LAD stands for Large-scale Attribute Dataset. It is a large-scale image dataset with attribute annotations. It is divided into 5 major categories and 230 subcategories, with a total of 78,017 images. 359 visual, semantic, and subjective attributes are defined and annotated in the instance images. The number of images in the LAD dataset is larger than that of the four mainstream attribute datasets (AHr , CUB , aP/aY and SUN) combined.

This dataset can be used to test advanced zero-shot learning (ZSL) algorithms.