Extreme Compression of Adaptive Neural Images

TL;DR

This paper introduces Adaptive Neural Images, a neural representation that significantly compresses neural images by 8 times without quality loss, advancing neural field compression techniques.

Contribution

It proposes a novel adaptive neural image framework that achieves high compression rates with minimal fidelity loss, setting new state-of-the-art performance.

Findings

Reduced bits-per-pixel by 8 times

Maintained image fidelity and details

Achieved state-of-the-art PSNR/bpp trade-off

Abstract

Implicit Neural Representations (INRs) and Neural Fields are a novel paradigm for signal representation, from images and audio to 3D scenes and videos. The fundamental idea is to represent a signal as a continuous and differentiable neural network. This new approach poses new theoretical questions and challenges. Considering a neural image as a 2D image represented as a neural network, we aim to explore novel neural image compression. In this work, we present a novel analysis on compressing neural fields, with focus on images and introduce Adaptive Neural Images (ANI), an efficient neural representation that enables adaptation to different inference or transmission requirements. Our proposed method allows us to reduce the bits-per-pixel (bpp) of the neural image by 8 times, without losing sensitive details or harming fidelity. Our work offers a new framework for developing compressed neural fields. We achieve a new state-of-the-art in terms of PSNR/bpp trade-off thanks to our successful implementation of 4-bit neural representations.