Get answers to common questions about FoodDX.
FoodDX is an AI-based model trained on a large dataset of foods scored by our nutrition experts. FoodDX mimics the nutrition experts food scoring by predicting what score a food will get based on what similar foods were rated by the human scorers. For this reason, FoodDX may be less accurate for uncommon foods or foods where the true score is highly variable based on ingredients, toppings, dressing type, and portion size. We continuously work to improve the model and regularly assess the accuracy with validation of scores by our dietitians.
Our team of dedicated nutrition experts built FoodDX upon proven concepts in behavioral and medical science. For more information on our nutrition experts, feel free to visit us here.
The food scoring accuracy is estimated to be 85%. This means that the model classifies images in the correct category (food vs. drink) and provides a food score within 1 point of what a human nutrition expert would have scored it, at least 85% of the time.
We are not available in all countries at the moment. We are currently working on expanding our services globally with partner organizations. If you would like to see FoodDX in your region, let us know by reaching out via firstname.lastname@example.org.
Ingredient switches are often not detectable by the human eye or in images, therefore it is difficult for an AI model to differentiate these differences as well. However, we are actively developing functionality for our users to flag photos or provide additional information.
At this time, food scores cannot be changed by users. We are developing functionality for our users to flag the images that feel are incorrectly scored for the purposes of improving the accuracy of our model.
We understand the need to provide scores for beverages and will consider incorporating this feature in the future. However, it is difficult to rate drinks based on the image alone due to the undetectable differences in the contents that impact nutritional value, such as sugar level, ingredients, food colouring, and so on.
Rating packaged foods is difficult because packaging tends to be similar for very different types of food (for example, cereal and laksa paste both being sold in cardboard boxes). FoodDX does not "read" labeling or text, but scores based on AI-trained detection of other image features.
Currently, FoodDX does not have the ability to allow users to enter additional details about their food, such as when healthier alternatives are used. We are developing functionality for users to flag images or provide additional information for the purposes of improving the accuracy of our model.
We apologise for the inconvenience caused. For better understanding, please contact the in-app customer service centre. Do share the technical issues you are currently facing and the customer service team will reach out to you for further support.