Floorplan-Jigsaw: Jointly Estimating Scene Layout and Aligning Partial Scans

Cheng, Lin; Li, Changjian; Wang, Wenping

doi:10.1109/iccv.2019.00577

Cited by 19 publications

(8 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Then, we provide our experimental results comparing with baselines and the ablation study. Since image sequences are not available in previous floor plan datasets [7], [8], [10], [30], [31], we collect and annotate a new dataset, MP3D-FPE. We synthesize users scanning through the scenes in Matterport3D [23] with the Minos simulator [24].…”

Section: Methodsmentioning

confidence: 99%

“…The most stable and robust solutions for floor plan estimation [7]- [9], [30], [31] mainly rely on active sensors (e.g., LiDAR or depth cameras) to directly acquire 3D information of the scene. For instance, early works such as [30] leverage multiple 3D LiDAR-based scans by estimating primitive geometries from a layered-up formulation.…”

Section: B Floor Plan Estimationmentioning

confidence: 99%

“…For instance, early works such as [30] leverage multiple 3D LiDAR-based scans by estimating primitive geometries from a layered-up formulation. In contrast, works like [8], [31] define the problem by acquiring 3D data from dense depth images. Floorplan-Jigsaw [31] constraints the problem to a set of data with small overlapping, searching for the most coherent alignment in a 2D grid map; FloorNet [8] estimates low-level geometry with multiple Convolutional Neural Network branches and aggregates them with integer programming, assuming a predefined room-shape.…”

Section: B Floor Plan Estimationmentioning

confidence: 99%

See 2 more Smart Citations

360-DFPE: Leveraging Monocular 360-Layouts for Direct Floor Plan Estimation

Solarte¹,

Liu²,

Wu³

et al. 2021

Preprint

View full text Add to dashboard Cite

We present 360-DFPE, a sequential floor plan estimation method that directly takes 360-images as input without relying on active sensors or 3D information. Our approach leverages a loosely coupled integration between a monocular visual SLAM solution and a monocular 360-room layout approach, which estimate camera poses and layout geometries, respectively.Since our task is to sequentially capture the floor plan using monocular images, the entire scene structure, room instances, and room shapes are unknown. To tackle these challenges, we first handle the scale difference between visual odometry and layout geometry via formulating an entropy minimization process, which enables us to directly align 360layouts without knowing the entire scene in advance. Second, to sequentially identify individual rooms, we propose a novel room identification algorithm that tracks every room along the camera exploration using geometry information. Lastly, to estimate the final shape of the room, we propose a shortest path algorithm with an iterative coarse-to-fine strategy, which improves prior formulations with higher accuracy and faster run-time. Moreover, we collect a new floor plan dataset with challenging large-scale scenes, providing both point clouds and sequential 360-image information. Experimental results show that our monocular solution achieves favorable performance against the current state-of-the-art algorithms that rely on active sensors and require the entire scene reconstruction data in advance. We will release our code and dataset.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: B Floor Plan Estimationmentioning

confidence: 99%

Section: B Floor Plan Estimationmentioning

confidence: 99%

See 1 more Smart Citation

360-DFPE: Leveraging Monocular 360-Layouts for Direct Floor Plan Estimation

Solarte¹,

Liu²,

Wu³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The consistency term 𝐸 𝑐𝑜𝑛𝑠𝑖𝑠 is calculated by the consistency cost term introduced in [66]. The term denotes the number of pixels used by the primitive boundaries of all the polygonal loops together.…”

Section: Graphic Reconstruction With a Polygonal Coordinate Descent S...mentioning

confidence: 99%

Automated Semantics and Topology Representation of Residential-Building Space Using Floor-Plan Raster Maps

Wang

Jiang

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Automatically representing the semantics and topology of indoor building spaces from floor-plans is necessary for many applications, such as architectural design and indoor renovations. Extensive studies have investigated reconstructing indoor spaces with semantics and topology using professional means (e.g., laser scanning and photogrammetry). Floor-plan raster maps are widely and freely available for various purposes. Nevertheless, there is little research on the semantic and topological representation of indoor elements from floor-plan raster maps. To fill this gap, we propose a method of automatically representing the semantics and topology of indoor spaces from floor-plan raster maps. The proposed method first identifies basic geometric primitives from floor-plans using a learning-based hierarchical segmentation approach. Second, the relationship between the detected geometric primitives is assembled into the planar structure representation with topological data using mixed integer programming. Finally, the floor-plan graph structure is checked and optimized to maintain consistency with a polygonal coordinate descent strategy, resulting in a correct representation of the semantics and topology of the indoor space. Comprehensive evaluations demonstrate that the proposed method effectively achieves superior performance in three different datasets. The proposed method allows for 3D model popups for better visualizations and direct architectural model manipulations of the interior building layouts computation.

show abstract

“…There is also a rich literature of house-scale 3D reconstruction and modeling methods. Different methods take as input a set of RGB images [4,9], RGBD videos [12,35], RGBD panoramas [23], dense point clouds [29] or partial reconstructions [19]. In contrast, our input is a floorplan and a sparse set of RGB images without precise camera pose partially covering the interior, which is the typical data available on real estate websites and we produce as output a 3D textured mesh for the entire residence.…”

Section: Related Workmentioning

confidence: 99%

Plan2Scene: Converting Floorplans to 3D Scenes

Vidanapathirana¹,

Wu²,

Furukawa³

et al. 2021

Preprint

View full text Add to dashboard Cite

We address the task of converting a floorplan and a set of associated photos of a residence into a textured 3D mesh model, a task which we call Plan2Scene. Our system 1) lifts a floorplan image to a 3D mesh model; 2) synthesizes surface textures based on the input photos; and 3) infers textures for unobserved surfaces using a graph neural network architecture. To train and evaluate our system we create indoor surface texture datasets, and augment a dataset of floorplans and photos from prior work with rectified surface crops and additional annotations. Our approach handles the challenge of producing tileable textures for dominant surfaces such as floors, walls, and ceilings from a sparse set of unaligned photos that only partially cover the residence. Qualitative and quantitative evaluations show that our system produces realistic 3D interior models, outperforming baseline approaches on a suite of texture quality metrics and as measured by a holistic user study.

show abstract

Floorplan-Jigsaw: Jointly Estimating Scene Layout and Aligning Partial Scans

Cited by 19 publications

References 49 publications

360-DFPE: Leveraging Monocular 360-Layouts for Direct Floor Plan Estimation

360-DFPE: Leveraging Monocular 360-Layouts for Direct Floor Plan Estimation

Automated Semantics and Topology Representation of Residential-Building Space Using Floor-Plan Raster Maps

Plan2Scene: Converting Floorplans to 3D Scenes

Contact Info

Product

Resources

About