The BSDS500 dataset is usually seperated into a training set of 432 images and a testing set of 68 images. Note that in the original BSD paper, the test set of 68 images were not defined. Rather, it seems that these 68 test images were first introduced by Roth and Black [1], where the test image IDs can be found here. In this repository, we collect the colored version of the training images and testing images. The 432 training images are collected from here and the testing images are collected from here.
[1] Stefan Roth and Michael J. Black. Fields of experts: A framework for learning image priors. InProceedingsof the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, volume 2, pages860–867. IEEE, 2005