The collection consists of 122 scanned floor plan documents divided in 4 different subsets regarding their origin and style. It contains documents of dierent qualities, resolutions, and modeling styles, which is suitable to test the robustness of the analysis techniques.
The dataset is fully groundtruthed for the structural symbols: rooms, walls, doors, windows, parking doors, and room separations. The GT not only makes specific their locations in the images, but also includes structural relations between them that is of special interest for analysis systems.
The complete database can be downloaded here: CVC-FP
Evaluation scripts for wall segmentation and room detection.
The Matlab evaluation scripts for wall and room detection are available for free download here:Evaluation. We recommend to read carefully the paper in here before running the scripts. For any help, you can contact the owner.
The SGT tool is thought to perform general purpose groundtruthing, not only to one specific domain, as most of the existing tools are. It grants full flexibility since the proprietaries of the databases can create, modify, and erase their own object classes. Additionally, it is
possible to define and declare n-ary properties for the objects, which allows to represent the ground truth as an attributed graph, where nodes are labeled objects and edges are relations between them. The tool is user-friendly, it allows 2 different labeling options, and the output is in the standard Scalable Vector Graphics (SVG). The tool is a cross-platform running on a web service, that enforces co-working without sacrificing security. It has been implemented in php5 and HTML5, and the collections are stored in a relational database like MySQL.
The SGT tool can be downloaded here: sgt_1_2
The user manual will be available soon.
SVG to PNG GT
Here we share freely the JAVA project that converts the grountruth in SVG file format into different PNG files –one for each object category in the GT–.SVGtoPNG
Floor plan bechmark
Here we detail the benchmarking results on wall and room recognition for all the images contained in this databes. ResultsComplete.