Thomas Brox (born 1976) is a German computer scientist specializing in computer vision and machine learning. He is a professor of pattern recognition and image processing at the University of Freiburg, where he heads the Computer Vision Group. Brox is known for co-authoring the U-Net architecture for biomedical image segmentation and FlowNet, an early convolutional neural network approach to optical-flow estimation. His publications have received more than 130,000 citations by 2026, according to Scopus.
Academic career
Brox received his doctorate in computer science from Saarland University in 2005. He then worked as a postdoctoral researcher at the University of Bonn and later at the University of California, Berkeley, in Jitendra Malik's computer vision group. He also headed the Intelligent Systems Group at TU Dresden as a temporary professor. In 2010, he joined the University of Freiburg, becoming professor of pattern recognition and image processing. He served as a programme chair for ECCV 2020 and became dean of the Faculty of Engineering on 1 July 2026.
Research on optical flow
An early focus of Brox's research was optical flow, the estimation of apparent motion between images. In 2004, Brox, Andrés Bruhn, Nils Papenberg, and Joachim Weickert proposed a variational method that combined brightness and gradient constancy with a discontinuity-preserving smoothness constraint and a nonlinear, multi-resolution optimization scheme. This work received the Longuet-Higgins Best Paper Award at ECCV 2004 and the Koenderink Prize in 2014.
In 2015, Brox co-authored FlowNet, which formulated optical-flow estimation as a supervised learning problem for convolutional neural networks, using entirely synthetically generated training data. This approach demonstrated the potential of deep learning for classical vision tasks.
U-Net and deep learning
Also in 2015, Brox co-authored U-Net with Olaf Ronneberger and Philipp Fischer. U-Net is a convolutional architecture for biomedical image segmentation that combines a contracting path with an expanding path, using skip connections to preserve high-resolution information from the encoder. The architecture became widely adopted in medical imaging and other segmentation tasks, influencing subsequent Neural network designs.
Brox's work sits at the intersection of Computer vision and Machine learning, contributing to the broader Deep learning movement. His research on visual representation learning has explored how Neural networks can learn useful features from images, often without explicit labels.
Awards and memberships
Brox received the Longuet-Higgins Best Paper Award at ECCV 2004 and the Koenderink Prize in 2014, both for the 2004 optical-flow paper with Bruhn, Papenberg, and Weickert. He has been a full member of the Heidelberg Academy of Sciences and Humanities since 2020.
References
- Source facts provided by Wikipedia (CC BY-SA).