Abstract:Low-level 3D representations, such as point clouds, meshes, NeRFs, and 3D Gaussians, are commonly used to represent 3D objects or scenes. However, human perception typically understands 3D objects at a higher level as a composition of parts or structures rather than points or voxels. Representing 3D objects or scenes as semantic parts can benefit further understanding and applications. In this paper, we introduce PartGS, part-aware 3D reconstruction by a hybrid representation of 2D Gaussians and Superquadrics, which parses objects or scenes into semantic parts, digging 3D structural clues from multi-view image inputs. Accurate structured geometry reconstruction and high-quality rendering are achieved at the same time. Our method simultaneously optimizes superquadric meshes and Gaussians by coupling their parameters within our hybrid representation. On one hand, this hybrid representation inherits the advantage of superquadrics to represent different shape primitives, supporting flexible part decomposition of scenes. On the other hand, 2D Gaussians capture complex texture and geometry details, ensuring high-quality appearance and geometry reconstruction. Our method is fully unsupervised and outperforms existing state-of-the-art approaches in extensive experiments on DTU, ShapeNet, and real-life datasets.

Superquadric Similarity Measure With Spherical Harmonics In 3d Object Recognition

3D Object Recognition by Fast Spherical Correlation Between Combined View EGIs and PFT.

Study On Similarity Measure Of Superquadrics

3D model similarity measurement with geometric feature map based on phase-encoded range image

Superquadric Matching Computation with Spherical Harmonics and 3D Moments

Matching 3D Models with Global Geometric Feature Map

Extracting Superquadric-based Geon Description for 3D Object Recognition

Shape Representation and Recognition in High Dimensional Feature Space

Superquadric-based geons recognition utilizing support vector machines

Survey of modeling and recognition using superquadrics for scattered data points

Superquadric-based 3D Scene Reconstruction and Interpretation

A similarity computing algorithm for volumetric data sets

Superquadric Object Representation for Optimization-based Semantic SLAM

Volumetric Part Based 3D Object Classification

Similarity Comparison of Triangular Mesh by Spherical Hybrid Curvature Image

Superquadric-based reconstruction from 2D images

3d Object Classification Based On Volumetric Parts

Content-Based 3d Model Retrieval Based On The Spatial Geometric Descriptor

SQ-SLAM: Monocular Semantic SLAM Based on Superquadric Object Representation

(μ + λ) evolutionary strategy for 3D modeling and segmentation with superquadrics

PartGS:Learning Part-aware 3D Representations by Fusing 2D Gaussians and Superquadrics