Computer Vision for Human-Computer Interaction Lab
The lab is directed by Prof. Dr. Rainer Stiefelhagen, who also supervises the KIT's Study Center for Visually Impaired Students (SZS). Together with the SZS, we develop new assistive technologies for visually impaired people. We also have a close collaboration with the Fraunhofer IOSB in Karlsruhe.
Our research focuses on the perception of people with applications in the following areas:
We are sorry to tell you that this year's ZEISS Computer Vision Hackathon - Next Gen Computer Vision which is co-organised by our team was cancelled due to the current health situation.
We were invited to present our paper by A. Roitberg written in collaboration with M. Martin from Fraunhofer IOSB features Drive&Act, the first large-scale dataset for fine-grained driver activity recognition.
At the 30th British Machine Vision Conference (BMVC) 2019 in Cardiff, our team won again the Best Industry Paper Award for its work on Image Translations with Spatial Profile Loss.
Authors: Vivek Sharma, Makarand Tapaswi, Saquib Sarfraz, and Rainer Stiefelhagen
Please be informed that our 'Computer Vision for Human Computer Interaction' lecture will not take place next winter semester. However, it will be resumed in WS 20/21.
The paper by M. Haurilet et al. presents a novel model based on a graph-traversal scheme for Visual Reasoning. The architecture searches relevant nodes in the scene graph to find information for answering the current question.
The paper by M.S. Sarfraz et al. introduces a highly-efficient approach for clustering using first neighbour relations. In comparison to other clustering algorithms, FINCH does not require any hyper-parameters, but is able to deduce the number of clusters automatically.
We will offer a new lecture 'Deep Learning for Computer Vision' from this summer semester on.
The paper by S. Sarfraz et al. presents a novel approach for person re-identification and an unsupervised re-ranking method for retrieval applications.
The paper by V. Sharma et al. presents a novel CNN architecture that can enhance image-specific details via dynamic enhancement filters with the overall all goal to improve classification.
We have moved back to
Monica Haurilet and Ziad Al-Halah participated in the textbook question answering challenge and won the first place on the text-based track and came second in the diagram-based track.
The winners were announced in CVPR17 Workshop for visual understanding across modalities - read more
Our new test lab for a barrier-free access to information for visually imparired persons was inaugurated on 3rd June 2016.
At the 26th British Machine Vision Conference (BMVC) 2015, our team received the best industry paper award for the work on thermal-visible face recognition
MIT Technology Review featured an article on our thermal visible face matching work in July 2015. Read more how 'Deep Neural Nets Can Now Recognize Your Face in Thermal Images'
Read more in 'In the Press'
We offer a new lecture 'Assistive Technologies for Visually Impaired Persons' from this semester on. Further details
At the 22nd International Conference on Pattern Recognition (ICPR) Prof. Stiefelhagen's team received the IBM Best Student Paper Award in the Track 'Pattern Recognition and Machine Learning' for the work on "High-Level Semantics in Transfer Metric Learning".Further details
Our research group receives a Google Research Award for its work on "A Mobility and Navigational Aid for Visually Impaired Persons". The "Google Faculty Research Award" is endowed with 83.000 USD for supporting research in computer science, engineering and related disciplines.
The video production team from the Department of Informatics visited our booth at the CeBIT 2013 and recorded a presentation of our demos there. You can watch it in their video channel KITInformatik.