SceneAligner: 3D-Grounded Floorplan Localization in the Wild

2026-05-21 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

SceneAligner is a novel approach for 3D-grounded floorplan localization designed for large-scale, unconstrained environments using rasterized floorplans. Unlike prior methods limited to small-scale settings and precise vectorized floorplans, SceneAligner reconstructs a gravity-aligned 3D scene from an unconstrained image collection. This 3D scene is then projected into a 2D density map, serving as a floorplan proxy. The system formulates floorplan localization as aligning this proxy with the input floorplan using a 2D similarity transform. To overcome the appearance differences between density maps and architectural floorplans, SceneAligner fine-tunes a 2D foundation model. This fine-tuning scheme promotes semantically aligned matches while maintaining structural consistency. Extensive experiments demonstrate SceneAligner's substantial improvements over existing methods, even in extremely sparse scenarios, such as those with only a single input image. Its code and data will be publicly available.

Key takeaway

For computer vision engineers developing indoor navigation or localization systems, SceneAligner offers a robust solution for "in the wild" floorplan localization. You should consider integrating its 3D-grounded approach, especially when dealing with unconstrained image collections or rasterized floorplans in large buildings. This method significantly improves accuracy, even with sparse input data like a single image, potentially simplifying data collection requirements for your projects.

Key insights

SceneAligner enables robust floorplan localization in complex, real-world settings by grounding the task in 3D scene reconstruction.

Principles

• Ground localization in 3D scene reconstruction.
• Use 2D density maps as floorplan proxies.
• Adapt foundation models for cross-modal alignment.

Method

Reconstruct a gravity-aligned 3D scene from images, project it to a 2D density map, then align this proxy with the input floorplan using a fine-tuned 2D foundation model for cross-modal correspondence.

In practice

• Orient visitors in large public buildings.
• Localize devices within complex indoor maps.
• Enable navigation with minimal visual input.

Topics

• Floorplan Localization
• 3D Reconstruction
• Computer Vision
• Foundation Models
• Indoor Navigation
• Sparse Data

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• GroundedPlanBench: Spatially grounded long-horizon task planning for robot manipulation — Jointly planning actions and spatial grounding improves robot task success and action accuracy for VLMs.
• Part 3: Simultaneous Localization & Mapping: Which SLAM Is For You? on OpenCV Live! — The webinar explores key considerations and challenges in Simultaneous Localization & Mapping (SLAM) applications.
• UI-Zoomer: Uncertainty-Driven Adaptive Zoom-In for GUI Grounding — UI-Zoomer adaptively zooms into GUI elements based on prediction uncertainty, improving localization without retraining.
• Heterogeneous LiDAR Early Fusion and Learned Re-Ranking Strategy for Robust Long-Term Place Recognition in Unstructured Environments — Heterogeneous LiDAR fusion and learned re-ranking enhance long-term place recognition in complex, repetitive environments.
• MatterDoor: Sampling Zero-shot Spatio-semantic Priors using Generative Models — Generative models can sample spatio-semantic priors for robotic planning in partially observed 3D environments.
• Expanding Spatial and Temporal Context for Robotic Imitation Learning With Scene Graphs — Scene graphs provide explicit, compact memory for robots to reason over long-horizon tasks under partial observability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.