Note that this dataset can be used only for non-commercial research purpose. 

If you publish a paper using our segmented food dataset, we'd glad if you could refer to the following paper:

****** 

Battini Sönmez, E., Memiş, S., Arslan, B. et al. The segmented UEC Food-100 dataset with benchmark experiment on food detection. Multimedia Systems (2023). https://doi.org/10.1007/s00530-023-01088-9  

******

This ZIP file contains the following files:

File Structure

--UECFOOD-100 Segmented/
-- class_file/
    -- ann/
        -- ann/  .json    //annotation for images
          -- img/
              -- img/  .jpg    //all images 
          -- masks_human/
              -- mask_human/  .png    //coloured human masks for all images
          -- masks_machine/
              -- mask_machine/  .png   //image masks 
-- obj_class_to_machine_color.json

In “ann”, annotations are saved as “json”
    - 0.jpg.json
          - 1.jpg.json
          - 2.jpg.json
In “img”, real train and test images are saved as “jpg”
          - 0.jpg
          - 1.jpg
          - 2.jpg
In “masks_human”, real image and mask comparisons are saved as “png”
          - 0.png
          - 1.png
          - 2.png
In “masks_machine”, mask images are saved as “png” (this is what we used)
          - 0.png
          - 1.png
          - 2.png

In “obj_class_to_machine_color.json”, labels and the rgb color distribution is stored. 
          "rice": [1,1,1],
      "fried rice": [2,2,2],
      "teriyaki grilled fish": [3,3,3]
… 
User can select any of the annotation selection for their training process whichever suits them. 
(annotations or machine_masks to extract annotated areas of the images)

Acknowledgements

Authors would like to thank to following undergraduate students for their contributions on manually segmenting the database:
İlteriş Mete Akdoğan, Melike Özlem Akmandor, Deniz Naz Demirtaş, Selen Fem Güngördü, Tuğçe Yücel, Ecem Suzan Ulaş