistant Professor in the Department of Computer Science at American University. She received her Ph.D. from the University of Pennsylvania in Neuroscience and postdoctoral training at Computer Science and Artificial Intelligence Laboratory (CSAL) at MIT. Her research centers on human color perception, material perception, computer vision, perception-driven graphics, tactile perception, and virtual reality. Recently, she has been working on human and machine inference of material properties of deformable objects in dynamic scenes and multisensory rendering and perception of material properties in VR.


Pieter Peers