RAPTR: Radar-based 3D Pose Estimation using Transformer

RAPTR, or RAdar Pose esTimation using tRansformer, introduces a two-stage pose decoder architecture that enhances pose queries with multi-view radar features. By utilizing only 3D bounding box and 2D keypoint labels, RAPTR simplifies the data collection process, which is particularly beneficial in cluttered indoor settings. The method's effectiveness is evidenced by its performance on two datasets, where it significantly outperformed existing techniques, reducing joint position errors by 34.3% on HIBER and 76.9% on MMVR. This advancement not only showcases the potential of radar technology in human pose estimation but also highlights the importance of developing methods that require less intensive data labeling, thus paving the way for broader applications in various fields such as robotics and augmented reality.