English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Diffusion Transformers have recently shown remarkable effectiveness ingenerating high-quality 3D point clouds. However, training voxel-baseddiffusion models for high-resolution 3D voxels remains prohibitively expensivedue to the cubic complexity of attention operators, which arises from theadditional dimension of voxels. Motivated by the inherent redundancy of 3Dcompared to 2D, we propose FastDiT-3D, a novel masked diffusion transformertailored for efficient 3D point cloud generation, which greatly reducestraining costs. Specifically, we draw inspiration from masked autoencoders todynamically operate the denoising process on masked voxelized point clouds. Wealso propose a novel voxel-aware masking strategy to adaptively aggregatebackground/foreground information from voxelized point clouds. Our methodachieves state-of-the-art performance with an extreme masking ratio of nearly99%. Moreover, to improve multi-category 3D generation, we introduceMixture-of-Expert (MoE) in 3D diffusion model. Each category can learn adistinct diffusion path with different experts, relieving gradient conflict.Experimental results on the ShapeNet dataset demonstrate that our methodachieves state-of-the-art high-fidelity and diverse 3D point cloud generationperformance. Our FastDiT-3D improves 1-Nearest Neighbor Accuracy and Coveragemetrics when generating 128-resolution voxel point clouds, using only 6.5% ofthe original training cost.

Fast Training of Diffusion Transformer with Extreme Masking for 3D Point Clouds Generation