Margret Keuper

University of Mannheim, University of Siegen

Papers

Total Citations

137

H-Index

About

Margret Keuper is a computer vision researcher whose work spans semantic scene understanding, depth-integrated perception, and human action recognition. She is perhaps best known for her contributions to RGBD semantic segmentation, most notably the STD2P framework, which introduced a superpixel-based multi-view convolutional neural network capable of leveraging spatio-temporal information across multiple views of a scene. This approach proved particularly effective for complex indoor video environments and has garnered over 130 citations, establishing it as a meaningful reference point in the field of depth-aware scene segmentation. Her research demonstrates a consistent interest in enriching visual representations through spatial and temporal context, pushing neural networks to exploit structural cues that go beyond single-frame RGB input. More recently, Keuper has explored skeleton-based human action recognition, investigating the role of local spherical harmonics as geometric descriptors to improve classification of fine-grained hand gestures. This breadth — from dense scene labeling to articulated body understanding — reflects her broader commitment to robust, structured visual perception. Her work is relevant to researchers working at the intersection of 3D computer vision, deep learning, and video understanding.