The dataset contains natural images of grocery items. All natural images were taken with a smartphone camera in different grocery stores. Total 5125 natural images from 81 different classes of fruits, vegetables, and carton items (e.g. juice, milk, yoghurt). The 81 classes are divided into 42 coarse-grained classes, where e.g. the fine-grained classes 'Royal Gala' and 'Granny Smith' belong to the same coarse-grained class 'Apple'. For each fine-grained class, downloaded an iconic image and a product description of the item, where some samples of these can be seen on this page below. The dataset was presented in the paper "A Hierarchical Grocery Store Image Dataset with Visual and Semantic Labels", which appeared at WACV 2019.

How to use the dataset

The files train.txt, val.txt, and test.txt in the folder dataset include the paths to the images in the training, validation, and test set respectively. Each row in these two files consists of the path to an image and its fine-grained label followed by its coarse-grained label, where both labels are represented as integers.

The 81 fine-grained classes and their coarse-grained classes can be found in classes.csv in the folder dataset. The classes corresponding label (an integer) is also included in addition to the paths to their iconic image and the product description.

Feel free to download the dataset and apply it to your model. When time allows, I will upload code that is related to the paper above, such that the results in the paper can be reproduced.

License: MIT License