Modeling 3D Surface Manifolds with a Locally Conditioned Atlas

Przemysław Spurek,Sebastian Winczowski,Maciej Zięba,Tomasz Trzciński,Kacper Kania,Marcin Mazur

2024-04-06

Abstract:Recently proposed 3D object reconstruction methods represent a mesh with an atlas - a set of planar patches approximating the surface. However, their application in a real-world scenario is limited since the surfaces of reconstructed objects contain discontinuities, which degrades the quality of the final mesh. This is mainly caused by independent processing of individual patches, and in this work, we postulate to mitigate this limitation by preserving local consistency around patch vertices. To that end, we introduce a Locally Conditioned Atlas (LoCondA), a framework for representing a 3D object hierarchically in a generative model. Firstly, the model maps a point cloud of an object into a sphere. Secondly, by leveraging a spherical prior, we enforce the mapping to be locally consistent on the sphere and on the target object. This way, we can sample a mesh quad on that sphere and project it back onto the object's manifold. With LoCondA, we can produce topologically diverse objects while maintaining quads to be stitched together. We show that the proposed approach provides structurally coherent reconstructions while producing meshes of quality comparable to the competitors.

Computer Science

What problem does this paper attempt to address?

### The Problem Addressed by the Paper

The paper "Modeling 3D Surface Manifolds with a Locally Conditioned Atlas" aims to solve a critical issue in 3D object reconstruction: **how to generate high-quality 3D meshes while maintaining local consistency**.

Specifically, existing 3D object reconstruction methods typically use an atlas to represent 3D meshes, approximating surfaces through a set of planar patches. However, these methods have limitations in practical applications because the surfaces of reconstructed objects contain discontinuities, which reduce the quality of the final mesh. This discontinuity is mainly due to the independent processing of each patch.

To address this issue, the authors propose a new framework—**Locally Conditioned Atlas (LoCondA)**. This framework avoids discontinuities by maintaining local consistency around the vertices of the patches. Specifically, LoCondA first maps the point cloud onto a sphere and then uses prior information on the sphere to ensure that the mapping is locally consistent on both the sphere and the target object. In this way, high-quality 3D meshes can be generated by sampling quadrilateral meshes on the sphere and projecting them back onto the object's manifold.

### Main Contributions

1. **Proposed a general patch-based reconstruction framework** that can generate high-fidelity 3D meshes from raw point clouds.
2. **Introduced the Continuous Atlas**, a new paradigm that generalizes existing atlas methods and can adaptively sample any number of patches to cover any part of the reconstructed object.
3. **Demonstrated a simple conditioning mechanism** that allows information sharing between atlas-based methods, thereby eliminating self-intersections and holes in the reconstructed mesh.

### Related Work

The paper also reviews existing 3D shape representation methods, including dense pixel depth maps, normal maps, point clouds, meshes, implicit functions, voxels, shape primitives, and parametric mappings. The authors particularly focus on polygon mesh-based methods due to their good balance between memory requirements and surface fitting accuracy.

### Experimental Results

The paper validates the effectiveness of LoCondA through experiments, including evaluations of generation and reconstruction capabilities. The experimental results show that LoCondA achieves performance comparable to or even better than existing methods in generating point clouds and reconstructing 3D meshes. In particular, LoCondA demonstrates significant advantages in generating high-quality, seamless 3D meshes.

### Conclusion

In summary, this paper addresses the common discontinuity issue in 3D object reconstruction by introducing the Locally Conditioned Atlas (LoCondA), thereby generating high-quality, structurally coherent 3D meshes. This method has significant implications for applications in computer vision and machine learning.