Infographics like graphic designs, data visualizations and graphs are pieced together by experts to contain visual and textual elements that come together to form a coherent story or a message. As a first step, we propose a cross-modal summarization task. Firstly, us- ing the extracted text we predict categories and tags representative of the main topics presented in the infographic. Secondly, we introduce the task of visual hashtag discovery - extracting visual elements from within an in- fographic that are most diagnostic of its topic.
A crucial goal of a graphic design or data visualization is to effectively communicate a message to a human observer. How effectively a message is communicated can be measured by how a design guides an observer’s atten- tion (eye movements, click patterns, and related behaviors), and how well the message is retrieved from memory at a later time point. In this work, we make automated predictions of importance for graphic designs and data visualizations.
The goal of this project was to do real time anomaly detection in ATM vestibules using intelligent CCTV cameras. The cameras were fitted with an ARM CPU with a custom version of caffe optimized for speed. Using a fully convolutional autoencoder, real time anomaly detection was performed on the embedded system.