Geolocalized Modeling for Dish Recognition
Abstract of Geolocalized Modeling for Dish Recognition
Geolocalized Modeling for Dish Recognition
Geolocalized Modeling for Dish Recognition,Food-related photos have become increasingly popular , due to social networks, food recommendations, and dietary assessment systems. Reliable annotation is essential in those systems, but unconstrained automatic food recognition is still not accurate enough. Most works focus on exploiting only the visual content while ignoring the context. To address this limitation, in this paper we explore leveraging geolocation and external information about restaurants to simplify the classification problem.
We collected a restaurant-oriented food dataset with food images, dish tags, and restaurant-level information, such as the menu and geolocation. Experiments on this dataset show that exploiting geolocation improves around 30% the recognition performanceer.
Conclusion
Unrestricted dish recognition is a very challenging problem, and addressing the problem only considering visual information is very hard, even for humans. In general, but particularly in this domain, contextual information can significantly improve performance over visual-only approaches. By exploiting geolocation and user-contributed information about restaurants, we can effectively simplify the problem from thousands to tens of candidate classes. In contrast to most approaches using geolocation for image recognition, we do not use retrieval techniques but explicit discriminative classification.
