TopoDiT-3D: Topology-Aware Diffusion Transformer with Bottleneck Structure for 3D Point Cloud Generation

TL;DR

TopoDiT-3D introduces a topology-aware diffusion transformer with a bottleneck structure that effectively incorporates global topological information into 3D point cloud generation, enhancing quality and diversity.

Contribution

The paper presents a novel topology-aware diffusion transformer that integrates persistent homology into feature learning using a Perceiver Resampler-based bottleneck, improving 3D point cloud generation.

Findings

Outperforms state-of-the-art models in visual quality and diversity.

Enhances training efficiency through adaptive filtering of local features.

Highlights the importance of topological information in 3D shape generation.

Abstract

Recent advancements in Diffusion Transformer (DiT) models have significantly improved 3D point cloud generation. However, existing methods primarily focus on local feature extraction while overlooking global topological information, such as voids, which are crucial for maintaining shape consistency and capturing complex geometries. To address this limitation, we propose TopoDiT-3D, a Topology-Aware Diffusion Transformer with a bottleneck structure for 3D point cloud generation. Specifically, we design the bottleneck structure utilizing Perceiver Resampler, which not only offers a mode to integrate topological information extracted through persistent homology into feature learning, but also adaptively filters out redundant local features to improve training efficiency. Experimental results demonstrate that TopoDiT-3D outperforms state-of-the-art models in visual quality, diversity, and training efficiency. Furthermore, TopoDiT-3D demonstrates the importance of rich topological information for 3D point cloud generation and its synergy with conventional local feature learning. Videos and code are available at https://github.com/Zechao-Guan/TopoDiT-3D.