Development of an open-access food atlas with nutrient composition and recipes in Argentina: The Smart24Food Atlas

Introduction Accurate dietary intake assessment remains challenging, particularly for portion-size estimation, and context-specific photographic aids remain limited in Argentina. This study aimed to develop an open-access digital atlas of foods and recipes consumed in Argentina, integrating portion-size photographs and nutrient-composition data, and to describe its performance for food and recipe retrieval and visual portion-size matching under controlled direct-observation conditions. Methods The Smart24Food atlas was developed as a digital resource integrating food photographs, standardized recipes, and nutrient-composition information. A controlled direct-observation study in a university setting included 108 participants, mostly young adults, completing 20 tasks: seven packaged-product selection tasks and 13 visual portion-matching tasks. Exact matching was analyzed using cross-classified logistic mixed-effects models, while weight-based performance was assessed using linear mixed-effects models and repeated-measures Bland–Altman analysis. Results The atlas contains 3,273 photographs representing 998 distinct foods through 1,052 presentation-specific items, with nutrient-composition information for up to 53 components. Food or recipe retrieval was successful in 99.81% of tasks (2,156/2,160), and overall exact matching was 86.5% (1,869/2,160). Food-level heterogeneity was greater than participant-level heterogeneity, whereas food category and task type were not significantly associated with exact matching. Across 1,404 visual portion-matching observations, mean signed error was −1.34 g (95% CI: −4.68 to 2.00; p = 0.447), with no evidence that signed error varied systematically with served portion weight across the evaluated foods (B=−0.011 g/g; p = 0.703). The 95% Bland–Altman limits of agreement ranged from −33.44 to 30.76 g. Conclusions The Smart24Food atlas is an open-access resource integrating photographic portion references with nutrient-composition information for foods and recipes consumed in Argentina. Under controlled direct-observation conditions, participants showed high retrieval and exact-match performance, with no evidence of systematic mean bias in visual portion matching. Further studies in more diverse populations and routine dietary-assessment contexts are needed.