Reproduce “Direct3D-V3: Latent-Free Diffusion Transformers in the Native Voxel Domain” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Shuang Wu, Jiachen Qian, Youtian Lin, Feihu Zhang, Philip Torr, Yao Yao, Xun Cao

Direct3D-V3 generates high-resolution 3D shapes directly as signed-distance fields (SDFs), rather than decoding a compressed geometry code. A sparse two-level transformer uses patch-level tokens for broad structure and voxel-level tokens for local detail, reaching a reported resolution of 1536³. On 102 input images it scores 0.4198 on Uni3D and 0.4535 on ULIP-2; these indicate mainly semantic alignment, not fine geometric precision.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.