pic2kcal
收藏资源简介:
pic2kcal数据集由德国卡尔斯鲁厄理工学院的人类学与机器人研究所创建,包含超过70,000个食谱和308,000张食物图片,涵盖从简单沙拉到蛋糕、比萨和汤等多种菜肴。数据集通过匹配食谱中的原料与食品数据库中的营养信息来生成精确的热量估计。创建过程涉及使用语义嵌入技术处理原料描述,以确保与营养数据库的准确匹配。该数据集主要应用于食物热量预测,通过多任务学习方法提高热量估计的准确性,旨在帮助用户更准确地跟踪和控制饮食,促进健康生活方式。
The pic2kcal dataset was developed by the Institute of Anthropology and Robotics at Karlsruhe Institute of Technology (KIT), Germany. It encompasses over 70,000 recipes and 308,000 food images, spanning a diverse array of dishes ranging from simple salads to cakes, pizzas, and soups. Precise calorie estimates are generated within this dataset by aligning ingredient lists from the recipes with nutritional data sourced from food databases. The development pipeline utilized semantic embedding techniques to process ingredient descriptions, ensuring accurate matching with the underlying nutritional databases. The primary application of this dataset lies in food calorie prediction, where multi-task learning approaches are employed to enhance the accuracy of calorie estimation. It is designed to assist users in more precisely tracking and regulating their dietary intake, thereby facilitating the adoption of healthy lifestyles.

- 1Multi-Task Learning for Calorie Prediction on a Novel Large-Scale Recipe Dataset Enriched with Nutritional Information人类学与机器人研究所,卡尔斯鲁厄理工学院,德国 · 2020年



