Patch-Grid: An Efficient and Feature-Preserving Neural Implicit Surface Representation

2023-08-26

arXiv.org

AI Summary

Traditional MLP-based implicit representations struggle to accurately reconstruct sharp edges, open boundaries, and sub-millimeter thin structures in CAD models, while suffering from low training efficiency. To address this, we propose Patch-Grid—a unified neural implicit signed distance function (SDF) representation that jointly leverages learnable patch-wise feature volumes and constructive solid geometry (CSG)-driven grids, enabling localized SDF fusion within a shared octree. Patch-Grid is the first method to integrate patch-level geometric priors with a structured CSG composition mechanism, thereby simultaneously ensuring high-fidelity preservation of sharp features and efficient optimization. Experiments demonstrate that Patch-Grid achieves state-of-the-art accuracy and robustness within seconds of training—significantly outperforming existing approaches on CAD-level reconstruction tasks involving sharp features, open surfaces, and sub-millimeter thin structures.

Abstract

Neural implicit representations are widely used for 3D shape modeling due to their smoothness and compactness, but traditional MLP-based methods struggle with sharp features, such as edges and corners in CAD models, and require long training times. To address these limitations, we propose Patch-Grid, a unified neural implicit representation that efficiently fits complex shapes, preserves sharp features, and handles open boundaries and thin geometric structures. Patch-Grid learns a signed distance field (SDF) for each surface patch using a learnable patch feature volume. To represent sharp edges and corners, it merges the learned SDFs via constructive solid geometry (CSG) operations. A novel merge grid organizes patch feature volumes within a shared octree structure, localizing and simplifying CSG operations. This design ensures robust merging of SDFs and significantly reduces computational complexity, enabling training within seconds while maintaining high fidelity. Experimental results show that Patch-Grid achieves state-of-the-art reconstruction quality for shapes with intricate sharp features, open surfaces, and thin structures, offering superior robustness, efficiency, and accuracy.

Problem

• Efficiently models 3D shapes with sharp features
• Preserves edges and corners in CAD models
• Reduces training time while maintaining high fidelity

Innovation

• Patch-Grid uses learnable patch feature volumes.
• CSG operations merge SDFs for sharp features.
• Merge grid organizes patches in an octree structure.