Broden

The dataset used in the paper is Broden, a dataset containing pixel-level concept annotations.

BibTex: