T
Tim Fingscheidt
Researcher at Braunschweig University of Technology
Publications - 250
Citations - 3048
Tim Fingscheidt is an academic researcher from Braunschweig University of Technology. The author has contributed to research in topics: Speech enhancement & Computer science. The author has an hindex of 22, co-authored 231 publications receiving 2140 citations. Previous affiliations of Tim Fingscheidt include Siemens & AT&T Labs.
Papers
More filters
Posted Content
GAN- vs. JPEG2000 Image Compression for Distributed Automotive Perception: Higher Peak SNR Does Not Mean Better Semantic Segmentation
TL;DR: It is concluded that distributed perception in future autonomous driving will most probably not provide a solution to the automotive bus capacity bottleneck by using standard compression schemes such as JPEG2000, but requires modern coding approaches, with the GAN encoder/decoder method being a promising candidate.
On Low-Bitrate Image Compression for Distributed Automotive Perception: Higher Peak SNR Does Not Mean Better Semantic Segmentation.
TL;DR: In this paper, the authors compare the image compression standards JPEG, JPEG2000, and WebP to a modern encoder/decoder image compression approach based on generative adversarial networks (GANs).
Proceedings ArticleDOI
An Unsupervised Temporal Consistency (TC) Loss to Improve the Performance of Semantic Segmentation Networks
Serin Varghese,Sharat Gujamagadi,Marvin Klingner,Nikhil Kapoor,Andreas Bär,Jan David Schneider,Kira Maag,Peter Schlicht,Fabian Hüger,Tim Fingscheidt +9 more
TL;DR: In this paper, an unsupervised temporal consistency (TC) loss is proposed to penalize unstable semantic segmentation predictions for improving the temporal consistency of DNNs for highly automated driving.
Journal ArticleDOI
DNN-Based Cepstral Excitation Manipulation for Speech Enhancement
Samy Elshamy,Tim Fingscheidt +1 more
TL;DR: The new approach exceeds the performance of a formerly introduced classical signal processing-based cepstral excitation manipulation (CEM) method in terms of noise attenuation by about 1.5 dB and shows that this gain also holds true when comparing serial combinations of envelope and excitation enhancement.
Posted Content
Self-Supervised Monocular Depth Estimation: Solving the Dynamic Object Problem by Semantic Guidance.
TL;DR: In this paper, a self-supervised semantically-guided depth estimation (SGDepth) method is proposed to deal with moving dynamic class (DC) objects, such as moving cars and pedestrians, which violate the static-world assumptions typically made during training of such models.