Chethan M. Parameshwara

Papers

Total Citations

H-Index

About

Chethan M. Parameshwara is a researcher at the forefront of computer vision and embodied AI, specializing in unsupervised learning from unconventional visual sensors and zero-shot object understanding. His pioneering work on event-based vision, particularly in his highly cited 2018 paper, introduced a lightweight, unsupervised learning pipeline that simultaneously estimates dense optical flow, depth, and egomotion from the sparse output of a Dynamic Vision Sensor (DVS). By developing a novel encoder-decoder neural network architecture (ECN), Parameshwara demonstrated that rich 3D scene understanding can be achieved without labeled data, a breakthrough for low-power, high-speed robotics applications. This work has garnered 29 citations, establishing a foundation for self-supervised learning with neuromorphic sensors. More recently, Parameshwara’s NudgeSeg (2021) tackles the challenge of zero-shot object segmentation by leveraging repeated physical interaction—a robot actively nudging objects to discover and segment them without any prior training on specific classes. This innovative, interaction-driven approach, with 7 citations, pushes beyond the limitations of static deep learning models, enabling robots to generalize to entirely novel environments. Parameshwara’s contributions bridge efficient visual perception and interactive learning, making him a key figure in advancing autonomous systems that can see and understand the world with minimal supervision.