NutritionVerse Direct
NutritionVerse is a research initiative focused on developing accurate and efficient methods for estimating dietary intake using computer vision. Current efforts center on creating and utilizing large-scale datasets of 2D and 3D food images (NutritionVerse-Real, NutritionVerse-Synth, NutritionVerse-3D) to train and evaluate deep neural networks, including vision transformer architectures, for directly predicting nutritional content from food images. This work aims to improve the accuracy of dietary assessment, particularly for vulnerable populations like the elderly, by automating a process currently reliant on error-prone self-reporting methods. The open-source nature of the datasets and associated tools is accelerating progress in this field.
Papers
NutritionVerse-Thin: An Optimized Strategy for Enabling Improved Rendering of 3D Thin Food Models
Chi-en Amy Tai, Jason Li, Sriram Kumar, Saeejith Nair, Yuhao Chen, Pengcheng Xi, Alexander Wong
NutritionVerse-3D: A 3D Food Model Dataset for Nutritional Intake Estimation
Chi-en Amy Tai, Matthew Keller, Mattie Kerrigan, Yuhao Chen, Saeejith Nair, Pengcheng Xi, Alexander Wong