Extreme Compression of Adaptive Neural Images

2024-05-27

arXiv.org

AI Summary

To address the low compression ratio of Implicit Neural Representations (INRs) and the challenge of achieving both high compression rates and high fidelity in image compression, this paper proposes Adaptive Neural Images (ANI), a novel framework enabling dynamic adjustment of representation granularity based on inference or transmission requirements. ANI achieves a 4× reduction in bits per pixel (bpp)—i.e., 75% bpp reduction—without perceptible loss of fine details. Its core contributions are: (1) a new paradigm for 4-bit low-bitweight quantization specifically designed for ANI; and (2) a joint optimization mechanism integrating adaptive structural pruning with reparameterization. Experimental results demonstrate that ANI significantly reduces bpp while fully preserving visual fidelity and sensitive textures. The framework is efficient, scalable, and generalizable, establishing a practical foundation for neural field compression.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

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 idea offers unprecedented benefits such as continuous resolution and memory efficiency, enabling new compression techniques. However, representing data as neural networks poses new challenges. For instance, given a 2D image as a neural network, how can we further compress such a neural image?. In this work, we present a novel analysis on compressing neural fields, with the focus on images. We also introduce Adaptive Neural Images (ANI), an efficient neural representation that enables adaptation to different inference or transmission requirements. Our proposed method allows to reduce the bits-per-pixel (bpp) of the neural image by 4x, without losing sensitive details or harming fidelity. We achieve this thanks to our successful implementation of 4-bit neural representations. Our work offers a new framework for developing compressed neural fields.

Problem

• Explores novel neural image compression techniques
• Achieves extreme compression of adaptive neural images
• Enables 4-bit neural representations without fidelity loss

Innovation

• Adaptive Neural Images enable compression adaptation
• 8x bits-per-pixel reduction without fidelity loss
• 4-bit neural representations achieve state-of-art PSNR/bpp