Mesh Compression with Quantized Neural Displacement Fields

2025-03-28

AI Summary

This work addresses the challenge of high-fidelity compression of unstructured 3D triangular meshes at extreme compression ratios. Methodologically, it introduces a lightweight implicit neural representation (INR)-based encoding framework that leverages a coarse mesh as geometric prior and parameterizes the vertex displacement field via a compact neural network. The displacement field is represented implicitly through quantized network weights, optimized end-to-end in a differentiable manner—eliminating the need for explicit displacement storage. This constitutes the first systematic application of INR to unstructured mesh compression; its core innovation lies in encoding intricate geometric details implicitly using an extremely small set of network parameters. Experiments demonstrate that the method consistently outperforms state-of-the-art approaches across compression ratios of 4×–380×, preserving fine-grained geometric textures and topological fidelity while reducing model size by one to two orders of magnitude compared to storing either raw displacement fields or original mesh data.

Technology Category

Data Mining & Knowledge Management: Data CompressionMachine Learning: Learning on the Edge & Model CompressionComputer Vision: 3D Computer Vision

Application Category

Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphsWeb Mining and Content Analysis: Large pretrained models with web dataUser Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Implicit neural representations (INRs) have been successfully used to compress a variety of 3D surface representations such as Signed Distance Functions (SDFs), voxel grids, and also other forms of structured data such as images, videos, and audio. However, these methods have been limited in their application to unstructured data such as 3D meshes and point clouds. This work presents a simple yet effective method that extends the usage of INRs to compress 3D triangle meshes. Our method encodes a displacement field that refines the coarse version of the 3D mesh surface to be compressed using a small neural network. Once trained, the neural network weights occupy much lower memory than the displacement field or the original surface. We show that our method is capable of preserving intricate geometric textures and demonstrates state-of-the-art performance for compression ratios ranging from 4x to 380x.

Problem

• Extends implicit neural representations to compress 3D meshes
• Encodes displacement fields using small neural networks
• Achieves high compression ratios while preserving geometric details

Innovation

• Uses quantized neural displacement fields
• Compresses 3D meshes via neural networks
• Achieves high compression ratios effectively