Related Experiment Video
Updated: May 8, 2025
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
8.9K
[Building a food image dataset based on intelligent recognition and weight estimation]
Weiyan Gong1, Fan Yuan1, Caicui Ding1
1National Institute for Nutrition and Health, Chinese Center for Disease Control and Prevention, Beijing 100050, China.
Wei Sheng Yan Jiu = Journal of Hygiene Research
|December 24, 2024
Summary
Researchers developed a large-scale food image dataset to enhance food recognition and weight estimation accuracy. This dataset, featuring over 1.15 million images across 2356 categories, supports advancements in intelligent food analysis.
Area of Science:
- Computer Vision
- Machine Learning
- Food Science
Background:
- Accurate food recognition and weight estimation are crucial for dietary assessment and intelligent systems.
- Existing food image datasets often lack scale, diversity, and comprehensive annotations.
Purpose of the Study:
- To establish a large-scale, meticulously annotated food image dataset.
- To improve the accuracy of food intelligent recognition and weight estimation technologies.
Main Methods:
- Web crawling, professional manual collection, and user uploads were employed to gather data.
- Data collection focused on diverse food items, ingredients, and dishes.
- Rigorous data cleaning, professional inspection, and iterative labeling were performed.
Main Results:
- A dataset comprising over 1.15 million annotated images and 2356 food/ingredient categories was created.
- The dataset includes detailed information: name, category, weight, images, nutrition, cooking method, and region.
- It encompasses 12 main categories and 73 subcategories, covering a wide range of food types.
Conclusions:
- This represents the largest-scale food image dataset for intelligent recognition to date.
- The dataset provides a robust foundation for developing and refining intelligent food image analysis systems.
