BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Probabilistic and Deep Models for 3D Reconstruction - Andreas Geig
 er\, Max Planck Institute for Intelligent Systems
DTSTART:20170926T120000Z
DTEND:20170926T130000Z
UID:TALK80361@talks.cam.ac.uk
CONTACT:Microsoft Research Cambridge Talks Admins
DESCRIPTION:3D reconstruction from multiple 2D images is an inherently ill
 -posed problem. Prior knowledge is required to resolve ambiguities and pro
 babilistic models are desirable to capture the ambiguities in the reconstr
 ucted model. In this talk\, I will present two recent results tackling the
 se two aspects. First\, I will introduce a probabilistic framework for vol
 umetric 3D reconstruction where the reconstruction problem is cast as infe
 rence in a Markov random field using ray potentials. Our main contribution
  is a discrete-continuous inference algorithm which computes marginal dist
 ributions of each voxel's occupancy and appearance. I will show that the p
 roposed algorithm allows for Bayes optimal predictions with respect to a n
 atural reconstruction loss. I will further demonstrate several extensions 
 which integrate non-local CAD priors into the reconstruction process. In t
 he second part of my talk\, I will present a novel framework for deep lear
 ning with 3D data called OctNet which enables 3D CNNs on high-dimensional 
 inputs. I will demonstrate the utility of the OctNet representation on sev
 eral 3D tasks including classification\, orientation estimation and point 
 cloud labeling. Finally\, I will present an extension of OctNet called Oct
 NetFusion which jointly predicts the space partitioning function with the 
 output representation\, resulting in an end-to-end trainable model for vol
 umetric depth map fusion.
LOCATION:Auditorium\, Microsoft Research Ltd\, 21 Station Road\, Cambridge
 \, CB1 2FB
END:VEVENT
END:VCALENDAR