Binary Coherent Edge Descriptors

C. Lawrence Zitnick





Binary Coherent Edge Descriptors

C. L. Zitnick

ECCV, 2010.



An exe can be found here.




The paper uses two different types of datasets.  The first datasets test invariance to illumination and pose changes using images taken of the Statue of Liberty and Notre Dame.  These dataset are provided by Simon Winder and Matt Brown here.


Description: \\research\root\web\external\en-us\UM\People\larryz\BiCE_files\image001.jpg

Category patch dataset


The second dataset measures invariance to intra-category variation.  The dataset contains a training set and testing set of image patch pairs.  A pair is said to “match” if they contain the same object part, such as the front wheel of a motorcycle or a person’s eye.  These patches were selected by hand from a subset of categories in the Caltech 101 dataset.  The data can be found here: (557 MB) 


Please read the accompanying Readme.txt file for the data’s format.


The patches grouped by category that were used to create the above patch pairs dataset can be found here: (101 MB)


When using the category patch dataset, please cite the paper above.


Any questions should be directed to Larry Zitnick (larryz at microsoft dot com).