Implicit Neural Representations
Also known as INR
Overview
Implicit neural representations (INRs) learn continuous functions from coordinates to values, such as image colors, 3D geometry, motion, or physical measurements. Common approaches use coordinate-based MLPs with sinusoidal or Fourier features, multiresolution grids, tri-planes, or hybrid designs that combine neural decoders with explicit spatial structure. Researchers use them for reconstruction, compression, super-resolution, rendering, and scientific inverse problems, where continuous querying can handle irregular samples and varying resolutions. A central challenge is balancing detail and accuracy against fitting time, memory, and query cost; adaptive representations and learned priors help, but generalization remains task-dependent, and explicit methods can sometimes be more efficient or accurate.
Latest papers
204 papers19 papers in the last four weeks, down 32% on the four weeks before. 0.1% of all new papers.