Abstract:Part-level 3D shape representations are crucial to shape reasoning and understanding. Two key sub-tasks are: 1) shape abstraction, creating primitive-based object parts; and 2) shape segmentation, finding partition-based object parts. However, for 3D object point clouds, most advanced methods produce parts relying on task-specific priors, such as similarity metrics and primitive geometries, resulting in misleading parts that deviate from semantics. To address prior limitations, we establish a foundation for joint shape abstraction and shape segmentation as formal linear transformations within a shared latent space, encapsulating essential dual-purpose membership information linking points and object parts for mutual reinforcement. We demonstrate that the transformations are underpinned by a derivation based on k-means, Non-negative Matrix Factorization (NMF), and the attention mechanism. As a result, we introduce Latent Membership Pursuit (LMP) for joint optimization of shape abstraction and segmentation. LMP utilizes a shared latent representation of object part membership to autonomously identify common object parts in both tasks without any supervision and priors. Furthermore, we adapt deformable superquadrics (DSQs) for primitives to capture variable part-level geometric and semantic information. Experiments on benchmark datasets validate that our approach enables mutual learning of shape abstraction and segmentation, and promotes consistent interpretations of 3D object shapes across instances and even categories in a fully unsupervised manner.

3D Shape Co-Segmentation Via Sparse and Low Rank Representations

Shape Representation and Recognition in High Dimensional Feature Space

3D Graph-S<SUP>2</SUP>Net: Shape-Aware Self-ensembling Network for Semi-supervised Segmentation with Bilateral Graph Convolution

3D Shape Co-segmentation by Combining Sparse Representation with Extreme Learning Machine.

Recurring Part Arrangements in Shape Collections

Scribble-Based 3D Shape Segmentation via Weakly-Supervised Learning

Unsupervised 3D Shape Segmentation and Co-Segmentation Via Deep Learning.

Unsupervised Co-Segmentation of 3D Shapes Via Affinity Aggregation Spectral Clustering.

A Unified Feature Representation and Learning Framework for 3D Shape

Shape Sparse Representation for Joint Object Classification and Segmentation

Simultaneous variational image segmentation and object recognition via shape sparse representation

3D Shape Segmentation Using Soft Density Peak Clustering and Semi-Supervised Learning

Transductive 3d Shape Segmentation Using Sparse Reconstruction

Shared Latent Membership Enables Joint Shape Abstraction and Segmentation With Deformable Superquadrics

Volumetric Shape Contexts for Mesh Co-Segmentation.

3D shape representation, matching and recognition

Cosegmentation and Cosketch by Unsupervised Learning.

Semi-supervised 3D shape segmentation with multilevel consistency and part substitution

Shape Topics: A Compact Representation and New Algorithms for 3D Partial Shape Retrieval

Segmentation Based on Shape Prior and Graph Model Optimization

Unsupervised Shape Co-segmentation Based on Transformation Network