Coordinate-based multi-layer perceptrons
Reinhard Heckel · 2025
Abstract Coordinate-based neural representations for images, 3D shapes, and other signals have recently emerged as an alternative to traditional discrete representations such as sparse representations and convolutional neural networks. In this chapter, this emerging class of signal models is discussed, its role for image reconstruction, and why multi-layer perceptrons with Fourier-feature inputs are good image models.