Event

SceneBloom: An Agentic Framework for Indoor Scene Generation

1 reports · 1 sources

Yuhan Liu, Lianhui Qin, and Shenji Wan propose SceneBloom, an agentic framework that combines visual layout sampling with contract-guided 3D compilation to generate simulated indoor environments. Its hierarchical sampler leverages diffusion priors to generate furniture layouts within functional zones, and the compiler then performs geometric fitting and local pose correction based on relations extracted from each proposal. The authors claim that goal-navigation policies trained in these scenes achieve higher success rates on held-out real scans than policies trained with an equal number of real training scans.