The BeiHang Brake Cylinder (BH-BC) Database

We introduce a challenging data set, namely the “BeiHang Brake Cylinder (BH-BC) Database”. In the proposed evaluation protocol, two kinds of data sets are composed from the BH-BC database: one training set and two query (probe) sets, and all the images that appear in the training set are excluded from the probe sets. The training set contains 2006 120x180 brake cylinder images and 6944 120x180 negative samples cropped from a varied set of train chassis images taken along railways in real application which makes the database more comprehensive to object detection. Two probe sets, namely D1, D2, are constructed. D1 consists of 283 images which are taken under relatively friendly conditions. D2 containing 180 images is much more challenging, for the images mainly focus on illumination and scale changes, where the targets and backgrounds are almost indistinguishable. The partitions can be used to address the performance variations associated with the changes in the probe set and to identify the strengths and weaknesses of a particular algorithm.

you may download the whole database by clicking here (224MB).

Related Publication:

(to be added)


CopyRight reserved
Designed by Machine Learning Group