TieredImageNet is a dataset for small sample classification tasks. It is a larger subset of ILSVRC-12 (Image Net Large Scale Visual Recognition Challenge), with 608 categories and 779,165 images. The categories of this dataset are more extensive and correspond to higher-level nodes in the ImageNet hierarchy, with a total of 34 major categories, including 20 training classes, 6 validation classes, and 8 test classes.

The dataset has a root-split hierarchy similar to ImageNet, which results in a more challenging but more realistic system.

Example image: