Reproduce “Any 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale” 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

Dongxu Wei, Qi Xu, Zhiqi Li, Hangning Zhou, Cong Qiu, Hailong Qin, Mu Yang, Zhaopeng Cui and 1 more

3DDiT generates novel scene views by denoising fixed-length 3D tokens built by 3DRAE from posed images and camera information. Tokens decode at queried camera poses. On DL3DV with one-view conditioning, it renders 16 views in 1.98 seconds versus 85.2 for SEVA, with better FID and pose-error values. This advantage is over tested 2D diffusion baselines, not regression methods.

Summary by alphaXiv.

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