A Multimedia Database for Automatic Meal Assessment Systems Dario Allegra 1 , Marios Anthimopoulos 2,3 , Joachim Dehais 2 , Ya Lu 2 , Filippo Stanco 1 , Giovanni Maria Farinella 1 , and Stavroula Mougiakakou 2,4(&) 1 Department of Mathematics and Computer Science, University of Catania, Catania, Italy {allegra,fstanco,gfarinella}@dmi.unict.it 2 ARTORG Center for Biomedical Engineering Research, University of Bern, Bern, Switzerland {marios.anthimopoulos,joachim.dehais,ya.lu, stavroula.mougiakakou}@artorg.unibe.ch 3 Department of Emergency Medicine, Bern University Hospital, Bern, Switzerland 4 Department of Endocrinology, Diabetes and Clinical Nutrition, Bern University Hospital, Bern, Switzerland Abstract. A healthy diet is crucial for maintaining overall health and for controlling food-related chronic diseases, like diabetes and obesity. Proper diet management however, relies on the rather challenging task of food intake assessment and monitoring. To facilitate this procedure, several systems have been recently proposed for automatic meal assessment on mobile devices using computer vision methods. The development and validation of these systems requires large amounts of data and although some public datasets already exist, they don’t cover the entire spectrum of inputs and/or uses. In this paper, we introduce a database, which contains RGB images of meals together with the corresponding depth maps, 3D models, segmentation and recognition maps, weights and volumes. We also present a number of experiments on the new database to provide baselines performances in the context of food segmentation, depth and volume estimation. 1 Introduction Automatic diet assessment refers to the use of information technology for the ad-hoc translation of food intake into nutrient information in an accurate and intuitive way. Over the last years there have been a number of systems that use visual meal infor- mation to output nutrient content, mainly calories and carbohydrates [1–5], with only few of them being validated by end-users [6, 7]. Typically, once the visual information is available, a number of computer vision steps is executed: food detection, segmen- tation, recognition, and volume estimation. By knowing the food type and its volume and by using food composition databases the contained nutrients are estimated. Key element in the development and technical validation of the computer vision steps is the © Springer International Publishing AG 2017 S. Battiato et al. (Eds.): ICIAP 2017 International Workshops, LNCS 10590, pp. 471–478, 2017. https://doi.org/10.1007/978-3-319-70742-6_46