Visual Inference

Visual Inference

The Visual Inference group at TU Darmstadt, led by Prof. Stefan Roth, conducts research in several areas of computer vision with an emphasis on statistical methods and machine learning. We develop mathematical models and algorithms for analyzing and processing digital images with the computer. For example, we work on image restoration, image motion estimation, object recognition & tracking, and semantic scene understanding.

Moreover, we regularly offer courses, seminars and labs in computer science, particularly in computer vision and machine learning.

Recent Highlight

Obtaining ground truth data for computer vision using games

News & Events

  • CVRSUAD 2016 Best Paper Award

    Junhwa Hur and Stefan Roth have won the Best Paper Award at the Workshop on Computer Vision for Road Scene Understanding and Autonomous Driving (jointly with ECCV 2016) for the paper “Joint optical flow and temporally consistent semantic segmentation”.

  • Cover story in MIT Technology Review

    September 12, 2016

    Our research on obtaining training data for computer vision using games has been featured as a cover story in MIT Technology Review.

  • “Playing for Data”

    We have released a new dataset for sematic segmentation that leverages computer games. Please also see the paper and the video above.

  • IV 2016 Best Paper Award

    Marius Cordts, Timo Rehfeld and Stefan Roth have won the Best Paper Award (First Prize) at the IEEE Intelligent Vehicles Symposium 2016 for the paper “Semantic stixels: Depth is not enough”.

  • MOT16 Challenge

    The 2016 challenge of the Multiple Object Tracking Benchmark (MOT16) has been released. Please visit for data, instructions & results.

  • Cityscapes Dataset Released

    The Cityscapes Dataset for sematic urban scene understanding has been officially released: Please visit for data, instructions & results

Recent Publications