Transformers for Efficient and High-Level Image and Video Understanding
MCML Authors
Rajat Koner
Dr.
* Former Member
Abstract
Rajat Koner
Dr.
* Former Member
Abstract
This dissertation develops efficient Transformer-based architectures for high-level image and video understanding, addressing the computational challenges of attention while adapting Transformers to complex vision tasks. It introduces novel methods for scene graph generation, visual question answering, out-of-distribution detection, and video instance segmentation, achieving state-of-the-art performance with improved efficiency and interpretability. (Shortened.)
BibTeXKey: Kon25a