Table 1.

Challenges and research opportunities in adopting image-based sensing methods.

Challenge	Research opportunity
Lack of publicly available large-scale food image datasets with comprehensive ground truth labels	Develop feasible method to annotate food images crawled from the Web or collected from nutrition studies that can scale up
Inaccurate food image–segmentation algorithms	Reduce the burden of requiring fine-grain pixel-level training data for image segmentation and leverage accurate image or specific image region level information to improve food segmentation performance
Nonrobust food image–recognition systems	Design deep neural network–based models to capture structures in the image that are associated with specific foods and incorporate contextual information to improve robustness
Inaccurate food portion size estimation in image	Develop methods that can directly link food images to portion size; explore 3-dimensional information from newer camera sensors on mobile devices