Radish: Radar Implicit Shapes Via Extreme-Wideband Sub-Terahertz Synthetic Aperture Radar

We explore synthetic aperture radar (SAR) 3D imaging capabilities in the sub-THz band using a novel 110-260 GHz experimental testbed. We propose RADar Implicit SHapes (RADISH), a post-processing method that leverages the SAR’s millimeter-level imaging resolution to estimate an object’s 3D shape. RADISH first c onverts high-resolution SAR images to a detailed point cloud of a scanned object. The point cloud is then used to fit an implicit neural representation to the object’s surface by approximating the signed distance function of the scene. We experimentally validate RADISH’s ability to represent the salient geometric features of real-world 3D objects.