Related Experiment Video
Updated: Aug 9, 2026
06:55
Synthesis of Cyclic Polymers and Characterization of Their Diffusive Motion in the Melt State at the Single Molecule Level
Published on: September 26, 2016
Disentangle-and-Diffuse: Structure-Aware Modeling with Diffusion-Based Generation
IEEE Computer Graphics and Applications
|August 7, 2026
Summary
Disentangle-and-Diffuse generates 3D shapes by breaking them into parts, encoding features, and using diffusion models for synthesis. This structure-aware approach enhances geometric quality and enables flexible shape manipulation.
Area of Science:
- Computer Vision
- Artificial Intelligence
- 3D Computer Graphics
Background:
- Generating structured 3D shapes is challenging.
- Existing methods struggle with geometric quality and structural consistency.
Purpose of the Study:
- Introduce Disentangle-and-Diffuse, a novel framework for structure-aware 3D shape generation.
- Improve geometric quality and structural consistency in generated 3D shapes.
Main Methods:
- Automatic semantic part decomposition of 3D shapes.
- Invariant/equivariant feature encoding for geometric and pose information.
- Dual-stream transformer for context fusion and a diffusion model for part synthesis.
Main Results:
- Achieved superior structural consistency and geometric quality compared to baselines.
- Demonstrated flexible 3D shape manipulation and interpolation capabilities.
- Validated effectiveness on challenging 3D benchmarks.
Conclusions:
- Combining structural and feature-level disentanglement is key for controllable 3D generative modeling.
- Disentangle-and-Diffuse offers a promising direction for advanced 3D shape generation.
