The UK Bench dataset from Henrik Stewenius and David Nister contains 10200 images of N=2550 groups with each four images at size 640x480. The images are rotated, blurred and have a tendancy for computer science motives. The original paper is based on two subsets of 1400 and 6376, due to lack of images and efficient implementations. The dataset is typically used for image retrieval, where one image of a group is used as query and the score how many of the other images are in the top-4 rank.

Related datasets