Adarsh Kowdle

Cornell University, Google (United States)

Papers

Total Citations

114

H-Index

About

Adarsh Kowdle is a leading researcher in computer vision, with a primary focus on holistic scene understanding and 3D perception. His most influential work introduces a groundbreaking framework for scene understanding that moves beyond isolated tasks. Kowdle pioneered the concept of feedback-enabled cascaded classification models, demonstrating how interrelated subtasks—such as scene categorization, depth estimation, and object detection—can mutually reinforce each other. This holistic approach, detailed in his highly cited 2011 paper (garnering over 100 combined citations), showed that feeding information from one classifier back into another significantly improves overall accuracy, setting a new standard for integrated scene analysis. More recently, Kowdle has advanced visual-inertial odometry by incorporating learned monocular depth priors, a contribution that enhances the robustness of initialization in complex environments. His work bridges classical geometric methods with modern deep learning, offering practical solutions for augmented reality and autonomous navigation. With a consistent record of high-impact publications, Kowdle’s research continues to shape how machines perceive and interpret the visual world, making him a key figure in the evolution of scene understanding.