This app is based on ResearchKit to survey the food preferences of the user (and potentially ask the user to complete other psychological surveys), for use in psychology or consumer studies such as those studying diet choice and attitudes towards different foods. For the beta-version of this app, the various functions are exposed in the main navigation window (e.g. overview of the study, consent form, joining/withdrawing from the study, completing surveys) -- in the final app, we plan to make the onboarding to the study and the flow of the app more integrated and structured.
Surveys are encoded as YAML files in the app bundle. Food preference is assessed in surveys that ask the user to select among 2-choice food images (from a large collection of food images included in the app) or by rating a single food image on a linear scale.
As a ResearchKit App, we anticipate that any specific studies undertaken with the app would undergo Human Subjects review and approval. The app also accesses HealthKit info (if the user has granted permission, of course), so that for example food preferences might be correlated with daily step levels.
Account login is not required to participate in the surveys of the beta app; ultimately users would enroll by providing their name, email and phone number as part of the informed consent onboarding. As described in the consent document, email and SMS via phone number would be used to receive study information and to request the user to take scheduled surveys (e.g. via custom URLs to launch the app from a SMS text).
The app also includes the Google Firebase framework to store survey results in a cloud database.
Presentation of a pair of food images, e.g. images from a database such as Food-pics database{Blechert:2014kx} (http://www.eat.sbg.ac.at/resources/food-pics) or the Food4Health image collection {Charbonnier:2016fa} (http://nutritionalneuroscience.eu)
A custom ImagePreferenceTask, question step, and answer format were developed to evaluate food image preferences in a series of forced choice questions.
The ImagePreferenceTask accepts the name of a bundled folder of image files (numbered sequentially, e.g. "1.jpg", "2.jpg", etc.). The task can be specified to present a certain number of random pairs from the image folder, or a pre-determined set of pairs can be passed to the task as an array of image number pairs.
During the task, each question step presents a pair of images and allows the selection of one image as the preferred option in a forced choice. (An optional flag also permits the presentation of a "no preference" button.) The subject's preference (e.g. "2>1" or "2==1" if "no preference" is allowed), along with the latency to make a choice, is recorded.
Alternatively, a complete matrix of preferences for a subject could be based on more-or-less obvious attributes across the assessed subset of food images (e.g. preference of "salty" food images vs "sweet food" images.) In this case, it may be helpful to present images of food "baskets" containing collections of emblematic foods, such as an ensemble of salty snack foods vs. dairy foods vs. vegetables, and so on.
De Bruijn,S.E.M. , Y.C. De vries, C. De Graaf, S. Boesveldt, G. Jager The reliability and validity of the Macronutrient and Taste Preferences Ranking Task: A new method to measure food preferences. Food quality and preference, 2017, vol 57, pp, 32-40.
Charbonnier,L. , F. van Meer, L.N. van der Laan, M.A. Viergever, P.A.M. Smeets, "Standardized food images: A photographing protocol and image database", Appetite, 2016, vol. 96, pp. 166-173.
As an alternative to the forced choice food preference task, we also implemented a custom question step for the subject to rate their response to a single food image. Images are again presented from a bundled folder of image files, but this time one at a time, along with a vertical slider scale for rating.
As options for rating scales, the food rating answer format can be set to to use the Labeled Hedonic Scale (LHS) developed by Lim et al. {Lim:2009du} or the older Natick 9-point hedonic scale{Jones:1955vb}.
A YAML schema was developed to encode common psychological surveys, in order to facilitate their re-use in ResearchKit.
Images databases available for academic research
Open Library of Affective Foods (OLAF) (96 images)
"Designed to allow simultaneous use with affective images from the International Affective Picture System (IAPS), which is a well-known instrument to investigate emotional reactions in the laboratory."
{Miccoli:2014jq}
Food4Health (80 images)
{Charbonnier:2016fa}
Food-pics database (568 images)
"A picture database comprising 568 food images and 315 non-food images along with detailed meta-data. A total of N = 1988 individuals with large variance in age and weight from German speaking countries and North America provided normative ratings of valence, arousal, palatability, desire to eat, recognizability and visual complexity. Furthermore, data on macronutrients (g), energy density (kcal), and physical image characteristics (color composition, contrast, brightness, size, complexity) are provided."
{Blechert:2014kx}
Foodcast Research Image Database (FRIDa)
"The database comprises open-source images belonging to one of eight different categories: (1) natural-food, (2) transformed-food, (3) rotten-food, (4) natural-nonfood items, (5) artificial food- related objects, (6) artificial objects, (7) animals, and (8) scenes."
{Foroni:2013ff}
Image-Net (1001K images)
"ImageNet is an image database organized according to the WordNet hierarchy (currently only the nouns), in which each node of the hierarchy is depicted by hundreds and thousands of images." As of 2010, there were 1495 food synsets, with an average of 670 images per set. URLs freely available.
{Deng:2009td}
Food-11 (16643 images)
"This is a dataset containing 16643 food images grouped in 11 major food categories. The 11 categories are Bread, Dairy product, Dessert, Egg, Fried food, Meat, Noodles/Pasta, Rice, Seafood, Soup, and Vegetable/Fruit. Similar as Food-5K dataset, the whole dataset is divided in three parts: training, validation and evaluation. The same naming convention is used, where ID 0-10 refers to the 11 food categories respectively." See also Food-5K, with 2500 food and 2500 non-food images for classification training.
ETH Food-101 (101000 images)
"a challenging data set of 101 food categories, with 101,000 images. For each class, 250 manually reviewed test images are provided as well as 750 training images. On purpose, the training images were not cleaned, and thus still contain some amount of noise. This comes mostly in the form of intense colors and sometimes wrong labels. All images were rescaled to have a maximum side length of 512 pixels."
Pittsburgh Fast Food Image Database (4545 images)
Amsterdam Beverage Picture Set (ABPS)
{Pronk:2015ge}
Geneva Appetitive Alcohol Pictures (GAAP)
International Affective Picture System (IAPS)
American Alcohol Photo Stimuli (AAPS)
The DIsgust-RelaTed-Images (DIRTI) database
Food-11 "Food/Non-food Image Classification and Food Categorization using Pre-Trained GoogLeNet Model
UNICT-FD889 European Conference on Computer Vision ECCV 2014: Computer Vision - ECCV 2014 Workshops pp 584-599 A Benchmark Dataset to Study the Representation of Food Images Giovanni Maria FarinellaEmail authorDario Allegra
UEC-256 UEC FOOD256, UEC FOOD100
Picture recognition of food by macaques (Macaca silenus). Judge PG, Kurdziel LB, Wright RM, Bohrman JA. Anim Cogn. 2012 May;15(3):313-25. doi: 10.1007/s10071-011-0455-9. Epub 2011 Sep 15. PMID: 21918870
Dyscontrol evoked by erotic and food images in women with bulimia nervosa. Rodríguez S, Mata JL, Lameiras M, Fernández MC, Vila J. Eur Eat Disord Rev. 2007 May;15(3):231-9. PMID: 17676693 Similar articles
Evaluation of a digital food photography atlas used as portion size measurement aid in dietary surveys in Greece. Naska A, Valanou E, Peppa E, Katsoulis M, Barbouni A, Trichopoulou A. Public Health Nutr. 2016 Sep;19(13):2369-76. doi: 10.1017/S1368980016000227. Epub 2016 Feb 26. PMID: 26917048
© 2017 T.A. Houpt. Last updated 2018-01-07.