The main goal of this database is to provide a framework for the evaluation of different symbol spotting methods in vectorized graphic documents.
All the images of this database have been converted by using a raster-to-vector algorithm implemented in the QGar library.

We can define the symbol spotting problem as the task of recognizing and locating symbols appearing within complex graphical documents without the need of segmenting those symbols from the backround beforehand. The FPLAN-POLY database consists of a set of real floorplans which have been vectorized and a set of symbols which are taken as models. The ground-truth of this database defining which symbols appear in which floorplans and their locations is also provided.

Database Description

The FPLAN-POLY database consist of three different folders:
  • The Floorplans folder contains 42 floorplan vectorial images in dxf format which can be viewed with the common CAD softwares or easily parsed. In these files, all the symbols are defined by just using the geometric primitive POLYLINE.
  • The Model Symbols folder contains 38 symbols cropped from the original floorplans which are used as models. The models are also in dxf format and defined by the POLYLINE primitives.
  • The GroundTruth folder contains 38 XML files relating each model symbol with their locations in the floorplans.

Ground-truth file format

In order to have an accurate localization of the symbols, we use the convex-hull as the primitive of our ground-truth.
The ground-truth of this collection is organized in 38 XML files defining the convex-hull coordinates where all the model symbols can be found in each floorplan image in the following way:

	< page page_id="p01c03.dxf" >
		< coords >
			< point x="26.0" y="528.0" >
			< point x="30.0" y="534.0" >
			< point x="159.0" y="534.0" >
			< point x="177.0" y="528.0" >
			< point x="193.0" y="512.0" >
			< point x="198.0" y="498.0" >
			< point x="197.0" y="461.0" >
			< point x="188.0" y="445.0" >
			< point x="177.0" y="435.0" >
			< point x="154.0" y="429.0" >
			< point x="33.0" y="429.0" >
			< point x="26.0" y="435.0" >
			< point x="26.0" y="528.0" >
		< /coords >
	< /page >
	< page page_id="p01c06.dxf" >
		< coords >
			< point x="74.0" y="1677.0" > 
			< point x="76.0" y="1681.0" >
			< point x="212.0" y="1682.0" >
			< point x="225.0" y="1677.0" >
			< point x="241.0" y="1661.0" >
			< point x="246.0" y="1646.0" >
			< point x="245.0" y="1610.0" >
			< point x="237.0" y="1594.0" >
			< point x="225.0" y="1584.0" >
			< point x="202.0" y="1577.0" >
			< point x="81.0" y="1577.0" >
			< point x="74.0" y="1584.0" >
			< point x="74.0" y="1677.0" >
		< /coords >
	< /page >

Download or Browse the FPLAN-POLY database

All the data is publicly available and can be downloaded. We invite you to use these vectorial images and the models for research in symbol spotting (for scientific, non-commercial purposes only). In this case, we would appreciate acknowledgement of the source. Please make the reference to the following paper
  • Relational Indexing of Vectorial Primitives for Symbol Spotting in Line-Drawing Images.
    M. Rusiñol, A. Borràs and J. Lladós. Pattern Recognition Letters, 31(3):188-201, 2010.
    [ Paper]   [ 10.1016/j.patrec.2009.10.002 ]

You can browse the ground-truthed database without the need of having any CAD software nor parse the ground-truth files.

Download the complete floorplan database, the model symbols and the ground-truth.

Performance Evaluation

We have proposed a performance evaluation protocol for symbol spotting methods which can be used with this collection. The details of this protocol can be found in
  • A Performance Evaluation Protocol for Symbol Spotting Systems in Terms of Recognition and Location Indices.
    M. Rusiñol and J. Lladós. International Journal on Document Analysis and Recognition, 12(2):83-96, 2009.
    [ Paper]   [ 10.1007/s10032-009-0083-y]

This page is best viewed with Mozilla Firefox - Last Updated September 2009
© 2009 Marçal Rusiñol