AnchoredDream: Zero-Shot 360° Indoor Scene Generation from a Single View via Geometric Grounding

TL;DR

AnchoredDream tackles the challenge of generating a complete 360° indoor scene from a single image in a zero-shot setting by grounding appearance generation in high-fidelity geometry. It introduces an appearance-geometry mutual boosting pipeline that uses a vision-language model to obtain a scene description, builds a geometry-rich 3D layout, and progressively warps, inpaints, and refines appearance while maintaining geometric plausibility through a Grouting Block and a 3D Gaussian representation. Post-optimization and multi-view consistency strategies further enforce cross-view coherence, yielding superior appearance consistency and geometric plausibility compared with strong baselines. The approach enables robust, geometry-aware single-view scene synthesis with potential applications in robotics, virtual content creation, and text-to-scene generation, highlighting the value of geometric grounding for high-quality zero-shot scene generation.

Abstract

Single-view indoor scene generation plays a crucial role in a range of real-world applications. However, generating a complete 360° scene from a single image remains a highly ill-posed and challenging problem. Recent approaches have made progress by leveraging diffusion models and depth estimation networks, yet they still struggle to maintain appearance consistency and geometric plausibility under large viewpoint changes, limiting their effectiveness in full-scene generation. To address this, we propose AnchoredDream, a novel zero-shot pipeline that anchors 360° scene generation on high-fidelity geometry via an appearance-geometry mutual boosting mechanism. Given a single-view image, our method first performs appearance-guided geometry generation to construct a reliable 3D scene layout. Then, we progressively generate the complete scene through a series of modules: warp-and-inpaint, warp-and-refine, post-optimization, and a novel Grouting Block, which ensures seamless transitions between the input view and generated regions. Extensive experiments demonstrate that AnchoredDream outperforms existing methods by a large margin in both appearance consistency and geometric plausibility--all in a zero-shot manner. Our results highlight the potential of geometric grounding for high-quality, zero-shot single-view scene generation.

Figures (18)

• Figure 1: Single-view to 360° scene generation. Given a single-view indoor image, AnchoredDream generates a complete 360° scene represented with 3DGS, while maintaining both appearance consistency and geometric plausibility throughout.
• Figure 2: Overview of the AnchoredDream pipeline. Given a single-view image, our method generates a complete 360° indoor scene through appearance-guided geometry generation, geometry-aware appearance synthesis, and post-optimization.
• Figure 3: Illustration of the Grouting Block. Our two-stage refinement process ensures seamless blending between the input view and newly generated regions.
• Figure 4: Effect of geometry alignment on scene geometry (case: "A vintage-style budget hotel room"). Geometry alignment effectively rectifies input view's geometry errors while maintaining the global geometry consistency of the full scene.
• Figure 5: Visual quality comparison for the case “A modern-style living room”. Our method faithfully preserves the appearance style of the input image, whereas the baselines exhibit visual artifacts, style drift, and inconsistencies in the generated results.