Neural approximations of scalar and vector fields, such as signed distance
functions and radiance fields, have emerged as accurate, high-quality
representations. State-of-the-art results are obtained by conditioning a neural
approximation with a lookup from trainable feature grids that