metadata

pretty_name: Snacks
task_categories:
  - image-classification
  - object-detection
  - computer-vision
license: cc-by-4.0

Dataset Card for Snacks

Dataset Summary

This is a dataset of 20 different types of snack foods that accompanies the book Machine Learning by Tutorials.

The images were taken from the Google Open Images dataset, release 2017_11.

Dataset Structure

Number of images in the train/validation/test splits:

train    4838
val      955
test     952
total    6745

Total images in each category:

apple         350
banana        350
cake          349
candy         349
carrot        349
cookie        349
doughnut      350
grape         350
hot dog       350
ice cream     350
juice         350
muffin        348
orange        349
pineapple     340
popcorn       260
pretzel       204
salad         350
strawberry    348
waffle        350
watermelon    350

To save space in the download, the images were resized so that their smallest side is 256 pixels. All EXIF information was removed.

Annotations

Included in the annotations folder are three CSV files with bounding box annotations for the images in the dataset, although not all images have annotations and some images have multiple annotations.

The columns in the CSV files are:

image_id: the filename of the image without the .jpg extension
x_min, x_max, y_min, y_max: normalized bounding box coordinates, i.e. in the range [0, 1]
class_name: the class that belongs to the bounding box
folder: the class that belongs to the image as a whole, which is also the name of the folder that contains the image

Data Splits

Train, Test, Validation

Licensing Information

Just like the images from Google Open Images, the snacks dataset is licensed under the terms of the Creative Commons license.

The images are listed as having a CC BY 2.0 license.

The annotations are licensed by Google Inc. under a CC BY 4.0 license.

The credits.csv file contains the original URL, author information and license for each image.