StrucADT: Generating Structure-Controlled 3D Point Clouds With Adjacency Diffusion Transformer

Area of Science:

• Computer Vision
• 3D Shape Generation
• Machine Learning

Background:

• Existing 3D generative models produce diverse shapes but lack user control.
• Controllable 3D point cloud generation is crucial for practical applications.

Purpose of the Study:

• To develop a novel method for controllable 3D point cloud generation.
• To enable generation of 3D point clouds based on user-defined structural requirements.

Main Methods:

• Introduced StructureGraph representation based on part adjacency.
• Developed StrucADT, a structure-controllable point cloud generation model.
• Utilized StructureGraphNet, cCNF Prior, and Diffusion Transformer modules.

Main Results:

• Generated high-quality and diverse 3D point cloud shapes.
• Achieved state-of-the-art performance in controllable point cloud generation.
• Demonstrated successful generation of point clouds based on specified structures.

Conclusions:

• The proposed method effectively addresses the lack of control in 3D point cloud generation.
• StrucADT enables precise control over 3D shape generation through structural specifications.
• This work advances the field of controllable 3D shape synthesis.