Traditional neural radiance fields (NeRF) rely on multilayer perceptrons to implicitly query the color and density of each spatial point, requiring dense sampling along rays during rendering, which is computationally expensive and whose representation is not explicit.