Jean Song

Papers

Total Citations

H-Index

About

Jean Song is a researcher at the intersection of human-computer interaction, computer vision, and robotics, with a focus on scalable 3D perception and human-in-the-loop systems. Her work addresses a critical bottleneck in machine learning: the difficulty of acquiring accurate 3D object pose annotations from widely available 2D images. In her highly cited paper, *"C-Reference: Improving 2D to 3D Object Pose Estimation Accuracy via Crowdsourced Joint Object Estimation"* (2020, 6 citations), Song introduced a novel crowdsourcing framework that leverages multiple annotators to jointly estimate object poses, significantly improving the accuracy of 3D pose estimation from monocular RGB data. This work has direct implications for accelerating training in spatial reasoning domains, including assistive robotics, augmented reality, and autonomous vehicles. Building on this, her paper *"Human-in-the-loop Pose Estimation via Shared Autonomy"* (2021, 4 citations) explores how to balance human and robot control for dexterous manipulation tasks. Song’s key contribution lies in designing shared autonomy systems that overcome the robot’s limited 6-degree-of-freedom perception, enabling more reliable and efficient human-robot collaboration. Her research is notable for its practical, data-driven approach to bridging the gap between 2D vision and 3D understanding, making her a rising voice in crowdsourced perception and interactive robotics.