DMESH: A Structure-Preserving Diffusion Model for 3-D Mesh Denoising.

Area of Science:

• Computer Graphics
• Artificial Intelligence
• Computational Geometry

Background:

• Denoising diffusion models excel at generating high-quality images by progressively removing noise.
• Applying diffusion models to mesh denoising is challenging due to the need to preserve structure while removing artifacts.

Purpose of the Study:

• To develop a diffusion-based mesh denoiser capable of removing noise while preserving the original mesh structure.
• To create a topology-agnostic model for versatile mesh denoising.

Main Methods:

• Formulated a structure-preserving diffusion process by diffusing mesh vertices into specific noise distributions.
• Proposed a topology-agnostic diffusion model using 2-D projections of mesh vertices for deep network learning.
• Employed reverse diffusion and refinement based on 2-D projections to obtain the denoised mesh.

Main Results:

• The proposed structure-preserving diffusion process effectively removes noise without compromising mesh integrity.
• The topology-agnostic model successfully handles meshes with irregular topologies.
• Experimental results demonstrate superior performance compared to state-of-the-art mesh denoising methods in both quantitative and qualitative evaluations.

Conclusions:

• The developed diffusion-based mesh denoiser offers a robust solution for artifact removal while preserving essential mesh structures.
• The topology-agnostic approach broadens the applicability of diffusion models in mesh processing.
• This method represents a significant advancement in the field of 3D mesh denoising.