The dataset VIPER-FoodNet (VFN) contains 82 categories of food images. Each food image has at least one bounding box indicating the location of the food item.
Every food category is the frequent intake food term selected from What We Eat In America (WWEIA) database. All of food images are online images uploaded by real users.
Version: VFN 1.0 Content: 82 categories, 14991 images, 22423 bounding boxes
The structure of archive file is described below:
|- <archive>.zip
|- Meta
|-annotations.txt
|-category_ids.txt
|-training.txt
|-validation.txt
|-testing.txt
|- Images
|-<image category>
|- <image id>.jpg
|- <image id>.jpg
|- ...
|-<image category>
|- ...
Description of structure:
File annotations.txt: Bounding box and food category information matching to image IDs.
File category_ids.txt: Food category information matching to food category IDs.
File train.txt, val.txt, test.txt: Image IDs corresponding to images in training/testing/validation sets.
Directory Images: Food images for training, validation and testing.
Directory <image category>: Food images of specific category
File <image id>.jpg: food image
Description of annotations.txt file:
Each line has 6 elements which are splited by one whilespace.
Each element from left to right is described below:
1. <image id>.jpg
2. x axis coordinate of upper-left corner of the bounding box
3. y axis coordinate of upper-left corner of the bounding box
4. x axis coordinate of lower-right corner of the bouding box
5. y axis coordinate of lower-right corner of the bouding box
6. Category ID of the bouding box
@article{mao2020, author="Mao, Runyu and He, Jiangpeng and Shao, Zeman and Yarlagadda, Sri Kalyan and Zhu, Fengqing", title="Visual Aware Hierarchy Based Food Recognition", journal="ICPR Workshops", year="2020"}