/** ******************************************************************* /* KTH Object Database (KOD) - Stripped version * /* * /* http://www.csc.kth.se/~kootstra/kod * /* Updated: June 16, 2010 * /* * /**********************************************************************/ /** ************************* !!OBSERVE!! *****************************/ This is a stripped version of the database, only containing the following files (for description, see below): FRR.ppm, FRR_mask_.pgm, FD_CV.dat FD_CV.ppm The complete version can be downloaded from the website /** *******************************************************************/ 1. For feedback, errors and questions, contact Gert Kootstra (kootstra@csc.kth.se) 2. When using the database in a publication you will need to reference the following publication: [1] Kootstra, G., Bergstrom, N., & Kragic, D. (2010). Using Symmetry to Select Fixation Points for Segmentation. International Conference on Pattern Recognition (ICPR), August 23-26, 2010, Istanbul, Turkey. (To be presented) /** ***************** DESCRIPTION OF THE DATABASE *********************/ The purpose of the database is to provide an object database containing table top scenarios with more or less common household objects for attention and segmentation. The goal is to provide objects with differing complexity regarding shape and appearance, as well as scenes with different complexity regarding numbers of objects and object positions. In addition, different backgrounds are given (textured or non-textured) and all images are taken in two different lighting contitions. In total 25 different objects have been used. The database is separated into single-object scenes and multiple-object scenes. The naming convention is: ____ where is the name of the object(s) in the scene is one of {c1,c2,l1,l2,r1,r2} for single objects and {p1,p2,p3,p4} for multiple objects is one of {on,off} is one of {cloth,notexture} is described below The database was captured using an Armar III head, which has one foveal and one stereo camera pair, and contains the following images: --- extension --- | --- description --- FCal.txt | Foveal calibration* FD_CV.dat | Foveal disparity using OpenCV (float array) FD_CV.ppm | Foveal disparity using OpenCV (for visualization) FD_CV.txt | Foveal disparity using OpenCV (text) FL.ppm | Left foveal image FLR.ppm | Left rectified foveal image FPointCloud.txt | Point cloud from foveal image** FR.ppm | Right foveal image FRR_mask_.pgm | Hand segmented mask of object corresponding | to the right rectified foveal image FRR.ppm | Right rectified foveal image WCal.txt | Wide-field calibration* WD_CV.dat | Wide-field disparity using OpenCV (float array) WD_CV.ppm | Wide-field disparity using OpenCV (for visualization) WD_CV.txt | Wide-field disparity using OpenCV (text) WL.ppm | Left wide-field image WLR.ppm | Left rectified wide-field image WPointCloud.txt | Point cloud from wide-field image** WR.ppm | Right wide-field image WRR.ppm | Right rectified wide-field image * the calibration files are on the following format: First and second row correspond to left and right internal parameters: [0-8] - internal calibration matrix [9-12] - distortion parameters Third row: [0-2] - translation [3-6] - rotation (quaternions) [7-15] - *unused* [16-17] - size of images [18-35] - *unused* Fourth row contains information on how to tune the right image rectified with the above calibration (the tuning is done for the FRR and WRR images): [1] - x-difference [2] - y-difference [3] - rotation [4] - scaling ** The point cloud files are in the following format [0-2] - x,y,z, position [3-4] - corresponding pixel [5-7] - corresponding pixel RGB value