Benchmarking Intelligent Agents for Visual Food Recognition Against Manual Logging
Primary Investigator (PI) Name
Maria Valero
Department
CCSE - Computer Science
Abstract
Accurate dietary tracking is crucial for nutrition management, but manual food logging can be time-consuming. This study benchmarks intelligent vision agents from Google, Anthropic, and OpenAI against MyFitnessPal user logs to assess their accuracy in food recognition and macronutrient estimation. We evaluate performance based on precision, recall, and mean absolute error (MAE) for calories, protein, carbohydrates, and fats.
Beyond accuracy, we analyze usability concerns, including ease of use, error correction, and adaptability to complex meals. Our findings highlight the strengths and limitations of AI-driven tracking compared to manual logging, providing insights into the future of automated nutrition assessment.
Disciplines
Computer Engineering | Nutrition
Benchmarking Intelligent Agents for Visual Food Recognition Against Manual Logging
Accurate dietary tracking is crucial for nutrition management, but manual food logging can be time-consuming. This study benchmarks intelligent vision agents from Google, Anthropic, and OpenAI against MyFitnessPal user logs to assess their accuracy in food recognition and macronutrient estimation. We evaluate performance based on precision, recall, and mean absolute error (MAE) for calories, protein, carbohydrates, and fats.
Beyond accuracy, we analyze usability concerns, including ease of use, error correction, and adaptability to complex meals. Our findings highlight the strengths and limitations of AI-driven tracking compared to manual logging, providing insights into the future of automated nutrition assessment.