Reproduce “Infinigen Indoors: Photorealistic Indoor Scenes using Procedural 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
Alexander Raistrick, Lingjie Mei, Karhan Kayan, David Yan, Yiming Zuo, Beining Han, Hongyu Wen, Meenal Parakh and 4 more
Infinigen Indoors, developed by researchers at Princeton University, procedurally generates diverse and photorealistic indoor scenes, extending the capabilities of the original Infinigen system. The system produces large-scale synthetic datasets with perfect ground truth annotations, demonstrating improved performance for models trained on its data in tasks like shadow removal and occlusion boundary detection.
Key insights
- A declarative, constraint-based system using a domain-specific language enables precise control over object placement and spatial relationships within indoor scenes.
- Fully procedural generation of assets and architectural elements, implemented using Blender's geometry nodes, allows for infinite variation and avoids reliance on static asset libraries.
- The inherent availability of pixel-perfect ground truth annotations in procedurally generated scenes directly addresses data scarcity issues for complex computer vision tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.