DART

Realistic radar simulation requires careful 3D modeling of a virtual world with accurate material properties and modeling all types of EM wave interactions with it. Previously we have seen explicit approaches for direct modeling or radar imaging followed by simulated view rendering. These explicit approaches are hard to design for practical, crowded scenes with heterogenous material composition and challenging multipath interaction. We propose an implicit neural rendering approach to produce accurate simulation of radar measurements from novel view points. As a by-product of our approach, we are also able to produce high quality radar images of scenes, akin to synthetic aperture imaging, but without any explicit modeling.

Full Paper | Code | Dataset | Video

CVPR Oral