Fig. 2: The neural level set representation model.
From: Shaping freeform nanophotonic devices with geometric neural parameterization

a Functional representation showing the multi-resolution hash encoding process. Input coordinates (x, y) are mapped through parallel hash functions at different resolution levels, each accessing a dedicated lookup table of trainable feature vectors. Features are interpolated based on the relative position within grid cells and concatenated. The resulting feature vector is processed by a multi-layer perceptron (MLP) to produce the final level set function value Fθ(x, y). b Level set operations showing the conversion from continuous functions to binary and grayscale structures through dense sampling and thresholding operations.