Abstract: Humans exploit prior knowledge to describe images, and are able to adapt their explanation to specific contextual information given, even to the extent of inventing plausible explanations when contextual information and images do not match. In this work, we propose the novel task of captioning Wikipedia images by integrating contextual knowledge. Specifically, we produce models that jointly reason over Wikipedia articles, Wikimedia images and their associated descriptions to produce contextualized captions. The same Wikimedia image can be used to illustrate different articles, and the produced caption needs to be adapted to the specific context allowing us to explore the limits of the model to adjust captions to different contextual information. Dealing with out-of-dictionary words and Named Entities is a challenging task in this domain. To address this, we propose a pre-training objective, Masked Named Entity Modeling (MNEM), and show that this pretext task results to significantly improved models. Furthermore, we verify that a model pre-trained in Wikipedia generalizes well to News Captioning datasets. We further define two different test splits according to the difficulty of the captioning task. We offer insights on the role and the importance of each modality and highlight the limitations of our model.

Abstract: Given a set of graphs, the median graph has been theoretically presented as a useful concept to infer a representative of the set. However, the computation of the median graph is a highly complex task and its practical application has been very limited up to now. In this work we present two major contributions. On one side, and from a theoretical point of view, we show new theoretical properties of the median graph. On the other side, using these new properties, we present a new approximate algorithm based on the genetic search, that improves the computation of the median graph. Finally, we perform a set of experiments on real data, where none of the existing algorithms for the median graph computation could be applied up to now due to their computational complexity. With these results, we show how the concept of the median graph can be used in real applications and leaves the box of the only-theoretical concepts, demonstrating, from a practical point of view, that can be a useful tool to represent a set of graphs.

Abstract: This paper presents a syntactic approach based on Adjacency Grammars (AG) for sketch diagram modeling and understanding. Diagrams are a combination of graphical symbols arranged according to a set of spatial rules defined by a visual language. AG describe visual shapes by productions defined in terms of terminal and non-terminal symbols (graphical primitives and subshapes), and a set functions describing the spatial arrangements between symbols. Our approach to sketch diagram understanding provides three main contributions. First, since AG are linear grammars, there is a need to define shapes and relations inherently bidimensional using a sequential formalism. Second, our parsing approach uses an indexing structure based on a spatial tessellation. This serves to reduce the search space when finding candidates to produce a valid reduction. This allows order-free parsing of 2D visual sentences while keeping combinatorial explosion in check. Third, working with sketches requires a distortion model to cope with the natural variations of hand drawn strokes. To this end we extended the basic grammar with a distortion measure modeled on the allowable variation on spatial constraints associated with grammar productions. Finally, the paper reports on an experimental framework an interactive system for sketch analysis. User tests performed on two real scenarios show that our approach is usable in interactive settings.