Ali Osman Ulusoy

I am a senior scientist at Microsoft, where I work on computer vision.

At Microsoft I work on HoloLens and Azure Spatial Anchors. Before to joining Microsoft, I was a post-doctoral researcher at the Max-Planck Institute for Intelligent Systems, where I was advised by Michael J Black and Andreas Geiger. I did my Phd in electrical engineering with Joseph Mundy at Brown University.

Email  /  Google Scholar  /  LinkedIn  /  CV

profile photo

I'm interested in computer vision and machine learning. Most of my research is about inferring structure and motion from images. Representative papers are highlighted.

clean-usnob Superquadrics Revisited: Learning 3D Shape Parsing beyond Cuboids
Despoina Paschalidou, Ali Osman Ulusoy, Andreas Geiger
CVPR, 2019
video  /  code  /  blog post

CNNs can learn to parse a 3D model into a small set of Superquadrics.

clean-usnob RayNet: Learning Volumetric 3D Reconstruction with Ray Potentials
Despoina Paschalidou, Ali Osman Ulusoy, Carolin Schmitt, Andreas Geiger
CVPR, 2018   (Spotlight Presentation)
video  /  code  /  project page

RayNet integrates a CNN that learns view-invariant feature representations with a factor graph that properly models the geometry of perspective projection and occlusion.

clean-usnob OctNetFusion: Learning Depth Fusion from Data
Gernot Riegler, Ali Osman Ulusoy, Horst Bischof, Andreas Geiger
3DV, 2017   (Oral Presentation)
video  /  code

Learning depth-fusion is much more accurate than traditional TSDF fusion and TV-L1 fusion.

clean-usnob OctNet: Learning Deep 3D Representations at High Resolutions
Gernot Riegler, Ali Osman Ulusoy, Andreas Geiger
CVPR, 2017   (Oral Presentation)
video  /  code

We accelerate 3D convolutional neural networks using octrees.

clean-usnob Semantic Multi-view Stereo: Jointly Estimating Objects and Voxels
Ali Osman Ulusoy, Michael J Black, Andreas Geiger
CVPR, 2017

We extend our 3DV 2015 paper with object-level 3D shape priors. Inference in this factor graph yields dense 3D geometry in addition to the 6DOF pose of objects in it.

clean-usnob Compression of Probabilistic Volumetric Models using multi-resolution scene flow
Octavian Biris, Ali Osman Ulusoy, Joseph L. Mundy
Image and Vision Computing, 2017

We extend Horn and Schunck's optical flow formulation to 3D volumetric models and use it for compression.

clean-usnob Patches, Planes and Probabilities: A Non-local Prior for Volumetric 3D Reconstruction
Ali Osman Ulusoy, Michael J Black, Andreas Geiger
CVPR, 2016
video  /  code

We extend our 3DV 2015 paper with non-local planarity surface priors. Much of the paper is about efficient inference in the factor graph.

clean-usnob Towards Probabilistic Volumetric Reconstruction using Ray Potentials
Ali Osman Ulusoy, Andreas Geiger, Michael J Black
3DV, 2015   (Best Paper Award)
video  /  code

We formulate multi-view stereo as inference in a factor graph. This allows exposing the inherent ambiguities in surface reconstruction from images, and incorporating surface priors.

clean-usnob Image-based 4-d Reconstruction Using 3-d Change Detection
Ali Osman Ulusoy, Joseph L. Mundy
ECCV, 2014

We use a stream of images captured over time to continuously update a 3D model with changes in the environment.

clean-usnob Evaluation of feature-based 3-d registration of probabilistic volumetric scenes
Maria Restrepo, Ali Osman Ulusoy, Joseph L. Mundy
ISPRS Journal of Photogrammetry and Remote Sensing, 2014

We study the problem of registering volumetric 3D models, by evaluating sensitivity to discretization, camera registration errors, and changes in illumination.

clean-usnob Dynamic Probabilistic Volumetric Models
Ali Osman Ulusoy, Octavian Biris, Joseph L. Mundy
ICCV, 2013

We reconstruct arbitrary dynamic 3D scenes accurately using a view-dependent volumetric modeling, and efficiently by exploiting redundancies in space and time.

clean-usnob Characterization of 3-D Volumetric Probabilistic Scenes for Object Recognition
Maria Restrepo, Brandon Mayer, Ali Osman Ulusoy, Joseph L. Mundy
IEEE Journal of selected topics in Signal Processing, 2012

We solve 3D object recognition in the context of volumetric 3D models.

clean-usnob High Resolution Surface Reconstruction from Multi-view Aerial Imagery
Fatih Calakli, Ali Osman Ulusoy, Maria Restrepo, Gabriel Taubin, Joseph L. Mundy
3D Imaging Modeling Processing Visualization Transmission (3DIMPVT), 2012   (Oral Presentation)

We fuse images from different viewpoints into a volumetric probabilistic model and extract a textured mesh from it.

clean-usnob Robust one-shot 3D scanning using loopy belief propagation
Ali Osman Ulusoy, Fatih Calakli, Gabriel Taubin
CVPR workshop, 2010

We use factors graphs and belief propagation to solve 3D scanning using structured light.

clean-usnob One-shot scanning using de bruijn spaced grids
Ali Osman Ulusoy, Fatih Calakli, Gabriel Taubin
3DIM workshop at ICCV, 2009

We project a grid pattern with unique spacings onto an object to reconstruct its 3D structure.