This data set contains 30 floormaps of libraries and corresponding ground truth as mentioned in the following paper:

Hima Bindu Maguluri, Qiongjie Tian, Baoxin Li, "Detecting Text in Floor Maps using Histogram of Oriented Gradients", in proceedings of the IEEE, International Conference on Acoustics, Speech, and Signal Processing (ICASSP), May 2013. 

The folder 'FloormapImages' contains JPEG images of the floor maps.

Note: Though the naming of images varies from 1 to 40, there are only 30 images. All the numbers in between 1 and 40 have not been used for naming.

The folder 'GtMask' contains marked ground truth in the form of mask images for all the floor maps.

The folder 'GtText' contains ground truth in the form coordinates of rectangles and text content inside each rectangle. In each text file, 'x1','y1' columns contain coordinates of top left corner and 'x2','y2' columns contain coordinates of bottom right corner of rectangles. The 'content' column contains text in each rectangle. The origin of the coordinate systme is top left corner.

The folder 'UpdatedGtMask' contains mask images corresponding to conditioned labels for all the maps. Refer the paper for details about conditioned labels and data conditioning. 

Note: In cases where text occurs in very poor quality, corresponding words have not been marked in the ground truth. Ex: The words 'Foyer' and 'Telephones' that  occur in bottom middle of image '40.jpg'. The reason for eliminating such words is even if detected, such text cannot be read by OCR.

If you use this data set, please cite the article mentioned above.

This material may not be redistributed.