Visual Inference

Visual Inference

The Visual Inference Lab 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 semantic scene understanding, image motion estimation, deep learning, probabilistic methods, image restoration, and object tracking.

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

Recent Highlight

Darmstadt Noise Dataset: Ground truth data for image denoising with real image noise
Darmstadt Noise Dataset: Ground truth data for image denoising with real image noise

News & Events

  • PhD Defense

    October 17, 2017

    Marius Cordts has successfully defended his PhD thesis “Understanding Cityscapes: Efficient Urban Semantic Scene Understanding”. Congratulations!

  • PhD Defense

    September 26, 2017

    Timo Rehfeld has successfully defended his PhD thesis “Combining Appearance, Depth and Motion for Efficient Semantic Scene Understanding”. Congratulations!

Recent Publications

  • NIPS 2018

    T. Plötz and S. Roth, “Neural Nearest Neighbors Networks.” [preprint coming soon], [code coming soon]

  • ECCV 2018

    M. Jin, S. Roth, and P. Favaro, “Normalized Blind Deconvolution.” [open access]

  • CVPR 2018

    F. Saeedan, N. Weber, M. Goesele, and S. Roth, “Detail-preserving pooling in deep networks,” oral presentation. [open access] [code]

  • CVPR 2018

    T. Plötz, A. S. Wannenwetsch and S. Roth, “Stochastic variational inference with gradient linearization.” [open access]

  • CVPR 2018

    J. Gast and S. Roth, “Lightweight probabilistic deep networks.” [open access]

  • CVPR 2018

    S. R. Richter and S. Roth, “Matryoshka networks: Predicting 3D geometry via nested shape layers.” [open access]

  • AAAI 2018

    S. Meister, J. Hur, and S. Roth, “UnFlow: Unsupervised Learning of Optical Flow with a Bidirectional Census Loss.” [preprint] [code] [slides]