ACETADA: Meal Images Enriched With Contextual Metadata

ACETADA is a public collection of 806 high-resolution \before-meal\ images captured with consumer smartphones in free-living settings. This dataset is the first of its kind to pair these free-living images with dietitian-verified nutrition labels (kilocalories, grams of protein, carbohydrates, fat, and edible-portion mass) and enriched with \contextual metadata\ in the form of precise GPS coordinates, timestamps, and fine-grained food-item annotations. Images are distributed as JPEG files; all annotations reside in a single CVS file that follows the Huggingface datasets schema for the possibility of one-line loading in Python. ACETADA enables controlled studies of nutrition and portion estimation, and metadata-aware prompting for Large Multimodal Models (LMM). The dataset is released under a CC-BY-4.0 licence for non-commercial use, with commercial licensing available upon request. ACETADA accompanies the manuscript \Evaluating Large Multimodal Models for Nutrition Analysis: A Benchmark Enriched with Contextual Metadata\ (IEEE JBHI, 2025).

ACETADA是一款公开数据集，收录了806张由消费级智能手机在自然生活场景下拍摄的高分辨率餐前（before-meal）图像。该数据集为首例将此类自然生活场景图像与营养师核验的营养标签（涵盖千卡、蛋白质克数、碳水化合物克数、脂肪克数及可食用部分质量）进行配对的数据集，并补充了以精准GPS坐标、时间戳和细粒度食物条目标注形式存在的上下文元数据（contextual metadata）。图像以JPEG格式分发；所有标注均存储在单个遵循Huggingface数据集规范（Huggingface datasets schema）的CVS文件中，支持Python一键加载。ACETADA可支撑营养与分量估计的对照研究，以及面向大型多模态模型（Large Multimodal Models, LMM）的元数据感知提示（metadata-aware prompting）相关研究。本数据集采用CC-BY-4.0许可证开源发布，仅允许非商业使用；商业授权可按需申请。ACETADA配套发表于2025年IEEE JBHI期刊的论文《Evaluating Large Multimodal Models for Nutrition Analysis: A Benchmark Enriched with Contextual Metadata》（面向营养分析的大型多模态模型评估：富集上下文元数据的基准数据集）。