Papers
1
Total Citations
3
H-Index
1
About
Keon-hee Kim is a researcher at the forefront of computer vision and deep learning, with a primary focus on advancing visual localization and scene understanding. Her most impactful work introduces a novel six-degree-of-freedom position regression convolutional neural network, built upon Google’s Inception-V4 architecture, which dramatically improves the accuracy of camera pose estimation—a critical capability for applications in robotics, augmented reality, and autonomous navigation. By achieving a 22% and 51% relative improvement over prior state-of-the-art methods, her 2017 paper, "Towards accurate kidnap resolution through deep learning," has garnered 3 citations and laid a foundation for more robust, real-world visual re-localization. Kim’s contributions are particularly notable for tackling the challenging "kidnapped robot" problem, where a system must recover its position without prior context. Her work demonstrates a keen ability to refine deep learning architectures for precise spatial reasoning, making her a rising voice in the field. For students and researchers exploring deep learning for geometric computer vision, Kim’s research offers a compelling example of how architectural innovations can yield tangible performance leaps.
Research Focus
Key Achievements
Top Papers
- 1Towards accurate kidnap resolution through deep learning3 citations · 2017