Dr. Stefan HINTERSTOISSER
stefan.hinterstoisser at gmail dot com
I'm an expert in ML (Deep Learning), Computer Vision and Software Engineering.
I'm currently working as a Team Lead / Manager (TLM) for X (formerly known as Google [X]) on Robotic Perception.
Current and Former Collaborators
(Universite Bordeaux) -
(TU Munich) -
Research / Publications
Synthetic Image Generation for Deep Learning
Transfer Learning for Object Detection
ECCVW'18 paper on transfer learning for object detection from synthetic data generated with CAD models.
Domain Adaptation for Deep Learning
ICRA'18 paper on multitask domain adaptation for deep learning of instance grasping from simulation.
Efficient Point Pair Features
ECCV'16 paper on efficient point pair features for detecting and estimating the 3D pose of CAD models.
Tracking Texture-less Objects
TVCG'13 paper on optimal local Searching for fast and robust textureless 3D object tracking in highly cluttered backgrounds.
Object Detection and 6D Pose Estimation
ACCV'12 paper on efficient template matching for detecting and estimating the 3D pose of CAD models.
The LINEMOD dataset can be found here.
Efficient Template Matching for Object Detection
Rapid Selection of Reliable Templates
CVPR'10 paper on rapid selection of reliable templates for visual tracking.
Fast Binary Template Matching for Object Detection
Real-Time Learning of Accurate Patch Rectification
Distance Transform Templates
CVPR'09 paper on distance transform templates for object detection and pose estimation.
Simultaneous Recognition and Homography Extraction of Local Patches
BMVC'08 paper on simultaneous recognition and homography extraction of local patches with a simple linear classifier.
Online Learning of Patch Perspective Rectification
CVPR'08 paper on online learning of patch perspective rectification.
Natural 3D Markers
ICCV'07 paper on N3M: natural 3D markers for real-time object detection and pose estimation.
Industrial Discrepancy Check
ISMAR'07 paper on industrial augmented reality solution for discrepancy check.
- Feb 2014 - now: working at Google (Google and X) on robotic perception
- Oct 2012 - Jan 2014: Industrial Perception Inc. (IPI - spin-off from WillowGarage): one of the first employees - acquired by Google
- Apr 2012 - Oct 2012: PostDoc at CAMP, TU Munich
- Jun 2010 - Jul 2010: internship at WillowGarage
- Aug 2007 - Mar 2012: PhD at CAMP, TU Munich (Prof. Nassir Navab) under the supervision of Prof. Vincent Lepetit, Dr. Selim Benhimane and Dr. Slobodan Ilic (passed with highest distinction - "summa cum laude")
- Mai 2007 - Jul 2007: 3 months research stay at EPFL at the lab of Prof. Pascal Fua
- Oct 2001 - Apr 2007: studied Informatik at TU Munich - finished with Diploma (equivalent to Master; main focus: Computer Vision; passed with high distinction - grade 1.2)
Theoretical Knowledge paired with Extensive Practical Experience
- Computer Vision: Object Detection, Pose Estimation, Pose Refinement, Tracking, Reconstruction, Feature Extraction, Description and Matching, Marker Detection, Outlier Rejection, Multiple View Geometry etc.
- Machine Learning: Deep Learning (e.g. Object Detection, Classification, GANs, Decoder / Encoder, Transfer Learning etc.), Decision Trees and Random Forests / Trees / Ferns
- Rasterization: efficient software rasterization
- Optimization: Mathematical / Hardware / Software
Programming Languages, Software and Build Systems
- I'm fluent in C++ (C++11, std, boost, tbb, Eigen, OpenCV, etc.) and thanks to ML I'm getting quite fluent in Python these days
- I know my tensorflow
- Experienced in SSE / AVX optimization
- Experienced in OpenGL and CUDA
- Experienced in GIT and other company based build systems
- Minor experiences in several other languages (Java, Matlab, Octave, Pascal).