Nutrition5k

该数据集名为Nutrition5k，包含3,493张具有顶视深度图数据的不同食物图片，这些图片中的食物由多种食材组成。在本次研究中，选取了一个包含909张图片的子集，重点关注那些含有不超过3种食物成分的图片，并排除了那些具有不正确地面真实数据（ground-truth data）的图片。在该子集中，能量值范围从10千卡到1,000千卡。训练集和测试集的比例为80:20，分别对应731张训练图片和175张测试图片。该子集共包含909张图片，其任务是进行食物能量估计。

This dataset is named Nutrition5k, which consists of 3,493 distinct food images with top-view depth map data, where each food item in the images is composed of multiple ingredients. In this study, a subset of 909 images was selected: we focused on images containing no more than 3 food ingredients, and excluded those with incorrect ground-truth data. The energy values of the food items in this subset range from 10 kcal to 1,000 kcal. The training and test sets are split at an 80:20 ratio, with 731 images allocated to the training set and 175 to the test set. This subset, totaling 909 images, is designed for the task of food energy estimation.

搜集汇总

背景与挑战

背景概述

Nutrition5k是一个包含约5,000盘食物视觉和营养数据的综合数据集，每盘食物提供多角度视频、RGB-D图像、详细成分列表及营养信息。数据集还包含官方训练/测试分割和评估脚本，适用于视觉营养理解研究。