NutritionVerse-Real
收藏资源简介:
NutritionVerse-Real是由滑铁卢大学创建的一个开放访问的手动收集2D食品场景数据集,旨在用于膳食摄入量估计。该数据集包含889张图像,涵盖251种不同的菜肴和45种独特的食品类型。数据集的创建过程包括在现实生活中手动收集食品场景图像,使用食品秤测量每种成分的重量,并利用食品包装或加拿大营养文件中的营养信息计算每道菜的营养成分。通过人工标注图像生成分割掩码。该数据集主要应用于膳食感知领域的机器学习模型开发,以解决膳食摄入量估计中的偏差问题。
NutritionVerse-Real is an open-access manually collected 2D food scene dataset developed by the University of Waterloo, intended for dietary intake estimation. This dataset comprises 889 images, covering 251 distinct dishes and 45 unique food categories. The dataset construction workflow includes manually collecting food scene images in real-world settings, measuring the weight of each ingredient with food scales, and calculating the nutritional content of each dish using nutritional information from food packaging or Canadian nutrition documents. Segmentation masks are generated through manual image annotation. This dataset is primarily utilized for developing machine learning models in the field of dietary perception, aiming to mitigate bias issues in dietary intake estimation.

- 1NutritionVerse-Real: An Open Access Manually Collected 2D Food Scene Dataset for Dietary Intake Estimation滑铁卢大学 · 2023年



