Fast Mesh Denoising With Data Driven Normal Filtering Using Deep Variational Autoencoders

Abstract

Recent advances in 3D scanning technology have enabled the deployment of 3D models in various industrial applications like digital twins, remote inspection and reverse engineering. Despite their evolving performance, 3D scanners, still introduce noise and artifacts in the acquired dense models. In this work, we propose a fast and robust denoising method for dense 3D scanned industrial models. The proposed approach employs conditional variational autoencoders to effectively filter face normals. Training and inference are performed in a sliding patch setup reducing the size of the required training data and execution times. We conducted extensive evaluation studies using 3D scanned and CAD models. The results verify plausible denoising outcomes, demonstrating similar or higher reconstruction accuracy, compared to other state-of-the-art approaches. Specifically, for 3D models with more than 1e4 faces, the presented pipeline is twice as fast as methods with equivalent reconstruction error. Index Terms-3D mesh denoising, data driven normal filtering, variational autoencoders.

XRSISE: An XR Training System for Interactive Simulation and Ergonomics Assessment

“…Moreover, as the authoring tools provided by XRSISE for 3D models creation are based on Unity's existing game-object logic, the de novo design of industrial training applications involving large scenes (e.g., for assembly tasks) would require significant manual effort. For such purposes, 3D models may be retrieved from other sources (such as databases with CAD objects for industrial applications), or 3D scanning technologies may be exploited along with point cloud denoizing (Nousias et al, 2021) and semantic segmentation techniques.…”

Analysing detail preserving capabilities of bilateral, laplacian and taubin mesh filtering methods

“…Deep learning-based mesh filtering [32,33], a burgeoning area, employs neural networks to learn optimal filtering parameters from data. While traditional methods rely on hand-crafted heuristics, these learnable filters adapt based on the input, making them versatile.…”

Denoising and Simplification in the Construction of 3d Digital Models of Complex Objects

“…Li et al [93] proposed a new deep normal filtering network that processes local patches and preserves feature regions. Other methods use convolutional neural networks in different ways for the mesh denoising problem [94,95,96,97]. Although data-driven methods generally do not require parameter definition, they are strongly dependent on the training dataset.…”