Bento800🍱

🎑 News

• [23/03/06] Bento800🍱_Text📝 is released.
• [22/10/22] Paper and Bento800🍱 (Version1.0) are released.
• [22/09/13] Add Bento800 creation process.
• [22/08/04] This repo is created.

🍀 Overview

Bento800 is the first manually annotated synthetic box lunch presentation dataset for novel aesthetic box lunch presentation design.

Bento800 is composited from 34 single food images in 6 categories:

Rice🍚(10) ; Fried Chicken🍗(5) ; Salt-grilled Salmon🐟(5) ; Tamagoyaki🧈(6) ; Croquette🍪(5) ; Fried Shrimp🍤(3)

with 3 different types of food presentations:

(1) Place fried chicken on rice;

(2) Place salt-grilled salmon and tamagoyaki on rice;

(3) Place croquette and fried shrimp on rice, the fried shrimp is on the croquette.

🌱 Dataset Creation Process

Bento800🍱_Text📝

• Image Details

Content	Tools	Details
Resize	-	Paste images in Bento800 onto a transparent image.

• Text Description Details

Content	Tools	Details
Text Paraphrasing	ChatGPT	Rephrase the above sentences with the template “Rephrase: [Input] with 25 examples” and “Rephrase: [Input]"+"Please try again"×24.

Bento800🍱

• Image Details

Content	Tools	Details
Image Collection	Google Image	These high-resolution images contain at least one complete food item.
Remove Background	removebg	Remove the background and unnecessary parts.
Image Compositing	-	Manually composite + random data augmentation.

• Annotation Details

Content	Tools	Details
Modal Segmentation	Dataloop	Manually marked visible parts and occluded parts.
Food Item Labels & Bounding Box & Ordering List	VGG Image Annotator (VIA)	Manually annotated several food item attributes.

🎁 Download

Content	Size	Files	Format	Details
Bento800🍱_Text📝 [Google Drive] [Tencent Cloud]	-	1600		Main Folder
├  Image	216MB	800	PNG	Food images from Bento800 of size 600×600
├  Text	793 KB	800	TXT	9 descriptions for each image in Bento800

Content	Size	Files	Format	Details
Bento800🍱v1.0 [Google Drive]	-	872		Main Folder
├  S1_collection	3.26 MB	34	JPG	34 food images in 6 categories
├  S2_removebg	2.81 MB	34	PNG	Remove Background
├  S3_train	410 MB	766	PNG	Training images
├  S3_val	18.8 MB	34	PNG	Testing images
├  S4_segmentation	8.72 MB	2	JSON	Semantic Segmentation
├  S5_annotation	403 KB	2	CSV	Simple text descriptions + Ordering id + Bounding Box + Food item labels

📚 Feedback

Please fill out the Bento800 Dataset Feedback Form.

I would greatly value your thoughts, suggestions, concerns or problems.

⭐ Citation

“We eat🍴 first with our eyes👀~ ”

If you find this dataset helpful for your research, please cite it as below:

@inproceedings{zhou2022able,
  title={ABLE: Aesthetic Box Lunch Editing},
  author={Zhou, Yutong and Shimada, Nobutaka},
  booktitle={Proceedings of the 1st International Workshop on Multimedia for Cooking, Eating, and related APPlications},
  pages={53--56},
  year={2022}
}