Reproduce “HLG: Comprehensive 3D Room Construction via Hierarchical Layout Generation” 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
Xiping Wang, Yuxi Wang, Mengqi Zhou, Junsong Fan, Zhaoxiang Zhang
HLG introduces a hierarchical approach for comprehensive 3D room construction, addressing the challenge of fine-grained object placement in indoor scenes. It systematically refines layouts from coarse furniture arrangements to intricate item placements, achieving highly detailed, functionally realistic, and controllable 3D environments.
Key insights
- Decomposing complex 3D scenes into a hierarchical structure through vertical and horizontal decoupling enables systematic optimization of object placements at different granularities.
- Leveraging LLMs for scene graph construction and detailed constraint extraction allows for flexible and precise control over fine-grained object relationships from multimodal inputs.
- A trainable optimization network, incorporating specific loss functions, can learn to resolve common placement issues (collisions, instability, incorrect orientation) in an end-to-end fashion.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.