DINER: Disorder-Invariant Implicit Neural Representation

TL;DR

DINER introduces a hash-table augmented implicit neural representation that rearranges input coordinates to mitigate spectral bias, enhancing generalization and performance across diverse inverse signal reconstruction tasks.

Contribution

The paper proposes DINER, a novel disorder-invariant INR method using hash-tables to improve frequency modeling and generalization in implicit neural representations.

Findings

DINER outperforms state-of-the-art methods in quality and speed.

It generalizes well across different INR backbones and tasks.

Significantly alleviates spectral bias in neural representations.

Abstract

Implicit neural representation (INR) characterizes the attributes of a signal as a function of corresponding coordinates which emerges as a sharp weapon for solving inverse problems. However, the capacity of INR is limited by the spectral bias in the network training. In this paper, we find that such a frequency-related problem could be largely solved by re-arranging the coordinates of the input signal, for which we propose the disorder-invariant implicit neural representation (DINER) by augmenting a hash-table to a traditional INR backbone. Given discrete signals sharing the same histogram of attributes and different arrangement orders, the hash-table could project the coordinates into the same distribution for which the mapped signal can be better modeled using the subsequent INR network, leading to significantly alleviated spectral bias. Experiments not only reveal the generalization of the DINER for different INR backbones (MLP vs. SIREN) and various tasks (image/video representation, phase retrieval, and refractive index recovery) but also show the superiority over the state-of-the-art algorithms both in quality and speed.