Computational Neuroscience Lab at Mind/Brain Institute, Johns Hopkins University

Code for Edge of Stability: Response Times and Delta Oscillations in Balanced Networks (PLoS Comput. Biol.) 

File: [Download]

Database of figure and ground patches 

Visual scenes consist of objects which are closer to us than their backgrounds. The background is occluded by foreground objects and the boundary between them is referred to as occlusion boundary (OB). Determining which side of an occlusion boundary is figure (closer to the observer) and which is ground (further away from the observer) is called Figure/Ground Organization (FGO). FGO is made through a combination of global cues, like convexity, and local cues, like T-junctions. 

An advantage of studying local cues is that FGO can be done by analyzing only a small part of the entire scene, hence can be fast and computationally more efficient. For people interested in studying the local cues for FGO, we provide a database of patches extracted along the OB. A pair of patches consists of small (16 x 16 pixels) figure and ground patches. We provide two databases of such figure-ground patch pairs, where one set of patches were extracted from the BSDS300 database of images and the other set from a subset of MIT LabelMe images. All patches have been rotated by appropriate angles such that it becomes easier for further analysis (see Figure). 


BSDS300 patch databse consists of 1475 patch pairs, whereas LabelMe patch database consists of 1761 patch pairs. Each figure and ground patch is a matrix of 8-bit, grayscale converted pixel intensities of size 16x16 pixels. Both databases are formatted as .mat files, and can be opened using MATLAB software. The MATLAB software can be obtained from here

The BSDS patch database is 745.5 kB in size. Patches were extracted from all 300 images in the BSDS300 database - training (200 images) and test (100 images) combined. For this database, images have not been included, but can be downloaded from the BSDS website. The details of the database are given below:

File: BSDS.mat [Download]

Description: A 1x1475 cell array of 'struct's (structured array)with the following fields:


The LabelMe patch database is 2.2 GB in size. Patches were extracted from 585 LabelMe images of different scene categories such as Indoor, Beach, Office, Street etc,. For this database, 8-bit grayscale converted images have been included. The details of the database are given below:

File: LabelMe.mat [Download]

Description: A 1x1761 cell array of 'struct's (structured array)with the following fields:

The details of databases and patch extraction can be found in the following papers:

  1. Ramenahalli, Sudarshan, Stefan Mihalas, and Ernst Niebur. "Extremal edges: Evidence in natural images." Information Sciences and Systems (CISS), 2011 45th Annual Conference on. IEEE, 2011.
  2. Ramenahalli, Sudarshan, Stefan Mihalas, and Ernst Niebur. "Figure-ground classification based on spectral properties of boundary image patches."Information Sciences and Systems (CISS), 2012 46th Annual Conference on. IEEE, 2012.