Papers
5
Total Citations
464
H-Index
3
About
Marius Leordeanu is a computer vision researcher whose work spans human action recognition, semantic segmentation, depth estimation, and graph-based learning. He is perhaps best known for his 2013 paper introducing the **Moving Pose (MP)** descriptor, a fast and elegant 3D kinematics-based approach for low-latency human action recognition and detection, which has accumulated over 450 citations and remains a landmark contribution in the field. The method's simplicity and effectiveness made it particularly influential in applications such as human-robot interaction, gaming, and surveillance. Beyond action recognition, Leordeanu has pursued challenging problems at the intersection of deep learning and aerial imagery. His more recent work addresses semantic segmentation and metric depth estimation from drone-captured video, developing unsupervised and semi-supervised techniques that reduce the burden of costly pixel-level annotations. His 2022 UFO Depth method elegantly combines odometry-based analytical solutions with deep learning for metric depth estimation from UAV footage. Leordeanu has also explored theoretical foundations of computer vision through unsupervised graph and hypergraph matching. Across his career, his research reflects a consistent drive to build efficient, principled solutions to complex visual understanding problems, bridging theoretical rigor with real-world applicability.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Towards Automatic Annotation for Semantic Segmentation in Drone Videos3 citations · 2019
- 5Unsupervised Learning of Graph and Hypergraph Matching2 citations · 2020