Reproduce “MultiCube: Compositional 3D Generation With Part-Level Semantic and Spatial Control” 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
Ava Pun, Kangle Deng, Yiheng Zhu, Jun-Yan Zhu, Maneesh Agrawala, Tinghui Zhou
MultiCube generates 3D objects as separate meshes from a text prompt, a part schema, and an axis-aligned box for each part. Its two-stage design conditions generation on per-part semantics and layouts. On 73 held-out PartObjaverse-Tiny examples, Stage 2 part-level Chamfer distance was 0.040 versus 0.183 for CubePart, with geometric match scores of 0.948 versus 0.787; this was not an end-to-end result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.