Towards Croppable Implicit Neural Representations

TL;DR

This paper introduces Local-Global SIRENs, an INR architecture enabling easy cropping and editing of signals by removing parts without retraining, while also improving encoding efficiency and performance.

Contribution

The paper presents a novel INR architecture that inherently supports cropping and editing, a feature not available in previous models, and demonstrates its versatility and efficiency enhancements.

Findings

Supports effortless cropping by removing weights without retraining

Accelerates training and improves encoding of signals

Enhances downstream task performance

Abstract

Implicit Neural Representations (INRs) have peaked interest in recent years due to their ability to encode natural signals using neural networks. While INRs allow for useful applications such as interpolating new coordinates and signal compression, their black-box nature makes it difficult to modify them post-training. In this paper we explore the idea of editable INRs, and specifically focus on the widely used cropping operation. To this end, we present Local-Global SIRENs -- a novel INR architecture that supports cropping by design. Local-Global SIRENs are based on combining local and global feature extraction for signal encoding. What makes their design unique is the ability to effortlessly remove specific portions of an encoded signal, with a proportional weight decrease. This is achieved by eliminating the corresponding weights from the network, without the need for retraining. We further show how this architecture can be used to support the straightforward extension of previously encoded signals. Beyond signal editing, we examine how the Local-Global approach can accelerate training, enhance encoding of various signals, improve downstream performance, and be applied to modern INRs such as INCODE, highlighting its potential and flexibility. Code is available at https://github.com/maorash/Local-Global-INRs.