HOME              RESEARCH         CV         AWARDS         BLOG?         BOOKS        CONTACT

Research:

Interests:  Computer Vision (Semantics x Geometry), Machine Learning, Robotics

Current Projects:  I am currently working on perception for Microsoft HoloLens. In the recent past I built computer vision systems for self-driving cars. This included basic research on exploiting deep neural networks for 3D reconstruction problems, as well as systems work for an embedded ADAS pipeline that combines monocular visual odometry, GPS fusion, object detection and tracking, detailed 3D object localization, among other capabilities. A bit earlier, I worked on analysis and implementation of dense SLAM algorithms for embedded and many-core platforms, and on marrying object recognition with 3D reasoning.

Publications: Google Scholar profile, Semantic Scholar page

Learning Descriptors and Matching   Learning Keypoint Descriptors and Matching

   Hierarchical Metric Learning and Matching for 2D and 3D Geometric Correspondences
   Mohammed Fathy, Quoc-Huy Tran, M Zeeshan Zia, Paul Vernaza, Manmohan Chandraker
   arXiv March/2018
   (PDF)

Injecting Inductive Biases into Neural Networks   Data-Driven Inductive Biases into Neural Networks

   Deep Supervision with Intermediate Concepts
   Chi Li, M Zeeshan Zia, Quoc-Huy Tran, Xiang Yu, Gregory Hager, Manmohan Chandraker
   Submission to TPAMI 2018
   PDF, KITTI-3D (159 MB), Labeled CAD models (1 GB), Rendered Images (Chair+Sofa, 14 GB), Rendered Images (Cars)

Object Reconstruction   3D Object Geometry with deep CNNs

   Deep Supervision with Shape Concepts for Occlusion-Aware 3D Object Parsing
   Chi Li, M Zeeshan Zia, Quoc-Huy Tran, Xiang Yu, Gregory Hager, Manmohan Chandraker
   IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017
   PDF, Supplementary Material, KITTI-3D (159 MB), Labeled CAD models (1 GB), Rendered Images (Chair+Sofa, 14 GB), Rendered Images (Cars)

Vehicle Reconstruction   Simultaneous Localization and Mapping (SLAM) x 3D Object Localization

   Monocular Reconstruction of Vehicles: Combining SLAM with Shape Priors
   Falak Chhaya, Dinesh Reddy, Sarthak Upadhyay, Visesh Chari, M Zeeshan Zia, KM Krishna
   IEEE Conference on Robotics and Automation (ICRA), 2016
   (PDF, Project Page and Supplementary Material, Spotlight talk by Falak)

SLAMBench
Design Space Exploration using Active Learning
   Simultaneous Localization and Mapping (SLAM) Architectures

   Comparative Design Space Exploration of Dense and Semi-Dense SLAM
   M Zeeshan Zia, Luigi Nardi, Andrew Jack, Emanuele Vespa, Bruno Bodin et al.
   IEEE Conference on Robotics and Automation (ICRA), 2016
   (PDF, Slides (interactive), Slides (spotlight))

   Integrating Algorithmic Parameters into Benchmarking and Design Space Exploration in 3D Scene Understanding
   Brudo Bodin, Luigi Nardi, M Zeeshan Zia, Harry Wagstaff, Govind S Shenoy et al.
   International Conf. on Parallel Architecture and Compilation Techniques (PACT), 2016
   (PDF, Code)

   SLAMBench: A performance and accuracy benchmarking methodology for SLAM
   Luigi Nardi, Bruno Bodin, M Zeeshan Zia, John Mawer, Andrew Nisbet et al.
   IEEE Conference on Robotics and Automation (ICRA), 2015
   (PDF, Slides, Project Page / Code)

   In Media:
   Jetson Hacks
   HiPEAC Newsletter

Journal14
CVPR 14
CVPR 13
TPAMI 13
3dRR 11
   3D Object Localization and Scene Understanding

   Towards Scene Understanding with Detailed 3D Object Representations
   M Zeeshan Zia, Michael Stark, Konrad Schindler
   International Journal of Computer Vision (IJCV): vol. 112(2), p.188-203, 2015
   (PDF, Code, Evaluation Script and Test set)

   Are Cars Just 3D Boxes? - Jointly Estimating the 3D Shape of Multiple Objects
   M Zeeshan Zia, Michael Stark, Konrad Schindler
   IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014
   Our IJCV'15 paper contains expanded explanations and results for this work.
   (PDF, Poster, Code, Evaluation Script and Test set)

   Explicit Occlusion Modeling for 3D Object Class Representations
   M Zeeshan Zia, Michael Stark, Konrad Schindler
   IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013
   (PDF, Supplemental, Poster, Code, Training set~1.5 GB, Test set~260 MB)

   Detailed 3D Representations for Object Modeling and Recognition
   M Zeeshan Zia, Michael Stark, Bernt Schiele, Konrad Schindler
   IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI): vol. 35(11), p.2608-2623, 2013  
   (PDF, Code, Annotations)

   Revisiting 3D geometric models for accurate object shape and pose
   M Zeeshan Zia, Michael Stark, Bernt Schiele, Konrad Schindler
   IEEE Workshop on 3D Representation and Recognition (3dRR/ICCV WS), 2011
   Best Paper Award from Microsoft Research
   (PDF, Slides, Annotations)

   High-Resolution 3D Layout from a Single View
   M Zeeshan Zia
   PhD thesis, Swiss Federal Institute of Technology (ETH-Zurich), 2014
   (PDF, Slides)

Color classification   Color and Pattern Matching on Mobile

   Exemplar-Based Color Classification
   Zeeshan Zia, Emilio Maggio, Qi Pan, Michael Gervautz, Zsolt Szalavari
   US Patent Application 20,150,279,047
   (Link)

ICRA/ICAR 09   3D Object Detection

   3D Model Selection from an Internet Database for Robotic Vision
   Ulrich Klank, M Zeeshan Zia, Michael Beetz
   IEEE Conference on Robotics and Automation (ICRA), 2009
   (PDF)

   Acquisition of Dense 3D Model Database for Robotic Vision
   M Zeeshan Zia, Ulrich Klank, Michael Beetz
   The 14th International Conference on Advanced Robotics (ICAR), 2009
   (PDF, Slides)

MS Thesis 09   Dynamic 3D Scene Understanding and Action Recognition

   Inside-Out Activity Analysis using 3D Hand, Object, and Scene Tracking
   M Zeeshan Zia
   MS thesis, Electrical Engineering, Technical University of Munich, 2009
   (PDF)