YorkUrbanDB - README Introduction The York Urban Database is a compilation of 102 images of urban environments consisting mostly of scenes from the campus of York University and downtown Toronto, Canada. The images are 640 x 480 in size and have been taken with a Panasonic Lumix DMC-LC80 digital camera. This public database is further broken down into 45 indoor and 57 outdoor scenes which conform to the "Manhattan Assumption", that is, the overlying structure in the scene conforms to a rectangular 3-D grid (right angle structure) (Coughlan and Yuille, 2001). Further details can be found in the ECCV conference paper (Denis, Elder & Estrada 2008), and in Patrick's MSc thesis. If you use this database, please cite the ECCV conference paper. References Coughlan, J. M., & Yuille, A. L. (2003). Manhattan world: Orientation and outlier detection by Bayesian inference. Neural Computation, 15 (5), 1063 - 1088. Denis, P., Elder, J.H. & Estrada, F. (2008). Efficient Edge-Based Methods for Estimating Manhattan Frames in Urban Imagery. Proc. European Conference on Computer Vision (5303), 197-211. Denis, P. (2008). Efficient Edge-Based Methods for Estimating Manhattan Frames in Urban Imagery. M.Sc. Thesis, York University, Canada. Organization In the main path there are 3 files, Manhattan_Image_DB_Names.mat, cameraParameters.mat and ECCV_TrainingAndTestImageNumbers.mat. 1. Manhattan_Image_DB_Names.mat PURPOSE: Contains the MATLAB filenames of all the images in the database. CONTENT: Manhattan_Image_DB_Names - [102,1] cell array. *********************************************************************** *********************************************************************** 2. cameraParameters.mat PURPOSE: Internal camera parameters used to compute the Manhattan frame for each image. See the ECCV paper and Patrick’s MSc thesis for details on the calibration methods. CONTENT: focal - [1] focal length of camera (mm). pixelSize - [1] The estimated size of the sensor pixels, in mm. pp – [1,2] Principal point in pixel coordinates. The principal point is stored in (x, y) coordinates where the origin (1, 1) is the TOP LEFT of the image. *********************************************************************** *********************************************************************** 3. ECCV_TrainingAndTestImageNumbers.mat PURPOSE: This file lists the randomly-selected partition of images into training and test sets used in Denis, Elder & Estrada, ECCV 2008. This may be useful if you wish to compare new algorithms directly against the results reported in this paper. CONTENT: trainingSetIndex – [51,1] Image numbers used for training set testSetIndex – [51,1] Image numbers used for test set *********************************************************************** *********************************************************************** Information for each of the 102 images in the database is stored in a separate folder. Each of these folders contains: 1. 2. 3. 4. <imageName>.jpg <ImageName>LinesAndVP.mat <ImageName>GroundTruthVP_CamParams.mat <ImageName>GroundTruthVP_Orthogonal_CamParams.mat 4. <imageName>.jpg PURPOSE: The original 640x480 image. *********************************************************************** *********************************************************************** 5. <ImageName>LinesAndVP.mat PURPOSE: Holds the MATLAB data pertaining to the labeled line segments and vanishing point association. CONTENT: vp_association - [n,1] Vanishing point associated with each of the n line segments. values are {1,2,3}. Convention used: Possible 1 – First horizontal vanishing point 2 - Vertical vanishing point 3 - Second horizontal vanishing point lines - [2n,2] Each line segment is represented by its endpoints, stored as a 2x2 array [x1 y1; x2 y2]. These are concatenated vertically, so that the kth line segment is accessed as lines(2*k-1:2*k,:). finalImg Image with especially accurately [480,640,3] the line segments overlaid. short ones, are not drawn due to aliasing. Please note that some lines, *********************************************************************** *********************************************************************** 6. <ImageName>GroundTruthVP_CamParams.mat PURPOSE: This file contains the vanishing point vectors independently estimated in the Gauss Sphere and estimated standard error. Since they were estimated independently, these vectors are not exactly orthogonal. CONTENT: vp - [3,3] matrix containing the three vanishing point unit vectors stored in column format. vp_std_error - [1,3] the standard error computed for each vector. *********************************************************************** *********************************************************************** 7. <ImageName>GroundTruthVP_Orthogonal_CamParams.mat PURPOSE: This file contains orthogonalized estimates of the vanishing point vectors. CONTENT: vp_orthogonal - [3,3] matrix containing the three orthonormal vanishing point vectors stored in column format.