Objective: To improve the accuracy of food intelligent recognition and weight estimation technology, establish a large-scale food image dataset.
Methods: Building large-scale food image and ingredient datasets based on web crawler technology, professional manual collection, and regular user uploads.
Results: A big dataset was constructed containing over 1.15 million annotated images and 2356 categories of food and dish ingredients, including information such as name, dish category, weight, images, nutritional content, cooking method, and region. The dataset includes 12 categories and 73 subcategories. The 12 categories include vegetarian dishes, meat dishes, meat and vegetable dishes, staple food, porridge, soup, snacks and desserts, milk and dairy products, fruits, nuts, beverages and food raw materials. And all data has undergone strict data cleaning professional inspection, and iterative labeling.
Conclusion: The largest-scale food image dataset currently used for intelligent recognition, providing a solid data foundation for intelligent recognition of food images.
Keywords: dietary database; food image; intelligent identification; web crawler.