New InSpace Framework Generates 3D Indoor Scenes from Single 360° Images

Researchers have developed InSpace, a novel framework for generating complete 3D indoor scenes from a single 360° image. This method addresses the limitations of existing single-image-to-3D techniques by focusing on structural layout in addition to asset generation. InSpace utilizes a three-stage process involving partial scene geometry estimation, coarse structure generation with view-selective cross-attention, and detailed layout and asset creation using a global-local hybrid attention mechanism powered by flow matching. The framework also introduces ERP-FRONT, a new dataset for indoor scene generation. AI

IMPACT Enables more comprehensive 3D scene reconstruction from limited input, potentially improving applications in virtual reality and robotics.

RANK_REASON The cluster describes a new academic paper detailing a novel method for 3D scene generation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• 360° image
• Equirectangular Projection (ERP)
• ERP-FRONT
• Flow Matching for Generative Modeling
• global-local hybrid attention
• InSpace
• view-selective cross-attention

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →