Hierachical Labels for CAPG-GP dataset

CAPG Grocery Product (CAPG-GP) is a grocery product dataset with 102 fine-grained classes. We organize these 102 classes into five categories based on brands to create hierarchical labels. The original CAPG-GP dataset was first published in 'Fine-Grained Grocery Product Recognition by One-Shot Learning' by Geng et al. in 2018. The original dataset is publicly available at http://zju-capg.org/capg-gp.html. The training dataset consists of 107 single-product images taken under ideal laboratory conditions, while the testing dataset contains 234 shelf images from two different stores, each showing multiple products captured with a mobile device. Our capg_labels.txt file contains lines where each line consists of a brand_id, instance_class_id, and at least one image_filename.