MoCA: Mixture-of-Components Attention for Scalable Compositional 3D Generation

2025-12-08

AI Summary

Existing partial-observation 3D generation methods rely on dense global attention, resulting in quadratic computational complexity with respect to the number of components—hindering scalable, fine-grained compositional 3D asset synthesis. To address this, we propose a scalable sparse global attention framework: first, components are ranked by importance; then, top-k routing selects salient components, while non-salient ones undergo semantic compression to preserve contextual priors with drastically reduced computation. Furthermore, we introduce a hybrid attention mechanism that jointly models local geometric details and global structural relationships. Experiments demonstrate that our method significantly outperforms existing baselines on compositional 3D object and scene generation, enabling high-fidelity, efficient synthesis for scenes comprising hundreds of components.

Technology Category

• Computer Vision: 3D Computer Vision
• Natural Language Processing: Generation
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Compositionality is critical for 3D object and scene generation, but existing part-aware 3D generation methods suffer from poor scalability due to quadratic global attention costs when increasing the number of components. In this work, we present MoCA, a compositional 3D generative model with two key designs: (1) importance-based component routing that selects top-k relevant components for sparse global attention, and (2) unimportant components compression that preserve contextual priors of unselected components while reducing computational complexity of global attention. With these designs, MoCA enables efficient, fine-grained compositional 3D asset creation with scalable number of components. Extensive experiments show MoCA outperforms baselines on both compositional object and scene generation tasks. Project page: https://lizhiqi49.github.io/MoCA

Problem

• Addresses poor scalability in part-aware 3D generation due to quadratic attention costs
• Introduces importance-based component routing and compression for efficient 3D asset creation
• Enables fine-grained compositional 3D generation with a scalable number of components

Innovation

• Uses importance-based component routing for sparse attention
• Compresses unimportant components to reduce computational cost
• Enables scalable compositional 3D generation with many components